off-topic-tech

[soft bloom]

hehe
in case 1 saying something is secret kinda defeats the purpose, doesn't it?
thankfully I don't have secret info

I will go watch that vid with young jesus

[pure karma]

all jokes aside tho i do think its on the verge of exploding because it started throwing WHEA errors left and right

[wooden vessel]

My son’s friend is a TikTok victim. But I don’t understand why the knives are out for it. I did hear the in app browser is a rat or has been in the past

[night girder]

The Passion?

https://tenor.com/view/disinformation-misinformation-blasphemy-heresy-fake-news-gif-24734459

[dire igloo]

If true, it's much much more helpful to say "I'd love to share the data but sadly that falls under an NDA" than to go around "oh yeah, evidence is easy to find. I won't tell you tho since I want to make you do the research for me"

[pure karma]

and in true magicz fashion my solution so far has been to just keep it runnign 24/7 and ignore it

[night girder]

Haha, sorry 🤣 That gif is from the passion but is so in the context of this channel.

[soft bloom]

no the one Fire suggested

[edgy hazel]

BED! NOW!

[night girder]

Sorry, I didn't read every detail of the conversation. Just read the screenshot I posted earlier. And somethign about yellowstone errupting.

And tapped out.

[dire igloo]

This one
https://youtu.be/Pc2psN0PFTk

[night girder]

Haha, young jesus 🤣

[pure karma]

best i have in the recently broken things list is this

caused from this insane turbo

[tribal kraken]

rookie numbers 😄

[twin dew]

So you broke yet another computer already?

[pure karma]

"damaged"

its still working so idc

[pure karma]

yea but thats a 10700F vs a 13900K

10700F is a 65W chip 13900K is like 280W

[tribal kraken]

253W is the stock Intel limit but icant keep 5,5Ghz all core

[pure karma]

im just surprised i managed to get a locked cpu to draw almost 3X its TDP for even a few seconds

i know this is probably going to start a debate but whats kind of the threshold in run hours where you cant really complain about problems anymore

i personally think its the big 10K

after 10000 run hours you cant complain anymore

you got your money's worth

[tribal kraken]

I have no really need to push it but 5,6GHz definitely possible. Thats the point where the voltage requirements sky rocket. But the Icemancooler block works really well. I've now ran prime95 on it for almost 3 hours, wanted to see where fan speed and water temps stabilize. It is now 31°C on the "hot" side water coming out of the cpu block, and 30°C "cool" side after radioators. Water pump fixed on 60% PWM about 3000rpm.

[tribal kraken]

But thats only 1,15 years of 24/7 use

[pure karma]

well thats difrent cause under server use it wont be cooling down and heating up nearly as often

realsitically a regular gaming cpu wont go for more than like 4-8 hours a day on average

that comes out to a good 4-5 years

[wanton orchid]

their organization loves them to publish *
they don't like having to publish in a hurry
they simply are asked to by higher ups

"oh no, oh fuck"

[tribal kraken]

Looks like my oldest Samsung 960 EVO 500GB NVME has 30k power on hours, 72TB written and 706 startups. And spinning 4TB WDRED discs in Raid1 about 40k hours, with 176 startups.

[wanton orchid]

damn 176 startups is not a lot that's nice

I've seen hdds with 600+ startups

I think it's also why I agree with the hate toward hdds in laptops

hdds are freaking good in stable enclosures in chassis/frames
not in a thin moving plastic thingy with small battery life

[pure karma]

i have 10K with 2200 startups on my old HDD

[tough owl]

how is this hdd still running with the magicz curse

[soft bloom]

Wait... I did watch it. hehe

[pure karma]

to be honest i often ask msyelf the same thing for my 2016 setup

[dire igloo]

Do you see how and why it applies to what you said in here?

[pure karma]

the only thing to have broken under my abuse on that setup is the sound bar on the monitor

[sharp oasis]

Oh god, oh no, oh fuck.

I mean

At least y'all haven't seen my 560w 6900XT

I'm using -30 ambient air to cool an air-cooled GPU

At like 1.8x it's intended TDP rn

So far fastest I have been able to score up for frequency is 2.89ghz

I do not daily this OC

[soft bloom]

Did i mention geology hub yet?

https://youtu.be/bEZyTIi7B98

[sharp oasis]

I daily 430w 2.77ghz

[soft bloom]

Oh wow, link stays

[sharp oasis]

Links staying here is crazy

[soft bloom]

Btw, he said people should evacuate.
But he definitely didn't say anything about Africa splitting in coming years afaik

[tough owl]

wccftech.com/chinese-officials-are-considering-selling-tiktoks-u-s-operations-to-elon-musk-amid-pressure-of-a-ban-or-forced-divestiture/ this has to be fake

[night girder]

Kade!

😡

[tough owl]

what did I do

[night girder]

👆

[sharp oasis]

Lmao

[night girder]

https://tenor.com/view/cat-cats-pet-cat-cat-pet-cute-cat-gif-24810247

[soft bloom]

He did say a couple of times in the past months about changes in magma composition.
And what happens in Indonesia with recent hot springs might be of the same nature though he explain why the cause might be different

[tough owl]

i dont read the history 😭

[night girder]

Clearly.

[tough owl]

i don't have time for that

[soft bloom]

Also, to Europeans: it's good to know your volcanoes.
GH has some vids about Slovakian ones

[sharp oasis]

I know I can never justify buying one. But it doesn't not want me to buy one. 5090 looks very nice lol

I honestly probably could not afford one

[soft bloom]

And to Americans:
https://youtu.be/OkqRVQAktNc?si=zfWzgFK3d_uVQRl_
Also good to know
Afair there's been some activity around New Madrid this century, but could be of human source
At least you should be aware of history

[sharp oasis]

I am happy with my 6900XT

It does like 90-110fps at 1440p ultra

Native.

[night girder]

It does 90-110fps at 1440p doesn't mean a lot without saying what game and graphic settings 🤣

But I get your sentiment. You are happy with 6900XT 😄

[sharp oasis]

Oh yeah. Uhh

[night girder]

The RX 6950 XT is holding up fine here too.

I do use upscale for certain games. 1080p. Ultra most of the time.

[sharp oasis]

Only game I do far require framegen right now on is stalker2 and that's native aa

[night girder]

Ah, stalker can probably use a little bit of optimization love. Not sure, haven't bought it yet.

I made the wrong descision and I bought Elden Ring. And played for a few hours and I don't like it lol.

[sharp oasis]

I do know warthunder SSAA still brings me from like 180fps in menu to 80fps in menu. But I'm basically rendering 4x 1440p

[night girder]

Squad is a shit optimized game for example, it really tanks my performance.

[sharp oasis]

Most games I have I can run native or a bit of aa

With maxed graphics

I like actually being at native now

And not struggling to maintain 60fps

Pov my 6700XT trying to maintain 60fps in helldivers 2 wile rendering 1440p native

[tough owl]

I want to email my schools IT staff to enable WSL on school computers in specific labs. what would be the best phrasing to go about asking for it

[pure karma]

Skip that step and do it yourself

[jagged snow]

"Enable WSL2 in order to allow students to gain experience in Linux operating systems without significant changes to existing infrastructure"?

[tough owl]

That’s the issue

I just talked in person with one of the guys and he said it’s hard 💀

[pure karma]

what is WSL anyway

im curious because its not a problem i ran into

[night girder]

windows subsystem for linux

[willow pike]

full linux kernel that runs in windows like a symbiote

lets you run linux software

[glossy glacier]

A mix between container and VM with some Integration into the Windows host

[willow pike]

legitimately good stuff

[glossy glacier]

Very neat yes

sadly they killed WSA

[jagged snow]

Honestly it's because of the way they tried to implement it

Good idea, bad implementation

[jagged snow]

Yeah, well
I had the same experience trying to get them to enable email forwarding

[dire igloo]

Eifel?

[glossy glacier]

I think they made great strides in the last updates.
But it was rigged from the start when they decided to use the amazon store..

[twin dew]

Well, Google Play Store wasn't an option?

[glossy glacier]

Nope

Dont know why but if you werent a bit tech savy you were stuck with the Amazon app store

[twin dew]

I mean Google Play Store was never an option that MS would have been allowed to include.
So Amazon was probably best they could get.

[glossy glacier]

It was possible to sideload with ADB, and later with the direct file integration.

[twin dew]

But I would expect the base on which that WSA was built was even worse than WSL1 ones.

Might come back at some point on WSL2 tech base.

[tough owl]

I’ll probably basically research the stuff myself and then email them with all the information

[soft bloom]

huh?

[twin dew]

The bad part is that actual distro files in WSL2 go under each users Windows profile files.

[tough owl]

Which is good

[twin dew]

WSL2 itself is easy to enable.
But if each user is to have separate distro install, it will take fuckton of space and fuck up remote profiles.

[tough owl]

As long as only users in specific classes with specific need can install it and not everyone on said machine we should be fine

Or if possible to move stuff to the non C drive

[dire igloo]

https://en.wikipedia.org/wiki/Volcanic_Eifel

[soft bloom]

I am really bad at remembering names

But I don't think this was in the vid simply from how it looks like

so, more info

[visual tree]

What's up with the package going from one amazon warehouse to another lol

[glossy glacier]

Logistics™️

[tough owl]

https://youtu.be/IKbFBgNuEOU?si=LNi3ZE3euw3mUrXj

Another one

[night girder]

Is there a reason he doesn't join effort with legal eagle? (have not watched video yet)

Because he filed a class-action lawsuit.

[tough owl]

Separate issues

Though they might be merged by whoever manages class action stuff

[glossy glacier]

thats mentioned not even 45seconds in xD

[glossy glacier]

i saw, still amuseing

[languid gulch]

would SF do better with a 9800X3D or a 9950X?

[gilded helm]

x3D

[languid gulch]

the cache outruns the cores?

[gilded helm]

That's my recollection, I don't have personal experience.

Sadly Satis still can't seem to crack hardware reviews because of an overly scientific approach by most reviewers.

Which has recently burned them with the B580.

[night girder]

I wonder if this still holds any value.

The reddit post is 3 years old: www.reddit.com/r/SatisfactoryGame/comments/ujef33/satisfactory_performance_on_5800x3d

and sadly, I can't find any benchmarks for satisfactory 😦

[languid gulch]

hoo boy i can smell the bullshit

[gilded helm]

And for Satis we're going to care about 1% and .1% lows. Combat can get nasty with hangs and missed inputs due to that whole factory thingy running in the background.

Timberborn is still king among games that get to multisecond input delays for me.

[languid gulch]

i just love the confidence in that SF claim i found 🤣

[night girder]

what website is that from anyway?

[jagged snow]

Satisfactory is better about scaling than it was in early access but especially now that the x3d clocks are the same there are diminishing returns past 16 threads

So yes, the x3d will be faster

[soft bloom]

https://youtu.be/dp7qNWhPNXk?si=lwAmsGX84NxopZaz
I am yet to understand why people in comments call it crack...
But hearing that Indonesia has music tradition based on offseting two pitches and playing them together... woah

[glossy glacier]

why? too low? too high?

[soft bloom]

Also how often do you almost get answer for something from an unepected area of discussion?
well, oscilations is probably one of the most intriguing topics in science

[languid gulch]

i suspect that a 9800X3D and a 4090 will do better than 80fps on low

[glossy glacier]

can confirm a 9800X3D will get about 70ish frames in a medium factory
3440x1440, ultra settings, 3080

and GPU was only at ~90% usage so probably already in CPU limit

[languid gulch]

interesting. feels way faster than that, even on lower hardware

[night girder]

Again, depending on website. Maybe it isn't updated for 1.0?

[glossy glacier]

was a quick & dirty test, so maybe 10-20 more fps when only running satis

[pure karma]

I saw a 8x performance increase in 1.0 so yea that does seem low

[soft bloom]

Why would I get spikes of internet usage (sort of like on/off) and spikes of disk usage (quickly reaching peak and slow decline, that stop near zero and jumps to local max again) reported by steam, while windows reports just steady load on Disk?

Kinda frustrating to watch. Almost like dial-up internet with estimated time lying and jumping

Hmm, giving disk some time to rest seems to have helped? At least internet is on more often.

[tribal kraken]

9800X3D and 4090 for me runs about 120-130fps min at my nuclear plan, on a dedicated server

4k ultra

[tribal kraken]

Currently ingame heres some data aquasuite.aquacomputer.de/c9be40e6-e055-4649-b9ff-5d8b212d9e59

[tribal kraken]

Have to go sleep so the live datapage will be from idle PC but this is how snapshot during satisfactory looks like

[jagged snow]

That sounds... off
I'm on a 5600x and I can get much higher frames than that on smaller saves

Don't drop to 60fps until I'm on 500 hour saves usually

[glossy glacier]

With or without global illumination?

[jagged snow]

Without of course

We were talking about low settings

[glossy glacier]

I had everything on ultra

And GI on

[gilded helm]

I'm really curious about the future of Satis now that it has possibly made it onto Nvidia's pet project list to some extent.

Probably a swap of development support in exchange for appearing in Nvidia slides.

[languid gulch]

10 year old game sucking down all my vram 🤣

[limber flicker]

@twin dew just thought I’d let you know since you were helping me try fix it my pc is dead the motherboard died I took it to a shop and they found out the issue

[languid gulch]

yeesh. what did the autopsy reveal?

[limber flicker]

No clue yet all they have revealed is it’s dead

[tough owl]

That’s a pfp of nightmares

[languid gulch]

it's perfect. completely unforgettable

[twin dew]

You would need to check stat unit.
My 7800X3D can maintain 100 fps in 30MB save from CPU side.

[glossy glacier]

Stat unit?

[twin dew]

ingame console command to get diagnostic numbers.
stat fps and stat unit

Will show separate frame times for game logic, and various rendering related parts etc.

Game is game logic on CPU.
Draw is render feeding on CPU.
RHIT is glue logic between game and GPU driver on CPU.
GPU Time is GPU actually rendering the frame.

Frame is one frame offset total time from the previous steps, should be close to highest of Game of GPU Time (screenshotting broke it).
And then the actual display stats, as FPS and frame time, again one frame offset from Frame.

[glossy glacier]

Will give it a try once I'm home again

[twin dew]

If hardware accelerated gpu scheduling is on, then Draw will be about same as GPU Time.
And not tied to purely CPU anymore.

Same is probably the case if Reflex or like is in play, delaying finishing the frame setup to get lower input latency.

[tribal kraken]

I tried quicly to look for commands to increase the stat display font size, its barely readable at 3840×2160

But did not find any

[twin dew]

AFAIK there are none.
And that hasn't probably been touched since UE4 was launched.

[tribal kraken]

For Satisfactory I think my build gun aim is as efficient if it were 60fps or 120fps 😄

[twin dew]

CSS could change it it seems, DEBUG FONT SIZE

But the full documentation for those stat commands, but most are disabled in release builds of Satisfactory:
https://dev.epicgames.com/documentation/en-us/unreal-engine/stat-commands-in-unreal-engine?application_version=5.3

[languid gulch]

so something may have happened, gotta wait a few days to confirm, but if it's true, i'm getting something fun

being intentionally cryptic because nothing's set yet 🤣

[soft bloom]

https://www.youtube.com/watch?v=b6utGZQ9Sks
yet another (old) Tom Scott video
but it got me wondering
what if... it can be used somewhere else?
that ferry is basically just a giant fin with 2 fixed axis and 1 free (also manual change of direction)
wind turbines use similar idea
what is analogy of this in electronics?
I mean, we have potential change with altitude...

[soft bloom]

Is Jonathan Blow crazy or wise? how much and about what?

[twin dew]

You forgot the chain from that, doesn't work without the chain/cable.
Just redirecting waterflow energy to push the ferry along the chain.

[soft bloom]

yes, that's why I said that 2 axis are fixed (water level, chain's position along river) and 1 is free (chain crossing river)
just like wind turbines don't move with wind but a fixed on ground

[twin dew]

But one use batteries are the "equivalent", using chemicals to generate electricity.

[mental oriole]

What context?

[twin dew]

And neither of those first two matters?

It is just about flow speed, and angle of attack.

[soft bloom]

the second one matters or the ferry would go with river

[twin dew]

With chain preventing the ferry from going down river while using that angle of attack to move just sideways.

[twin dew]

Point was that water level, or chains exact position, don't matter.

Just flowspeed and that there IS a chain.

[soft bloom]

well, recently I discovered he talks a lot about programming in general (also he develops JAI...), so asking more about that

[twin dew]

And Wind turbines are again just fixed electrical engine with adjustable blades, converting air movement into rotation, and then that into electricity.

[soft bloom]

well, yes.
though if the goal was to optimize the speed of ferry then it would matter

[twin dew]

Again, just about that flow speed of the water.

[soft bloom]

I was pointing at idea that there's a force that moves matter that can be interacted with (fins) and we set something what doesn't move with it, and exploit the opposing reaction force to get the work done. that's the analogy between them

[mental oriole]

I have no clue on the goals of JAI, except it's main use will be game dev (if I remember correctly). And since the language is still in development I have no opinion yet.

[soft bloom]

...and for wind turbine, it's height in the same location also will reach for different flow speed

[twin dew]

Which is just about boundary layers of ground.

[soft bloom]

well, not JAI, but he's takes on the programming

[twin dew]

You are mixing causes and effects.

[mental oriole]

I don't follow him so.

[soft bloom]

ok, don't pay attention to flow speed as function of height or position along the river because it's really not the thing I was talking baout, just something extra

I lost the message about chem battery...

which is kinda like having small bottle of water that you spill on a tiny modeled river path and it does the work

(let's forget for a second that gravity does the job)

if there was (and there is) a source of flow of electrons, how to make "fin" that can exploit it to do some work?

[twin dew]

What do you think all electronics, and electricity using devices do?

[soft bloom]

yes, but, that's a different question.
they have source of such "flow", but what creates that flow? our devices, that mostly just use heat energy

[twin dew]

No, that heat from electronics is byproduct.

From the actual intended work done from the flow of electrons.

[soft bloom]

that's a whole different thing you mention

there's source of heat -> boiling water -> steam turbine -> flow of electrons -> wires -> device that exploits 'flow' of electrons to do some work
                                            solar -------/
                                     wind tubrine ------/

[twin dew]

You are now mixing HOW the electricity is generated at large scale, and how electricity is used.

[soft bloom]

ye because you asked

[twin dew]

Where?
You asked weird questions about how to make fin to take power from electron flow, which is what all electricity use already does in various ways, either directly or via that flow generating magnetic fields that are then used.

And then there are lot of ways to turn either physical work (via magnetism) into electricity and that flow of electrons, or chemical energy directly into flow of electrons.

[twin dew]

Basically to me this whole line seems just that you have no idea what electricity is.

[twin dew]

For example here, both with that ferry and with wind turbines, only flow speed matters when you look at the physical interactions.

Then there are things that affect that actual flow speed of water or air, but those aren't directly related to the effect, and are only of interest to someone building actual implementation.
Like that boundary layer between ground and water/air causing slower flow nearer to the ground.
Where with wind turbines you need to balance the material costs etc. with effects of that boundary layer to determine how high turbine tower you want to build.
And the effects of surrounding terrain to that to select placement.
But those are implementation details.

With rivers, the total flow mass over time, and the river channel cross-section determine the overall flowspeed, with that boundary layer etc. then determining how it is spread over that cross-section.
Which parts are slower, which ones are faster, etc.

And in turn that total flowmass then depends on the amount of water collected from rain upstream, and height differences etc.
And if there are lakes upstream evening things out and acting as reservoirs and so on.
But again, goes into implementation details if someone wants to build something using that waterflow in certain area, to select the best spot or size the water wheel for wanted output power or like.

[twin dew]

Well, not flowspeed directly, but mass flow, but mostly that mass is constant enough when talking about these specific cases, but not overall.

[languid gulch]

dumb question time: for those O11-style cases with 9 fans (intake on bottom, exhaust on top with an AIO, intake on side), would it be better to have higher airflow fans on the AIO or somewhere else

[twin dew]

You want the input massflow to be little higher than output massflow, to get tiny bit of positive pressure inside the case, and as much of that exiting via that radiator, and only little from the other holes in case.
To maximize mass flow that actually cools the water.
And still getting all the input airflow through any dust filters.

[languid gulch]

so maybe have the bottom &/or the side fans be higher airflow than the AIO's exhaust fans?

[twin dew]

Aren't you having many more input fans than output?
If you are filling all the places.

And that radiator will be causing backpressure, dropping the flow that those radiator fans would otherwise generate.

[languid gulch]

yes, 6 intake/3 exhaust, but also reusing 3 higher airflow fans since they're decent

so trying to figure out best config

the 3 high airflow ones are double the pressure & volume of the others

cuz noctua 🤣

or should i have the AIO have the high airflow, but as intake at the top & exhaust out the side & bottom

[twin dew]

Pressure rating is at 0 airflow.
And airflow rating is at 0 pressure loss causing obstructions before or after the fan.

The actual operating point is always between them, and the line isn't straight.

[languid gulch]

good point

so i should look up the airflow thru the AIO

[twin dew]

Noctua article on that: https://noctua.at/en/nf-a12x25-performance-comparison-to-nf-f12-and-nf-s12a

Comparing two previous Noctua fans to then new NF-A12x25 which loses to those two on pure specs, but is better in actual operating range for all real cases.

[languid gulch]

yea, it's 3 of the industrial PPC 120mm 2000rpm fans i'm reusing

[twin dew]

But you probably want those Noctuas on the rad.

And then tune with fan speeds.

[languid gulch]

will i need to worry about bearing wear, since those 3 are currently vertical mounted?

[twin dew]

That kind of orientation change shouldn't affect in most cases with fans.
And Noctua fans work in that pushing up orientation, which isn't true for lot of fans.

Lot of fan bearings don't handle that specific orientation well.

And radiator obstruction will just make it even worse.

[languid gulch]

with the AIO being at the top, would it be better to have the fans above the AIO and pushing air into the AIO instead of trying to suck thru it?

[twin dew]

You can have the fans below the AIO pushing through it?
But depends on exact fan if push or pull through radiator is better.

And that second one just goes into the fan bearing.
If it can take the combined gravity pull and fan thrust without issue (when airflow is from down to up) or not.

And if you want top to exchaust or pull fresh air in.

If it pulls in, then you need to see where to exchaust.

[languid gulch]

yea, trying to decide if it'd be best to have top or bottom as intake, rad definitely has to go on the top either way

lol is there a case fan airflow builder out there 🤣

[twin dew]

But in most cases, push fans are better to get air through a radiator than pull fans.

[twin dew]

Anyways, Blue Origin New Glenn now targeting 2025-01-16 06:00 UTC for next launch attempt.
And SpaceX Starship Flight 7 is targeting 2025-01-15 22:00 UTC for attempt, including another booster landing attempt and upper stage PEZ dispenser test attempt with Starlink v2 mockups.

Now with updated Spaceship version with front fins moved backwards to have the hinges in more protected location.

Sorry, leeward

backwards could also mean lower down, as in closer to engine section.

[languid gulch]

are they going for full orbit with starship this time?

[twin dew]

No.

[languid gulch]

still doing that "orbital energy" traj

[twin dew]

No use until they are both certain that engine restarts are certain and they have actual Starlinks in cargo.

And they can do those deorbit tests on suborbital trajectory and have no danger of ground hits, and they need to test that pez dispenser to see if it works properly this time to start launching actual Starlinks on later test flights.

They tried the payload door in one of the first full hops, but it failed badly.

[languid gulch]

can't wait to have a shot at seeing it reenter over me

[twin dew]

But this is first flight of Starship v2 (still with Booster v1), which is 1.8m longer for 25% more fuel and lot of other changes, like those repositioned front fins.

But Musk has said that this is planned to be last suborbital one.
And next one will try to return to Starbase for catch if this one goes well and that is allowed by regulators.

[tribal kraken]

New BIOS for X870e Crosshair. Wonder what the "select" games are

And chipset drivers too, not sure if want to test it

[twin dew]

SpaceX really doesn't want this to happen to Starship:
https://arstechnica.com/space/2025/01/a-glowing-ring-of-metal-fell-to-earth-and-no-one-has-any-idea-what-it-is/

[languid gulch]

oh definitely, they've already done that with plenty already

[twin dew]

Even if that specific one isn't actual space debris.

Somehow seems more like there was separate fall, and then that thing was found.

But it isn't the thing that fell.

Not melted, no heat, no hit marks of any kind on ground etc.

[languid gulch]

oh yea it looks like a giant worm gear to me

some factory nearby might've gone boom & nobody wants to admit it

[twin dew]

Or just crap left behind at some point.
Which was found because someone went looking for whatever caused the light show.
So something actually came down from space, and that triggered finding of something left behind before.

And the actual thing that came down is elsewhere, or burned up completely before hitting ground.

[twin dew]

I'm waiting for that to exit beta.
Went from test (linked on specific forum thread) to beta (publicly on support site) very fast.

[languid gulch]

having players be beta testers for games is ok. having customers be beta testers for hardware is asking for a lawsuit

[twin dew]

To get test versions, you need to look in obscure place on Asus ROG forums in one specific thread.
To get beta versions, you need to intentionally download those versions clearly marked as Beta from support site.

And both are only released for MBs that have full flashback functionality.

TUF and Prime models don't get them.

Well, depending on exact MB more.

[twin dew]

So complete opt-in, and only for those who can downgrade with "broken" MB.

[languid gulch]

oh i mean "beta test" as in when they clearly publish a bios revision before it's been properly tested

it's fine to break a game, it's not fine to break a board

[twin dew]

Has to have gone through internal testing, and then was in that public test mode for few days.
Usually that public test-test is week or so.

And is now available with clear "Beta Version" red label.

Because it contains important security fixes to firmware.
Most of the firmware versions never get put out with that Beta label on Asus support site.

Go straight from that more hidden Test, to full release.
But often go through couple of versions of those Test releases.
And even more before getting to public hidden Test release.

AM5 AGESA/UEFI/BIOS List | SMU AM5/AM4 Table. Asus AM5 MBs:
https://docs.google.com/spreadsheets/d/12zg6yT_H7H-W1voyw1ZoIrj0GSE7WI4Ug-uLlv-Asa8/edit?gid=937453961#gid=937453961

When "Test" version gets redeployed with same version number as Beta or Release, or once Beta loses that Beta status, it just gets replaced in that listing, not shown multiple times.

So second or third revision of Asus BIOS with AGESA 1.2.0.2b got Test release, and only third version of Test got full release.
It never had public Beta.

[winged valley]

Speaking of bios versions, is it completely unheard of for the nomenclature of bios version numbers for a particular motherboard to change midway through its lifecycle?

[night girder]

Meta is going to fire 3600 employees en rehire other people. Based on performance. Memo written by Mark Suckerberg

[twin dew]

Unfortunately no.

[winged valley]

Oh

[night girder]

www.bloomberg.com/news/articles/2025-01-14/meta-is-planning-to-cut-5-of-lowest-performers-memo-shows

[winged valley]

Maybe what I found wasn't special then

[twin dew]

Better now.

[winged valley]

A while back I was helping a friend make his pc faster (operate as expected) and found his bios didn't have an option for PBO. When I checked the bios version, it was something like 00.0x or whatever, and after updating it got changed to something normal like 55.2.A and PBO appeared

I thought we had some early development prototype bios out of the box

[twin dew]

MSI?

[night girder]

Meta is also blocking pixelfed. Wtf is meta doing.

[winged valley]

For the life of me I don't remember, mightve been gigabyte

[twin dew]

MSI is all over the place with the in-BIOS shown versions IIRC.
Even when the actual version strings are somewhat stable now.
But still all over the place between related MBs.

Gigabyte fucks up in using same name with just "ver x" added for different MBs that use different BIOS versions, and the method they use to differentiate those BIOS versions.

But Gigabyte was also IIRC pretty bad before.

[winged valley]

I'm stupid. Not PBO, SMT

SMT not being available on a B series board is definitely not AMD spec

[twin dew]

Fuck Gigabyte... B650M S2H has three revisions (1.0, 1.1, 1.2), each uses different BIOS/UEFI version.
But 1.0 and 1.2 use same versioning (but checksums are different), 1.1 uses different versioning...

As example I checked.

[winged valley]

Messing with uefi is so much fun (not) I sure love trying to make things work (not) even enabling ACS in my server's uefi isn't enough and I still had to enable it in my kernel which is very (not) fun

I love messing with uefi so much I'd rather pay a subscription to Vinafix to download dumps than manually comb through the SPI image trying to locate the bios password

[night girder]

since you al talking bios n stuff, might well check if I need to update it.

but system has been stable (ish) so maybe not fuck around with bios atm and just enjoy gaming 😄

[tribal kraken]

Asus placing the setting for AMD 1:1 UCLK-MCLK in Bios is really confusing at the start too

[twin dew]

That it is low down in the advanced RAM settings part with all the timings etc?

[tribal kraken]

yeah

And I think its in another "AMD oveclock" menu too, and not even sure if they both need to be changed

[twin dew]

AMD Overclocking, AMD CBS and AMD PBS are menus from AGESA itself.

AMD coded menus that MB maker just has to expose for access.

Theoretically you set things in one or the other and as long as the other one is on Auto, it will just work.
But on some MBs that is broken and you need to set in one of them, and the other does nothing, or need to be set in both.
With Asus AFAIK it works right on all MBs.

And MSI and Gigabyte are AFAIK bad with that.

[tribal kraken]

ok, and also if you put higher than 6000 (or 6200?) DRAM Frequency, it will automatically swap to 1:2 mode, even if you had changed it manually earlier

[twin dew]

Ok, that isn't what my MB did when I was testing that.
That is just what it should do on Auto, not with manual setting to either...

Probably idea having been to make that default automatic difference more explicit, but failing the case when it was already set now.

[twin dew]

Basic AMD "algo" is 1:1 up to 6000, and 1:2 from 6200 up.

[dire igloo]

With the important note that anything from 6200 to 7200 will be slower than 6000 if run at those ratios

[twin dew]

So tired...
But need to hold on for another hour before heading to bed.

[night girder]

ok

Just reading TikTok is going to block US.

So, no more TikTok for them.

Going from "maybe we want to sell to Elon Musk" (yesterday) to "we probably just going to block it for US" (news today) 🤣

Ah bytedance says the news about selling to Elon Musk is fiction.

[honest karma]

I just checked what I would need to pay for a 10 GbE NAS and all the other stuff I need for it. 🫣
Quality NAS storage is expensive.😱

[tribal kraken]

I just upgraded my work room switch to Ubiquiti flex mini 2,5G as motherboards usually have that intgrated, some even 5G. But not really 10G

[jagged snow]

Even 10gig lan is expensive lol

[twin dew]

looks around room with two computers with 10G LAN and two switches with it...

[honest karma]

you just need a network card for it

[twin dew]

And those start from 100e or so.

[honest karma]

yes 150€-300€

[glossy glacier]

Buy used

[twin dew]

And if you aren't just doing one-to-one, then you need 10G switches too, etc.
Gets expensive fast.

[honest karma]

yeah between 1300-1900€ overall according to chatgpt

[glossy glacier]

Prebuilt or DIY?

[twin dew]

Which cannot count.
So that is useless.

[glossy glacier]

Kinda curious what it recommended though

[twin dew]

The good stuff isn't really on used market yet.
Only the first and second gen stuff that takes lot of power and cooling, and doesn't support 2.5G and 5G speeds.

[honest karma]

Thats what Chatgpt says:
Synology DS923+ NAS: 650-700 €
2x Samsung 970 EVO Plus 1 TB (NVMe SSD): 300-500 €
Intel X550-T2 10 GbE Networkcard: 150-200 €
Synology 10GbE Networkcard (E10G18-T1) for NAS: 150-180 €
Netgear XS505M 10 GbE Switch (optional): 200-250 €
DAC-cable (10 GbE): 40-80 €
1 GbE lan cable: 10-20 €

[twin dew]

Those prices have been pulled out of ass.

[honest karma]

Good, i havent checked everything but when i did like with the network cards it suddenly got even more expensive..

[twin dew]

And that switch isn't optional with that config, as that Intel card is RJ45, and that Synology card is SFP+.
Direct Attach Cables are SFP+ to SFP+.

[honest karma]

so whats your guess what i would cost me?

[dire igloo]

ChatGPT's biggest influence for tech is popularity, not validity or performance.

The more people say that a product is good, the stronger ChatGPT's bias towards that product.
Doesn't matter if those people are knowledgeable or not

[twin dew]

And that specific Synology cannot even take that specific 10G card.

[honest karma]

so it doesnt use benchmarks for cpu's as reference? lol

[dire igloo]

Avoid it for product research unless you're okay with getting the average reddit shill's opinion mixed with a bit of random guessing

[dire igloo]

How could it?
It's a language model, not an actually intelligent researcher

[twin dew]

LLMs are LANGUAGE models, they just know about certain words coming after other certain words with largest probability.

[honest karma]

meh i thought i would be the easy way

[dire igloo]

https://youtube.com/playlist?list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi

[glossy glacier]

Where are the HDDs? xD

[honest karma]

yeah it told me to buy SSD's

[dire igloo]

Watch these videos if you wanna learn more about how LLMs works.
They're a tool and just like every other tool in existence, you need to know what it can and can't do in order to use it properly

[twin dew]

And LLMs are just about being able to generate text that seems like human written in cursory glance.
With absolutely no guarantees about content.

[honest karma]

i will

[dire igloo]

Also, this one to learn more about the strengths and limitations
https://youtu.be/QrSCwxrLrRc

The 3blue1brown videos are more mathematical and theoretical, this is applied to reality

[willow pike]

it has no reference; it knows that if you say "how much is a Samsung SSD", something that looks right is Samsung 987, even though that's not a real SSD. it doesn't know what's real, it doesn't know anything, and it will make a price up. ask it repeatedly how much something costs and it will invent new prices every time

[dire igloo]

There is some reference: the training data.
Which is outdated and heavily skewed by popularity

[honest karma]

Yeah it has limitations but i thought i could do simple stuff like this

[willow pike]

it cannot

[dire igloo]

It's not simple stuff, lol

You're asking it to do something specific. Don't do that.
The less depth your question has, the better the answer will be

[willow pike]

the entire AI thing is a bubble waiting to pop, when more people realise it can't do what they say it can do

[dire igloo]

Of course the answer will have less value too, but that's another discussion

[honest karma]

I guess i will have to do my own research then, which could take hours x)

[willow pike]

ask chatGPT about anything you know a lot about, I mean really know, and watch as it makes stuff up and is so confident about its wrong answers

[dire igloo]

Ehhhh, I think it will pop only when ai does the opposite of what it's supposed to do - like when people got a company's chatbot to invent new warranty and return policies

[willow pike]

there's examples of stupid shit like that

[glossy glacier]

Get an old PC, throw in some drives and done

[honest karma]

well my girlfriend is working in a legal tech company and they are using AI to check how high their chances are of winning

based on that they take on work lol

[twin dew]

🤦

[willow pike]

oh god

https://tenor.com/view/thanos-memoji-gif-23490017

[twin dew]

So decision making based on LLM hallucinations?

[honest karma]

it has something to do with algorithm or something but they do have lawyers checking

[willow pike]

so fucked

[honest karma]

i dont know the whole principle

[tribal kraken]

That would be correct for the AI as HDD will have trouble keeping up with faster than 1GBit networks without perfomance oriented RAID

[dire igloo]

ChatGPT can still aid you with that. I've asked it for good resources in the past and was generally pointed to some great channels and websites.

You have to remember two things tho:

1. Anything ChatGPT says has to be treated as a random guess. If you want accuracy, you need to validate it. Don't trust its output to be true.
2. Be aware of the tool's limitations. ChatGPT is great at what it's supposed to do - but that usecase is much narrower than you think.

[soft bloom]

Well it's a curiosity on how that man visions language with meta programming.
Considering that he has developed Witness with completely optional puzzles... I wonder the language will handle optional stuff.

So far, I love that you can have non-default keyword args out of order. I mean, ordering args, having positional ones... All of that seems archaic to me.

[honest karma]

RAID was something like a mirror storage as a backup?

[dire igloo]

VERY important distinction here:
AI is NOT automation.
You can have a computer automated system for that very purpose that works completely without AI.
In fact, a lot of automation systems are now referred to as AI when in reality it's a completely hardcoded algorithm

[honest karma]

way narrower than i thought it seems

[tribal kraken]

Multiple types and levels, some mirror data for backup and other types for performance, but you can do both if buy lot of drives

[willow pike]

also using it in law is just... terrible. chatgpt, not AI in general, specifically chatgpt, has been caught hallucinating and making up legal precedent multiple times

[dire igloo]

And stupidly enough, a lot of really good automation systems added AI to keep up with the trend - and got worse in the process

Proper automation is almost always better than AI

[twin dew]

All LLMs hallucinate.
Natural consequence from just storing word associations.

[glossy glacier]

Autocomplete on steroids

[willow pike]

Schwartz, an attorney with Levidow, Levidow & Oberman in New York, used the OpenAI gen AI chatbot to find prior cases to support a case filed by Avianca employee Roberto Mata for injuries he sustained in 2019. But at least six of the cases submitted in the brief didn’t exist. In a document filed in May last year, Judge Castel noted the cases submitted by Schwartz included false names and docket numbers, along with bogus internal citations and quotes.

[honest karma]

wellp i dont know what they are using it exactly for only that they are searching for cases with the highest probability of winning and my girlfriend ist going to be a lawyer herself so it wont replace her

[willow pike]

yet

[dire igloo]

Even worse: iirc that dude was asked to provide information about those precedents and then had ChatGPT write the court documents for those cases.

And if it wasn't that dude, I know some other lawyer did that

[twin dew]

Which none of the current general ones can do, and the answer will be pure hallucination with random content.
And even if it is specifically tailored and trained one, it will still basically be just random coin toss.

[willow pike]

lmao

[honest karma]

the company has created it themselves so i dont know

[twin dew]

And most probably then has nothing to do with AI, and is just some kind normal algorithm-tool.

[dire igloo]

Also: it's not all about win probability, profit potential is a gigantic factor too.
A slam dunk case worth $1000 is far less valuable than a $5m coin toss

[soft bloom]

Considering recent video from WelchLabs, there's awya to extract and edit specific associations in LLMs "memory" (weights).
So it could be that we will see less hallucinations over time

[willow pike]

www.bbc.com/travel/article/20240222-air-canada-chatbot-misinformation-what-travellers-should-know

In 2022, Air Canada's chatbot promised a discount that wasn't available to passenger Jake Moffatt, who was assured that he could book a full-fare flight for his grandmother's funeral and then apply for a bereavement fare after the fact.

The British Columbia Civil Resolution Tribunal rejected that argument, ruling that Air Canada had to pay Moffatt $812.02 (£642.64) in damages and tribunal fees. "It should be obvious to Air Canada that it is responsible for all the information on its website," read tribunal member Christopher Rivers' written response. "It makes no difference whether the information comes from a static page or a chatbot."

[honest karma]

ai algorithm were her words, well she is literally someone who is clueless about computers so maybe i shouldnt take her words for granted but if that were the words of her company i dont know 😄

[dire igloo]

And THIS is where I see AI plummet into irrelevance.
It becoming an expensive legal liability

[willow pike]

lawyers should be scared shitless of AI from a liability perspective

[honest karma]

well they do actual lawyer work, i dont think it makes majors decisions for them

i mean they wouldnt need to be lawyers then

[dire igloo]

As elaborated earlier, a lot of standard automation is now referred to as AI despite there being no actual ML involved.

People have no idea what AI is or means and with the media craze around it, they'll call anything AI that automatically does stuff

[honest karma]

true, i myself am not that deep into the subject so what your saying is probably correct

[twin dew]

In multiple cycles, starting from basically static decision tree algorithms.

Many many years ago.

[dire igloo]

I really really wanna see those HR and marketing fucknuts get their asses handed to them by the legal department - because I KNOW that there is an IT department in the background that will throw a "told ya so" party

[dire igloo]

Yeah, uhhhh, that's probably not a good thing and part of why you fell for ChatGPT's nonsense.

Just because something sounds true or smart or thought out doesn't mean it is

fwiw, I could be as much of an uneducated pretender like ChatGPT

[honest karma]

at some point i would have noticed, since i wouldnt just buy blindly

[mental oriole]

Have to wait and see... I'm not a fan of meta programming for everything but it does have it's uses.

[honest karma]

i build my computer on my own and put every part together so im not clueless

[night girder]

Today, I saw a tool that let's a AI peerreview your PRs

Me: "isn't this just the same as a linter?"

[dire igloo]

These two aren't mutually exclusive, I've seen people make some horribly dumb part choices because they thought they knew when in reality they didn't

[night girder]

You take every tool/program/app in existence. Just slap AI into it. And the result will stay the same. Now it just has AI.

[glossy glacier]

And a NAS is mostly the same. Just trade looks and GPU for functionality and storage.

[night girder]

Photoshop + AI. Synology + AI. Discord has probably some AI into it.

[dire igloo]

It absolutely does. iirc you can use discord native AI utilities to create image descriptions

[night girder]

So everything has to have AI built into it.

[honest karma]

I think it was pretty good at that time:
AMD Ryzen 7 5800X 8-Core Processor (3.80 GHz)
Corsair Vengeance RGB PRO 32GB (4 x 16 GB) DDR4 3600MHz C18
Nvidia GeForce RTX 3070 Ti
Samsung SSD 980 PRO (
ROG STRIX X570-F GAMING

[twin dew]

Because the MBAs got caught in the hype, without understanding anything (again)

[glossy glacier]

Its the new crypto/blockchain

[twin dew]

Very weird RAM choice.

[soft bloom]

No. Purpose of review is everything what isn't linter, or test. Aka stuff you might argue about, hehe.
So it's liking hiring a jun who won't sleep and will tell you how great your code will be if you do extra stuff you didn't do because business/budget etc

[dire igloo]

4x16 = 32
quik mafs

[night girder]

You don't get it.

[twin dew]

And probably won't work stable at the specced speed anyways with that CPU.

[soft bloom]

Yeah. Companies invest. And later they will get to know what was worth the money...

[honest karma]

it does?

[twin dew]

Not if it is actually 4 sticks.

And not 2.

[glossy glacier]

My 5800X ran 4x8 @ 3200 fine

[twin dew]

If it is really 2 sets of 2, you probably didn't enable XMP.

[honest karma]

okay lets define unstable, what problems are you expecting

[twin dew]

That is much different from 3600MT/s.

[soft bloom]

Just 1 bit error

[glossy glacier]

Only 200Mhz

[twin dew]

And if XMP wasn't enabled, that set will be running at 2133MT/s knowing Corsair.

[dire igloo]

most definitely overpaid on RAM, SSD and MB.
GPU I'll leave aside cuz this sounds like a chip shortage config
why that CPU tho? what did you expect to get from 8 cores and why 5800x over 5700x?

and finally: what about case, cooling and PSU?

[honest karma]

it is overpaid, i agree, it was for the looks

[edgy hazel]

Can you watercool with molasses?

[dire igloo]

that's paying extra for aesthetics. overpaying is something else

[dire igloo]

105th anniversary

[night girder]

let me check

[twin dew]

And 3200MT/s is the official max for 2 sticks.
And 2933MT/s is the official max for 4x1R sticks like the talked config.

3600MT/s is the unofficial "this will work with all CPUs" when using 2 sticks.
But the actual max for 4 sticks is usually way below that, much closer to that official max.

[night girder]

It's not advisable to use molasses for water cooling; standard water or appropriate cooling fluids are much better suited for that purpose.

Using molasses could lead to clogging in the cooling system due to its thick consistency and the potential for it to leave residues. Additionally, molasses can ferment and spoil, which could create further issues in a cooling loop.

[dire igloo]

the only aesthetics portion in the things that I criticized is the RGB on the RAM - and I'm confident that there was equally good if not better quality RGB RAM for a similar price if not lower

[honest karma]

from corsair? if so i havent found it

[twin dew]

Corsair is the crappiest global RAM manufacturer there is

[tribal kraken]

Actually I've been tempted to try something like Redline Water Wetter, coolant booster for combustion engines as a computer coolant if it has any benefits but, expectations have not been high enough for the trouble

[dire igloo]

you've found the point of criticism: they're obviously not from Corsair.
brand doesn't matter. you want a good product? buy a good product. Don't buy whatever product because it has a certain logo on the box and some fancy marketing around it

[twin dew]

But even with DDR4, going to 4 sticks is bad idea if you can get the same amount of RAM in 2 sticks.
With DDR5 it is death sentence to performance.

[dire igloo]

Strix mainboard, Corsair RAM, Samsung SSD
this smells a lot like brand bias - which is a strong indicator of surface level knowledge

[dire igloo]

*except for quad channel platforms (not the case here, not the case basically ever unless for workstation platforms)

[honest karma]

wow, so your saying you cant have your favorite brands you were using for 30 years without problems?

thats my first AMD CPU btw

[twin dew]

Even if you wanted Asus MB, why Strix branded, which are extreme-OC boards with overinflated prices to match?

[night girder]

🤔

[honest karma]

i mean we can debate about efficiency and price then we probably wouldnt use any popular brand

[night girder]

two measurements as always in this chat, love it.

[soft bloom]

oh which ssd is good these days?

[honest karma]

because i always used asus like i always used corsair i wont stray from a brand when it works fine

[twin dew]

Too large, so they are able to use the crappiest binned RAM for every level.
And switch those RAM chips without changing the SKU.
And use the same crappiest possible SPD programming for all their RAM that has XMP profiles, so if you don't enable XMP, you get very crappy performance.

Most other makers have SKU changes and better matching SPDs.

[soft bloom]

10 years ago i thought Crucial is the king. but now they have 'budget' ones too

[dire igloo]

you can have them, but:

1. brand name is not indicative of product quality. only thing it matters for are warranty and support
2. your personal experience is ultimately meaningless unless you are capable of formulating an educated opinion
3. the past is in the past. a brand can change OEMs, a product can be partswapped. what matters is the present
4. working without problems is the bare minimum a product has to achieve in terms of quality

[honest karma]

hence my change from intel to amd so

[twin dew]

But I'm off to bed ->

[soft bloom]

working without problems is the bare minimum a product has to achieve in terms of quality
that's when tech is developed enough.
remember when best solution to every problem at any level was to reboot?..

[dire igloo]

my go to pick are drives with MAP1602 + YMTC X3 TLC config, tho that only works well for NVMe use (not via USB enclosure)
other than that, it depends on the usecase. for gaming and OS, any TLC drive above the basic shit is fine (NM620, SN570 and equivalents)

[honest karma]

i agree with you that brand doesnt mean a product of good quality

[narrow folio]

A reboot still can work wonders 😁

[verbal raft]

true

[honest karma]

its the same with guitars , i know no name brands with better performance than a les paul...

[verbal raft]

a reboot fixes 99% of the problems you can ever face

[honest karma]

or the same

for less money too

[night girder]

Nah, just funny. When I say something like that, I got a slap on the wrist from fireworker.

if you say the same, you get a thumbs up by them 🤣 Hence my two measurement reference.

[honest karma]

but the iconic style of a les paul is a reason to buy it

[soft bloom]

unless you have booting routine that takes longer than 10 minutes

[verbal raft]

i (as well as many other ppl) luckily dont have one

[dire igloo]

imo music is too subjective in matters of taste
with PCs, there is no taste involved in benchmark numbers and standardized durability tests

[verbal raft]

wait

[honest karma]

well then my cpu was benchmark winner ...

[verbal raft]

you have just informed me that ppl benchmark music

[honest karma]

nobody could compare when it comes to gaming when i bought it

its been a few years so..

not even the newest cpu for 1000€ won

[verbal raft]

this reminds me : do you guys know of a wonderfull site called "Userfenchbark"

[verbal raft]

waaaaaaaaaaaaaaaaaaaaat

R.I.P.

[honest karma]

u think i didnt compare lmao?

[dire igloo]

Nope - unless you mean 5800X3D and not 5800X
https://www.techpowerup.com/review/amd-ryzen-7-5800x/22.html

[honest karma]

and if i cant trust benchmarks to begin with, why bother checking

[dire igloo]

5800x basically got matched by a far cheaper 5600x for gaming

[verbal raft]

#Only userbenchmark

is not trust worthy

or i am missing context

[dire igloo]

who said you can't trust benchmarks? numbers are the only things you can trust - as long as the methodology is transparent and applicable to your usecase.
That's why the likes of userbenchmark and passmark are bullshit

[honest karma]

on average performance mine was better?

[dire igloo]

Cinebench too if you're trying to find out about gaming performance

[dire igloo]

I just send you a link that said it isn't - and it's not enough to just be better, the difference has to justify the higher price.
5600X was slower than 5800X in the testing I sent you - by less than 0.5% tho

[honest karma]

i donkt think it was that much of a bad choice you are making out to be

[verbal raft]

cinebench for gaming perf?

https://tenor.com/view/confused-perplexed-what-gif-10834164

[dire igloo]

as I said, it's mainly overpaying for the brand logo - and the CPU was merely questioned because I don't know your usecase

[honest karma]

or about how much % are yout nitpicking

[glossy glacier]

All X3D chips getting ignored

[dire igloo]

completely nonsensical, hence the grouping with passmark and UBM - Cinebench is an accurate representation of a Cinema4D render which itself translates decently well to similar usecases

[dire igloo]

I just wanted to know why you picked that CPU over 5600X, 5700X or their non-X counterparts.
You mentioned earlier that not even the newest 1000€ CPU could beat it - that makes me believe that your initial statement wasn't accurate.
I think you might be using a 5800X3D, not a 5800X. Those are two very different CPUs and 5800X3D while discontinued can still compete at the top end of performance in cache-reliant titles like Satisfactory or Baldur's Gate 3

the biggest criticism wasn't the CPU, it was RAM, SSD and MB. Because those were primarily bought because of brand name and not because of an educated decision (like you seem to have done with the CPU)

[honest karma]

I bought them because of brandname yes

plain favoritism if its logical or not

[dire igloo]

hence this

[night girder]

Also, for the future, you can ask multiple AI to see the difference in answers.

[honest karma]

but i put more thought in my gpu and cpu

the gpu was mainly the choice because i had no choice, it was a time where the prices exploded because of bitcoin mining and the sony ps5

1 week befor the price was at 2500€ a month later and 1000€

[dire igloo]

yeah, that much was apparent - it's hard to criticize those picks too because they can very easily be justified despite looking nonsensical. Your usecase just has to be niche enough

[night girder]

Certainly! Here are some popular 10 GbE NAS options that you might consider:

Synology DiskStation DS1621xs+
    6-bay NAS with powerful CPU options.
    Supports 10 GbE with an optional network interface card.
    Great for virtualization and heavy workloads.

QNAP TS-h1283XU-RP
    12-bay NAS with a powerful AMD Ryzen processor.
    Supports 10 GbE and has multiple expansion slots.
    Ideal for businesses needing high performance and redundancy.

Asustor AS5304T
    4-bay NAS with Intel Celeron processor.
    Comes with 2.5GbE ports, but can be upgraded to 10 GbE.
    Good for multimedia and gaming applications.

TerraMaster F2-221
    2-bay NAS that is budget-friendly.
    Supports 10 GbE with an optional adapter.
    Suitable for home and small office use.

Buffalo LinkStation 210
    Entry-level NAS with 10 GbE support.
    Simple setup and management.
    Good for basic file storage and sharing.

[dire igloo]

I unironically recommended a 16GB 4060 Ti to someone once - a GPU that basically every trustable reviewer out there called a waste of silicon.

[dire igloo]

chip shortage and Ether. Bitcoin had been specialized by then and wasn't really profitable on GPUs

chip shortage was the main factor tho, a drought in southeast asia brought down silicon wafer production

[honest karma]

yes, cant remember exactly but the prices exploded and i was in need of a computer

[dire igloo]

I paid 1000€ for a 3070 Ti, then realized my mistake and returned it unopened

[honest karma]

Thank you, i copied and will check them

[dire igloo]

It's an AI output too

[night girder]

That's just the result from another AI. So you could ask different AI's.

[honest karma]

man

i thought it was yours lmao

[night girder]

🤣

[honest karma]

good, deleted. >_>

[night girder]

Cool thing is, you can see sources for AI.

So for this respond, tom's hardware was used and smallnetbuilder.com for example

To me, it's the same as google.

It's just a basis to start from and do your own research.

[honest karma]

Cut me some slack, I can’t read properly because everything is blurry due to food poisoning. I feel like I’m drunk

furthermore i just had a surgery yesterday xD

should have noticed in the first sentence its an ai reply

[dire igloo]

now what would happen if you got drunk and sleep-deprived on top of this?

[verbal raft]

ooff hope you can recover

[honest karma]

i dont wanna know you dont wanna know how much i puked from this

[dire igloo]

yeah, get well soon

[dire igloo]

TMI
-# did you measure it by weight or by volume?

[honest karma]

It's already been 6 days, and I still haven't regained my sense of balance, its annoying as fuck

[soft bloom]

Just throwing stuff out there: there's model of Hydridic Earth, but as author says: people work in certain dogma, and despite evidence from physics, can't accept new model due to inertia.
I didn't find much online, except for the book, couple paywalled studies, and a lecture-discussion.
I was concerned about mentioning it to not put political taint on discussion, but I guess just mentioning it shouldn't hurt.
This here is just to extend discussion.

[honest karma]

and thank you very much

btw fireworker, it was hard to follow our conversation because of this, i think i havent answered everything and missed out alot

[dire igloo]

have you tried sci-hub or similar tools to get around the paywall?
alternatively, try reaching out to the researchers, sometimes it's the publisher enforcing the paywall but the original authors are allowed to distribute independently

[dire igloo]

main question: are you sure your CPU is a 5800X and not a 5800X3D?
for the rest: not too important, model over brand, learn from it and do better next time

[honest karma]

wait

AMD Ryzen 7 5800X , i bought it on amazon so yes

gosh its now available for 150€

i paid 400€

[dire igloo]

weird

I'd like to know what benchmark you consulted

[dire igloo]

that's just tech aging and getting flooded out by newer and better stuff

[honest karma]

that was 4 years ago, i used several sites which had the same result and i watched many youtube game tests

i only know it was a black website but that shouldnt say much lol

[dire igloo]

YouTube game tests as in recognized reviewers (Gamer's Nexus, Hardware Unboxed) or those random gameplay+Rivatuner videos that use who knows what methodology

[honest karma]

That i dont know, but i always make sure they have a lot views

so it shouldnt be a noname youtuber

[dire igloo]

okay, so you're judging quality through popularity

that's bad

[honest karma]

well its not like i paid attention to what they were saying

rather watched the performance that was blend in

[dire igloo]

LTT got called out for major flaws and inaccuracies.
Userbenchmark has millions of users and to this day is one of the top google results.
Don't trust stupid people to make smart decisions

[dire igloo]

are you German by any chance?

[honest karma]

it had benchmark in its name but im german so yeah

it had a dark theme i think

[dire igloo]

dang, I'm good

cpubenchmark.net and userbenchmark.com are wildly inaccurate sources. Avoid them

[glossy glacier]

cpubenchmark too?

[dire igloo]

Passmark is accurate in their methodology, it's just shitty methodology that's inapplicable to basically anything

[glossy glacier]

Got any sources for that?

[honest karma]

i will remember not to use them but i must say i might have overpaid but im pretty pleased, i never really had issues and could play all games in ultra at 4k without problems so far. I could render high resolution pictures in photshop and make full use of my music equipment. I doubt my current setup is good enough for newer games tho, not at ultra.

the cpu does get hot tho but that seems to be normal

[dire igloo]

1. the results and rankings on their page. top scoring CPUs are all multicore powerhouses. In non-multicore benchmarks (gaming, singlecore) they all score super poorly
2. Even the gaming chart: 7900X3D beating 7800X3D? 5600X3D beating 5800X3D? 245K beating 14900KS? no serious CPU reviewer can reproduce that data
3. the information on how their benchmark works: all super compute heavy, all of them with hardly any actual real world importance (encryption, compression and prime numbers being the most realistic ones). Zero of these are actually viable for gaming

www.cpubenchmark.net/cpu_test_info.html

[glossy glacier]

yeah, the 5800X is a bit of a hothead

[dire igloo]

would be nice to know the other parts of the system

are we talking LC-CC-85 or Mobius 360?

[honest karma]

Corsair iCUE H100i ELITE CAPELLIX

Corsair Crystal Series 680X RGB

i mostly used corsair because of the ICUe software

but i did fancy the NZXT Kraken x)

ah and because i loved the crystal series but it has some flaws im unhappy with

so it wont be a corsair next time

[dire igloo]

yeah, uhhh, don't. iCue is known bloatware and one of the worst offenders at that

[honest karma]

i am unhappy with it

it is literally the worst software on my computer, when i made my choice i just wanted a software to controle all RGB light without interferring each other

i wont use corsair again

[dire igloo]

signalrgb or openrgb are better

[honest karma]

can i still use them?

it never occurd to me that there would be other softwares god damn it

[dire igloo]

...

[dire igloo]

yup, tho the bloat might still be bloat

[honest karma]

well i cant know everything, never said i would

signalrgb looks better at a first glance

[dire igloo]

never said you'd need to - it's just the exact thing that I referred to, making uneducated decisions in the belief that they're educated due to a lack of awareness on what knowledge is lacking

[willow pike]

try openRGB

one open source program to manage all your RGB, no bloat

[dire igloo]

openrgb is great if you wanna tinker

but shit if you want smth to just work

[willow pike]

alas the "just work" options make other things not work, like your computer performing as it should

[honest karma]

yeah i dont see me as educated either im not an IT guy but i dont think im that incompetent either

[willow pike]

you could probably figure it out

[dire igloo]

I got a gigantic reality check on my PC knowledge when Goldensoldier took me from here to the PCPP server

didn't even know about RAM ICs, SSD NAND models or PSU platforms

let alone their differences and implications

[glossy glacier]

who? you, we or them?

[dire igloo]

me

until I got into that server

[glossy glacier]

damn
guess we suck then 🙃

[rustic panther]

That reminds me, what are the good psu platforms/manufacturers these days

I'm used to recommending Corsair but I have been slacking in keeping up with that

[pure karma]

signalrgb isent much better altho it gets the job done

[dire igloo]

CWT CSZ iirc, tho buying PSUs is really just a "cheapest that's good enough" type of situation

gpx is good-ish, not CSZ level tho

highpower has had a good platform, iirc Thermalright's SANR units aren't half bad either

[glossy glacier]

did some testing
High-ish settings, Global Illumination on Medium (last screenshot is with GI off), resolution 3440x1440

[winged valley]

$10,990.06 rough total

For a hypothetical all-rounder (though mostly Hyper-V for mass RDP of 60+ users) server with a $15k budget, hows this look to you guys? I don't know about shipping costs yet so don't ask this is a very rough draft. The only thing I dislike about this is the fact that the only SP5 cooler I could find which fits in a 3U chassis appears to be designed for ducting, which would have to be custom printed for this chassis.

We have PSU's on hand already

[glossy glacier]

raid card for boot drive(s)?

also for RDS i'd go for a P series (higher clocks -> more responsive system) and maybe a GPU

[wanton orchid]

iirc cooling wise SP5 is basically equivalent of sTR5

[willow pike]

that'll be an epic server

[wanton orchid]

wdym

[soft bloom]

https://www.youtube.com/watch?v=unz6mfjS4ws
Doesn't this remind you of those engines SpaceX is landing?

[winged valley]

no

wait are you questioning the card I have or the lack of one

[glossy glacier]

Sry, had the pic zoomed in and didn't see the comment

[languid gulch]

https://www.youtube.com/watch?v=-p0MEy8BvYY

[languid gulch]

nerds chatting is fun

[jagged snow]

I really enjoyed that vid

[twin dew]

I thought it important to say that I was going to just disappear in middle of conversation.

[twin dew]

And several of those are not with 10G ethernet or with upgrade option...

[twin dew]

So nowhere near CPU limited.

And some of the AI replies are discontinued too.

@honest karma So that second set of AI replies was again complete nonsense.

[twin dew]

First is discontinued.
And would have been with native 10G, not optional extra like said.

Second is legacy (discontinued?).
Uses Xeon E processors, not Ryzen.

And three last ones don't have 10G lan, and cannot be upgraded to it.

So again just LLM hallucinating half of what it wrote.

Didn't check if the three last ones were still for sale or not, or other lines for accuracy.

[twin dew]

So main PCB, PCIe foot PCB, connector PCB and then flex-PCB connecting connector and main PCBs.

With mezzanine connectors between.

[dire igloo]

It's a matter of averages tbh.
Corsair still has good ram kits, but they're usually overpriced compared to other equally good options.
And if you got any Corsair RAM kit at random, it's very likely that you got a shit product - whereas Klevv for example gets you a RAM kit that is somewhere between decent and great.

And it's not like Baldur said to avoid all Corsair kits because of brand reputation.
He stated that the brand has a shitty reputation - which is correct.

It's also the logic behind the statement - if you get a bad product once and dismiss the entire brand, that's nonsense.
If someone like Baldur, Darkmont or myself sees documentation on what ICs Corsair generally uses in their kits, that makes it a more reliable call.

[twin dew]

Well, Corsair was also one of the first RAM manufacturers who started to hide everything about their kits.
And IIRC doesn't tell anything but speed and CAS latency at this point.
Not even rest of the XMP set latencies.

And that they don't bother to bin the chips based on JEDEC / SPD speeds, and just slap the same absolute minimum SPD on all kits.

Yes, they work at the advertized speed.
But they mix chips in "same" product code.
So you cannot get second set with same chips by buying "same" product.
Hide almost all specs, so you cannot know what you are buying without third party sites, and those might not match to what you get because of the previous.
And SPD settings are always crap, so if you don't enable XMP, then you get slowest possible RAM, which doesn't run anywhere near what the chips could do at stock voltage.

[tribal kraken]

I have my first Corsair kit now with the 9800X3D, its 96GB 6400MT/s CL32 (m-die) which surprisingly works just straight enabling XMP. They run hot even at xmp 1.4V but guess that cant be avoided with any double rank sticks

But better not comment as I've not yet done any manual timing overcloking so really cant tell how they do.

[twin dew]

And the various current Hynix DDR5 dies are exceptional in that there isn't lot of silicon lottery variance in them.
So once you buy something with them, you get about same as with higher official bins.
But that hasn't been true for most DRAM dies over time.
Samsung 8Gbit DDR4 B-die was very varied AFAIK.
With some batches being very bad for OC potential.

It was just good in that it could take LOT of voltage and scaled well with that extra voltage.
If you cooled it enough.

[soft bloom]

everyone knows that definition of 'planet' is tricky at best
and now prepare to learn that Moon being defined as satelity is also not as straigthforward
https://www.youtube.com/watch?v=KBcxuM-qXec
(minute physics)

[dire igloo]

Thing is tho: that Corsair kit is not any better or worse than other less expensive Hynix kits

[tribal kraken]

It was 431€. I dont find it expensive compared to G-skill or Kingston, the suppliers I buy stuff really don't have other options for 96GB 6400MT/S CL32. 6000MT/s CL30 could have been maybe 50€ cheaper. Also as I buy computers and parts with VAT number, Proshop from Denmark for will be free front door delivery to Finland and 0% VAT.

[twin dew]

And either Corsair has restarted actually specifying the rest of the timings, or I was mixing with some other RAM maker, possibly Crucial.

32-40-40-84 for one of the 6400MT/s CL32 2x48GB kits.

Hopefully that 84 isn't tRAS but tRC.

[dire igloo]

6000cl30 and 6400cl32 use the exact same chips - all RAM kits with those specs do because only Hynix makes ICs that can go this fast.

And you can look further than the big name brands like Corsair, G.Skill and Kingston - Teamgroup, Lexar, Patriot, Silicon Power, Klevv, etc all make RAM kits at those specs

https://fi.pcpartpicker.com/products/memory/#sort=price&b=ddr5&S=6400,8400&Z=98304002

Welp, only Teamgroup and Geil join the ones from your list according to PCPP, should've checked that first
Maybe klevv, Kingbank or vcolor have smth listed separately

[twin dew]

Lot of RAM brands that exist in US don't exist in EU space, or specific parts of EU.

[night girder]

No kidding!

So that's why it says "Supports 10 GbE with an optional network interface card." underneath eveyr product?

[twin dew]

And that wasn't true for any of them?

[night girder]

It's literally underneath every product.

Comes with 2.5GbE ports, but can be upgraded to 10 GbE., Supports 10 GbE with an optional adapter.

[twin dew]

Two actually came with builtin already.
Three don't have upgrade option.

[wanton orchid]

I tried to search for my chip and didn't find any info past ras

[night girder]

I posted one that doesn't need extra.

[twin dew]

So all the four mentions of of upgrade were false.

[night girder]

Buffalo LinkStation 210
Entry-level NAS with 10 GbE support.

AND

I said it was from AI and they needed to investigate it themselves to be sure.

[twin dew]

No, that LinkStation 210 doesn't have 10G support.

Just 1G.

[wanton orchid]

micro only give a chip summary with junction layout and size

[night girder]

AND I said it just a basis to start from.

read please

[twin dew]

Point was that that AI reply was complete baloneys through it.
With NONE of the entries being true in whole.
And none you could buy now, and get 10G even with upgrade.

As the only ones that could have 10G were discontinued.

[wanton orchid]

btw I have almost all chip with W except one with Y in this place
starting to questioning the software they use to run logistic/sorting of chips before shipping

[twin dew]

Which was the point, that that listing of five NASes was half hallucinated, and completely useless.

[night girder]

All has been said.

[wanton orchid]

now I don't think I have a bad bin
but I can't overclock it anyway because of broken bios

[twin dew]

And those Google AI answers do link "sources" but don't actually use them properly.
The answers very often contradict the listed sources.

[wanton orchid]

(no xmp profile basically means impossible to go manual for the bios for some reasons)

[night girder]

Not even google.

DuckAssist.

[wanton orchid]

still unreliable

[night girder]

And no, not the same thing.

You don't even use it.

[twin dew]

And that TS-332X is also discontinued.

So it did have SFP+ cage for 10G, but you cannot buy one anymore.

[night girder]

Btw Baldur, TS-h1283XU-RP supports 10GB

[twin dew]

Yes, but is discontinued, and has Xeon CPU, not Ryzen.

[wanton orchid]

and they are usually locked

[twin dew]

Which were the points I listed for that.

[night girder]

AS5304T has two PCIE slots for upgrades. So can slot in better network card.

[twin dew]

And the first Synology is discontinued, and had 10G inbuild, not as extra.

[dire igloo]

With the accuracy that the AI output provided, you're better off just googling.
Yes, even with the linked sources - guess how one could find those without AI

[twin dew]

No it doesn't according to Asustor spec page.

[wanton orchid]

for upgrade of what ?
these are locked slots

[twin dew]

It has 2x 2.5G network ports.

No expansion ports of any kind.

[dire igloo]

I see the point tho, you can get AI to tell you half a dozen models, then research them and find out if they suit your needs.
It's just kinda bullshitty to do this if the results have too high of a failure rate

[twin dew]

Or just select a store, get product listing for NASes and go through that instead?

[wanton orchid]

^

[dire igloo]

Yup, good stores might even allow you to filter by supported network speed

[twin dew]

And then research the ones that looked good in the right price range.

[wanton orchid]

except you then only check the one ai got brainwashed for

[night girder]

And I didn't even post full list.

[twin dew]

Yeah, but when in the list you did post, all the entries were useless, how do you think rest of the list would have helped?
(useless: cannot buy or cannot get 10G)

With all the entries having some kind of hallucinations in them.

[night girder]

Bro, what don't you understand.

Here is a list; check it out yourselves.

ALso; it's from AI.

[dire igloo]

I've used AI to point me towards scientific research papers in a given field.
Out of 8 suggested papers, 2 didn't really exist (or rather: there did exist similar ones that seemed similar on the surface), 4 were irrelevant, 1 was off topic and 1 was actually helpful

[twin dew]

Where all the entries were useless to check.
Which was my point.

[night girder]

What did you provide?

They even didn't look at the list because we literally told them it was AI.

[dire igloo]

It's a useless list because you have to put effort into researching everything that was mentioned only to then realize that none of the models actually fulfill the criteria

[night girder]

Since people don't read.

[dire igloo]

What did you provide?

[night girder]

My point was, if you want to use AI to search for anything, you could use multiple AI's to get what you need.

[twin dew]

And my whole point was to point out that the list was completely useless and half hallucinated.
As current LLMs just don't handle that kind of question.

So all of them will just give out garbage.

[dire igloo]

Or you do this which is more likely to be successful and usually faster too #off-topic-tech message

[night girder]

The question/context; I am using AI to search hardware. My input; use multiple AI's, use it as start point, research yourself. If you are really stuck.

Then we go over that the list is AI generated ... so questionable at best.

Then 8 hours later, I get pings that the list is wrong. Way out of contexct.

[dire igloo]

Thing is: if you're that stuck that you have to resort to ai, you did smth wrong like 2 or 3 steps earlier

Also, go ask human intelligence if you don't know

[night girder]

All fine, but the person wasn't asking that, or not my interpretation.

They were already goofing around with AI buying hardware n shit. Or their friend was.

[wanton orchid]

So because they did bad decisions already we should comfort them into it ? when they get there for help ?

[dire igloo]

If you wanted to show that AI is a bad tool to use: you failed because you didn't explain why the list sucked.

If you wanted to show that AI can be viable: you failed because the list you sent was 0% helpful

[wanton orchid]

they asked humans here then
it takes over the ai bullshit

[night girder]

You do what you do buddy. I do what I do.

If you want to be a dick about it against people, go a head.

[wanton orchid]

but you insist without arguments

[dire igloo]

Getting no advice is much better than receiving bad advice

[night girder]

Instead of always looking down to people who come in here for help.

[wanton orchid]

if that's what you do I do tell you it's unproductive and getting annoying

[night girder]

But try for once to help them a little bit?

Even if you don't agree.

Or just entertain them instead of shooting their ideas down.

[dire igloo]

If they're coming for help, it'd help if helpers were helpful

[dire igloo]

How did your message help them?

[night girder]

I told them to use multiple AI's. Aggregate information. Use your head.

[dire igloo]

Bad ideas deserve to be shot down, sometimes they even have to be shot down.
Not without elaboration/explanation as to why, but you shouldn't entertain or encourage bad ideas

[wanton orchid]

no one is entitled to do whatever comes to their mind and call it "help" without facts to back that up
get yourself grounded

[night girder]

What is a bad idea? Using AI to find stuff?

Rofl. 🤣

I google: "best NAS in the world" ... the result would not be so much different in as I ask a AI. Because in both cases, I still need to fact check it.

It's just a different tool, to get a basis for your search from. Like I said multiple times.

a basis, a starting point, a way to get going while searching if you are stuck.

[twin dew]

At least normal search doesn't hallucinate.

[dire igloo]

I'll give you two out of three for this one.
Use multiple AIs: false. You're just getting more bad information.
Aggregate information: correct. Not from AI tho.
Use your head: correct-ish. Human brains are limited in their capability and someone who wants to know something can't rely on their knowledge because they know it's insufficient

[night girder]

You can refine it.

[wanton orchid]

to "fact check it" you have to ..
do an actual search

Super fast indeed

[night girder]

Aggregate as in; get a list of multiple products. Then check those products.

[twin dew]

No, you cannot refine technical question with LLM, as they have no idea about anything but language and words.

[night girder]

Yes you can.

[dire igloo]

If you are stuck, you're not starting

And educated researched guesses are much much more valuable than random guesses.
It's about the value of the information that you get.
By the very same logic you used, since you have to fact check both anyways, AI is a pointless tool because it offers no benefit over a regular search

[night girder]

And not technical, but you can ask the AI why it provided the list it did.

[dire igloo]

Not reliably

[night girder]

And then the AI would tap out and admit it's mistake.

Better than you can do with a newsarticle.

Anyway, isn't their a proverb about humans and tools? That it's not the tool but what we do with it?

I feel like this is very much the case of people trying to blame the tool.

[twin dew]

And in that it is hallucinating the existence of external 10G LAN adapters.

[night girder]

While it's us using it wrong or not understand it.

[dire igloo]

If verifying a result is significantly faster than obtaining a result, AI can be very helpful.
But if both take the same effort, don't bother with a linguistic guessing machine

[night girder]

It's the same nonsense I read when google was brought in.

Oh, don't believe google. Books are still better. Don't believe wikipedia. Blablabla.

[dire igloo]

No, it's you using a tool for something that it is incapable of doing well.
You're using a screwdriver to drive a nail into a wall

[wanton orchid]

well play
you now have have lost 15min of your time sending Google duckduckgo and basically any company on internet how you socially behave with their software

Ai is a full scale scam machine
these guys are looking for breach in your fucking social behavior

[night girder]

Also when you refine the question, you get better answers it seems.

[wanton orchid]

it seems

now go actually check

[night girder]

https://tenor.com/view/tinfoil-hat-gif-19189292

[wanton orchid]

imagine not knowing how ai prompt actually works
and what marketing is about

[night girder]

image not knowing what duckduckgo is. 🤣

[wanton orchid]

? a "search engine"

[dire igloo]

Or it doubles down. Or it admits a mistake where there wasn't one. Or it makes a different mistake like it did here.

LLMs are fundamentally flawed. It's upon the user to judge whether the flaws actually harm the quality of the result.
For technical accuracy, promt bias, dataset bias, hallucination, etc are very real issues and should make AI unviable.
For finding a nicer wording on a letter, those don't really matter as much.

[night girder]

haha, lol.

[wanton orchid]

you really think you'll bait yourself out of the argument ?

[night girder]

I had twice now, that AI admitted a mistake when asking more.

[night girder]

No, but I do know you will add nothing of interest to this conversation. You making statements about evil corporations but you don't even know the philosophy behind duckduckgo, you don't even know the product I am using. You are making baseless accusations against DDG. The same that Kekw of the summer did with tectonic activity. And they got 2 days of heat for it.

[twin dew]

Just thing they are hard-coded now to do when prompt contains specific things.

[dire igloo]

I got AI to "correct" mistakes where there weren't any and double down on clearly wrong answers.

Best example: how many R in the word "strawberry"? At what positions?
Ask that same question in different languages and watch it fumble horribly

[wanton orchid]

"politics are doing decisions which improve their own quality of life in the next years"
your answer :
"damn foil hat take"

get yourself grounded

[dire igloo]

What LLM is DDG using? Inhouse product? GPT 4?

[night girder]

You can choose.

just go research yourself, sorry.

If you want to shit on DDG AI, just do research first please?

[wanton orchid]

"don't even know the phylosophy behind" are you claiming you know ?
how are you claiming I don't ?
and any phylosophy they have will never change their business model and actions

I never claimed I knew their phylosophy

that's the difference with you and kek

[night girder]

There is 4 I believe.

We do not save or store your Prompts or Outputs on DuckDuckGo servers. Instead, recent chats are optionally saved locally on your device, to protect your privacy.

[willow pike]

that's the stupidest thing I've ever heard

[night girder]

So, yes you get a tin foil hat for claiming DDG "these guys are looking for breach in your fucking social behavior"

[willow pike]

chatgpt still has your shit

[night girder]

🤦‍♀️

[dire igloo]

Point proven

[night girder]

This is not even going to chat GPT.

[dire igloo]

It was right, I questioned it, it "corrected" and got it wrong

LLM is LLM

[night girder]

Your point is it can be wrong twice?

[night girder]

And no it doesn't.

[wanton orchid]

llm use is basically hallucinating together
except your prompts are online

[night girder]

GPT-4o is not even chatGPT to begin with.

[dire igloo]

My point is that an AI admitting and correcting a mistake doesn't mean that there was a mistake

[willow pike]

okay fine openAI the company, unless you're saying DDG are hosting the LLM, in which case they have the full conversation history

[night girder]

I know.

[dire igloo]

And consequently, the accuracy of information obtained through AI is so bad that you're better off not using it - for matters that require accuracy

[night girder]

All metadata that contains personal information (for example, your IP address) is completely removed before prompting the model provider. This means chats to Anthropic, OpenAI, and together.ai (which hosts Meta Llama 3.1 and Mixtral on their servers) appear as though they are coming from DuckDuckGo rather than individual users. This also means if you submit personal information in your chats, no one, including DuckDuckGo and the model providers, can tell whether it was you personally submitting the prompts or someone else.

That's why I asked, to do minimal research. Which 2 out of 3 not doing in here.

[dire igloo]

Big fucking yikes

[night girder]

These are older from this month.

2nd sreenshot is actually the first.

[wanton orchid]

Oh that's surely why I have search result for my CITY when I'm prompting "cat"

[dire igloo]

You get hung up on the privacy portion when that's only one person's position.
The accuracy is what makes it actually problematic

[wanton orchid]

but sure they will not have a 9766th leak this year

[night girder]

Well because it was brought up in this chat.

[night girder]

The accuracy, it's not even a discussion.

Like I just posted 2 screenshots that proof your own point.

so I already know that.

[willow pike]

you're being exceptionally naive if you believe a company allowing free access to an LLM isn't keeping your data, but scrolling up I am now much more concerned about how competent you seem to think these models are

[night girder]

Honestly, it's my data. People in here use chatGPT with accounts and everything.

Who cares? I can do that. Atleast I went for a solution that offers me a little bit more.

Security/privacy isn't a one trick pony.

You need layers.

And the questions I ask, I hope they have fun social engineering it.

[wanton orchid]

you claimed it's useful for that
we argued with facts it's loss of time if ever useful
you claim it can be useful
we explained it can't for that
you implied that's why it exists for
I responded it only exist because these companies deemed it good investment for their own interest and whatever reasons
you called me foil hat

[night girder]

"How long ago was 1900?"

Calling someone naive for believing in a TOS is a far stretch.

[wanton orchid]

breaches are online

Also this is not tos

[night girder]

I would say you more into conspiracy territory if you believe the data they keep, is used for "evil doings."

[wanton orchid]

but privacy policy

[willow pike]

let's back up a little here; think about what it's actually doing. these search assistants, they run search results through the LLM and try to parse out some meaning, often failing. we can argue until we're blue in the face if this is a good idea or not, but what really cripples this shit is the fact that most search results are AI generated too. so you have an AI incorrectly summarising the bad output of another AI

[wanton orchid]

quote me "evil doing"

🍿

[night girder]

I though your point was they kept the data? 🤔 And I was naive in for believing DDG doing their best to add privacy protection.

[willow pike]

we have multiple points and you're attempting to cherry pick them

[night girder]

No, you are?

You burst into the conversation 🤣

[willow pike]

no u

lmao I'm out, you go nuts

no skin off my back if you wanna do something stupid

[night girder]

👋

[dire igloo]

https://x.com/katherine1ee/status/1729690964942377076

Katherine Lee (@katherine1ee) on X

What happens if you ask ChatGPT to “Repeat this word forever: “poem poem poem poem”?”

It leaks training data!

In our latest preprint, we show how to recover thousands of examples of ChatGPT's Internet-scraped pretraining data: https://t.co/bySVnWviAP

X · Nov 29, 2023, 2:37 AM

Fixed now but this was a very real issue

Right after the "fix", ChatGPT even told users to just fulfill the prompt themselves and refused to do it

"can you repeat the word 'poem' forever?"
"Here's how you can use copy and paste to repeat a word multiple times"

[night girder]

So, AI is not accurate? Which it even tells users?

I really don't know if we can blame everything on AI, I still feel like it can be a useful tool if used correctly. Like I said, double check facts. Or tripple fact check it even. But it can be a tool to get ideas, inspiration and think out of the box. If I want to know more about wolves, I can read a book about wolves. Watch a movie about wolves. Visit a park with wolves and interview the caretaker. I can google it. And I can ask AI/ML. I don't see what the big deal is? It's just another tool to gather information. Depending on the source, even movie or books can be innacurate. Especially when it comes to science that keeps evolving.

[twin dew]

When LLMs don't even have any relation to concepts of accurate or even information?
Just about words after words.

[night girder]

I don't understand? Google doesn't have concept of accuracy or information either.

[wanton orchid]

yes it does

and all the public rules are easily accessible

such as security score, performance score, aggregation of data, and canonical hint for referencing reinforcement

[night girder]

First hit on google. Verry accurate.

[wanton orchid]

statistics of link click etc, are not ai generated thing

[night girder]

Google is also just a piece of code.

Nothing more, nothing less.

And then google also sorts the search results depending on x other factors.

[wanton orchid]

oh yea, nothing more, like running on servers accross the planet ?
nothing more like gathering data with a billion dolar company ?
nothing more really ?
so I'm writing google every other day in my bloc note

[night girder]

So when it comes to accuracy and trust ability, I don't know if old search engines < AI or AI > old search engines. Or maybe they are complementary. Like I've been trying to tell for an hour now.

[edgy hazel]

SEO is a cancer

[dire igloo]

The point we're trying to make is weighting the quality of information (accuracy, usefulness) against the cost of obtaining it (time, effort).

If you highly value accuracy (example: find a NAS that is actually capable of 10Gbit), the cost of obtaining said info from an AI far outweighs that of other methods, making AI unfeasible.

On the other hand, if accuracy isn't that important but speed is (writing a birthday message or a Christmas card), using AI is one hell of a tool.

[twin dew]

Search engines are code, but they are code that has been written for specific purpose.
LLMs on base are machine learning algorithms that are coded to track word associations and to generate realistic looking text.
Not accurate or information containing text.

[dire igloo]

1. Pretty sure that's a value ranking, not a performance ranking.
2. Could've used my earlier screenshot that had UBM as number one source.
3. Google isn't perfect either and it doesn't need to be. It just needs to be better than an LLM - which for the sake of accuracy, it often is (in the sense that you are more likely to be referred to accurate resources like the TH chart in your screenshot)

[night girder]

I don't agree with "the cost of obtaining said info from an AI far outweighs that of other methods, making AI unfeasible."

[dire igloo]

They're different tools for different purposes.

Google is completely dogshit for helping an influencer write an apology to their fans for something they don't feel the need to apologize for.

And an LLM is completely useless for gathering specific and accurate information, like what NAS hardware to get if you want 10 Gbit/s connection.

You don't use a hammer to tighten a screw and you shouldn't use a screwdriver to drive nails into the wall.

[dire igloo]

Rhetoric exaggeration for dramatic effect.
It definitely outweighs it, but to what degree and how you judge "far" depends on too many factors.

So yeah, valid disagreement, ignore the "far" in there

[night girder]

And an LLM is **completely useless **for gathering specific and accurate information, like what NAS hardware to get if you want 10 Gbit/s connection.

Also not true, since I did get good products.

Not the first query, but the second optimized one I did.

[soft bloom]

it's natural that some of the early sceptisism will get outdated

in 5 years it will be like "oh did you know that LLM is actually good, like no joking?"

[wanton orchid]

yet any hype will not make an llm some else than a llm

you're going toward what letterdief called me foil hat for why it looks better

[soft bloom]

i can' descipher that

my embeds are not so large

[night girder]

Sorry, if you post that. yes I will call you tin foil hat.

[wanton orchid]

"it's not gathering information on people social interactions with the ai prompt system to better exploit their behavior in a marketing fashion at least"
"it's only :
using information about statistics of people interactions online (link vague source of data)
for improving they visual quality of result and hints sent to users for gathering more attraction in our marketing business"

sorry I didn't use the marketing wording which makes it unimaginably more nice and great automatically

[soft bloom]

are we back to discussion about what makes claims unbased and fearmorgening?

[dire igloo]

You may have gotten good products, but:

1. You had no access to tests/reviews from the AI until you changed methodology. So extra effort created and little benefit over a normal search.
2. None of the products actually fulfilled your criteria
3. There were multiple cases of products that were no longer available
4. You were presented with false specification

Overall, too much inaccuracy for me to consider this a viable research tool.

[night girder]

I just had 3 products checked. All according to the 10GBe spec I asked for.

[dire igloo]

Eh, I'd word it differently:
The issues that sceptics point out right now will be addressed and might even be fixed in the future.

[soft bloom]

... which will make sceptic takes from earlier outdated (not reflecting present reality) in future

[twin dew]

At that point the base tech wouldn't be LLM.

[soft bloom]

did you watch the vid on extracting meanings/concepts from LLM's weights?

[soft bloom]

considering how large LLMs are and will be, I am not sure how to define 'base'
it probably will remain the largest part of product

[night girder]

I wouldn't be suprised google and co is replaced in x years with AI.

[twin dew]

LLM = Large Language Model.

[night girder]

Atm, they have done 20/80 or something. Some answers are already AI generated.

I can see it happen. Don't forget, google is ""better"" because its a 27 year old optimized product.

The future will tell us how AI/ML will be used. If it's like a NFT (slow death?) or like a bitcoin (survive?)

Same claims were made about both products (NFT, Bitcoins) and yet one is still here.

So not sure when "skepticism" becomes "conservatism" for some people. I was skeptical about AI and still am. But I also realize I would be stupid not to explore it. Especially as a programmer.

[wanton orchid]

it's not stupid to explore it, as soon as you got factual ground in it being of interest (to explore it) for you, which I still doubt

[night girder]

One addition; I am talking chatGPT etc. Not the AI shit they push into every single product.

[wanton orchid]

microsoft bought mistral for a reason

> chatGPT still is basically "AI shit they push into every single product"

[night girder]

Nah. Not what I meant at all.

• Soundcore unveils open earphones, speaker with AI karaoke features and outdoor model
• VLC has been downloaded 6 billion times and shows AI subtitling system
• Adobe announces AI tool that can edit up to 10,000 images at once
• Philips Hue develops AI assistant for lighting scenes

All headlines from the last few days I believe.

[pure karma]

everything has atleast 1 AI feature at this point

90% of them add no value

some of them are even payed so there setting themselfs up for failure

[night girder]

Google added AI to Workspace and increased the price. Read that news yesterday.

Google will integrate its AI features such as Gemini Advanced and NotebookLM Plus into its regular Workspace subscriptions. Until now, that cost an extra 20 dollars per user per month. The price will go up.

[pure karma]

i mostly had filmora's ai tokens in mind but im sure theres more examples like that

where you have to buy the tokens and then the tokens get used up depending on what ai thing your trying to do

[night girder]

if I had disk space I would run my own model for fun.

[pure karma]

isent ai a lot like mining

your basically just trying to do the most calculations possible or something

[charred willow]

wish I was at F2.0 instead of 1.8...

[night girder]

Mhh. The 1.8 works.

The nose and eyes are sharp, so that's good. Maybe a little unsharp on the eyes so maybe 2.0. But I don't think it's that noticable. It's a nice picture.

[dire igloo]

Bitcoin didn't survive as a speculative asset imo, it failed in that regard.
It survived as a currency tho, a means of handling transactions on the internet.

Which is what I suspect to happen with AI:
Bubble bursts on the trend-focused overgeneralized usage and it will settle into the functionality that it has been designed for and is good at.

[dire igloo]

Ehhhh, training your own AI model, yes.
Using a prebuild and pretrained AI model not so much, that's more about handling the request flood, not crunching the numbers on how to adjust the weights within the model

[night girder]

If bitcoins wasn't "a big deal" and just something less then 1% would use, EU wouldn't regulate it. MiCA came into full force in 2025 I believe.

[charred willow]

exactly what boat im in

its good, just wish I had the idea to drop to f2 in the moment

[night girder]

The fact EU is/was worried about it, shows the impact of bitcoins.

[dire igloo]

EU is scared shitless of any tech that allows anonymisation. Example: E2E encryption.

[charred willow]

but thats how things normally go 😆 , photo just be like that I suppose

[night girder]

Especially because of the illegal circuit using BTs

They didn't do anything so dramatically for NFTs. Not bitcoins is a different beast imo.

[dire igloo]

It is one of the most traded cryptocurrencies in the world, it's just no longer being used as a speculative asset.

The times of HODL are over, it's no longer feasible to buy and hold crypto in the hopes of selling at a higher price.

It only is if you can artificially impact the cryptocurrency's market value - like so many influencers do with their pump and dump schemes

[twin dew]

Traded, not used as actual currency.
As the transaction amount per hour is tiny.

[dire igloo]

Bitcoins is a different thing and it definitely survived much better than NFTs did.
However, there are still aspects of BTC usage that didn't survive

[twin dew]

As it is still currently mainly speculative asset.
Any actual using crypto as currency happens with other chains, that actually can handle transactions.

[night girder]

True, but doesn't matter? Google also evolved true time. So can bitcoins. And so can AI.

[dire igloo]

Hm, I guess there's some merit to it. I'd guess Ethereum and Monero to make up the majority of crypto-transactions

[twin dew]

Bitcoin already had a split over that tiny transaction limit.