off-topic-tech

[cyan crescent]

How is warframe on linux?

[edgy hazel]

Dual boot is a thing you know

[edgy hazel]

Like 10% faster

[cyan crescent]

Supported?

[edgy hazel]

Idk i just turned on proton and it works without issues

[dire igloo]

Dual boot takes effort and I'm lazy

[cyan crescent]

How do you dual boot?

[dire igloo]

You have two partitions, one for each OS.
Then you set it up in BIOS what partition to boot from by default - and some way of telling it "actually no, today I wanna boot this instead"

[edgy hazel]

grub my beloved

[narrow estuary]

I'm curious if there are any cpu benchmarks for satisfactory 1.0, ie wether the 14900 or 7800 lead the way or another

[twin dew]

Much simpler today.
One partition per OS where they actually live, then there is the "hidden" UEFI partition that has all the boot loaders, and UEFI can itself select between those.
Or you can set it to boot the GRUB by default and that then allows to also select Windows.
And Windows doesn't nuke that GRUB on the UEFI partition anymore like it did with old MBR.

[dire igloo]

X3D all the way for SF

Also:
14900?

[edgy hazel]

Except that Windows still fucking does that sometimes lol

[narrow estuary]

@dire igloo Thanks. I got a 7900x3d and 7900XT. Testing to see the limits of the game. Seem to find around 100k objects (scim) is when the cpu limit starts kicking in.

[twin dew]

Just sets the UEFI default boot to itself and not GRUB.
And can just be changed in BIOS/UEFI again.

Doesn't completely nuke the GRUB like it used to.

[edgy hazel]

https://www.bleepingcomputer.com/news/microsoft/august-windows-security-update-breaks-dual-boot-on-linux-systems/

[twin dew]

That was different.

[edgy hazel]

So windows didnt break dual boot?

[dire igloo]

7900x3d is worse than 7800x3d and 7950x3d

[edgy hazel]

It still did.

[dire igloo]

Also, 9800x3d is rumored to be announced early next month

[narrow estuary]

Has anyone hit the cpu limit with a 7800 or 14900 yet?

[dire igloo]

You can do that easily

4090 GPU, 720p Valorant, all low

Or you take a hyper endgame Satisfactory save and run that at 720p all low

[edgy hazel]

Just play city skylines

[narrow estuary]

Then be ram limited hehe

[tribal kraken]

All I need to do is stack/pile logs in Farming Simulator and that will load one thread to 100% and drop the physics engine calcuting the collisions down to 10fps. On a 13900K...

[gilded helm]

I've been kinda disappointed in CPU reviews for a while. Stuff like this is way more interesting than 500 versus 600 FPS in eSports games.

Meanwhile building/strategy games can get to multisecond hangs.

[twin dew]

GN at least somewhat tried laterly, but the selected game just doesn't work as it seems to now floor at 30 seconds per turn.

[wanton orchid]

games having multisecond hangs bave nothing to do with cpu
but poor game design
you don't want to see cpu saturation at all 'cost'
you want to see CPU scaling in relatively optimized workloads (which is what any critically cpu software will ultimately be or it will be a project management problem and not a cpu problem)

you don't want to benchmark the games on cpu

you want to benchmark the cpu with games

then adapt to whatever you know
it will be consistent because of the scaling benchmark consistency

[twin dew]

Point was that no-one is reviewing CPUs and including a test that has very high preference for the extra cache.

Like for example Factorios benchmark mode has, Satisfactory has, many other simulation games have etc.

GN tried, but they chose turn based strategy game, and measuring enemy turn times.
But that isn't deterministic, and they chose too "early" save so it floors with current CPUs already.
As the late game saves were too chaotic in run-to-run variance.

[pure karma]

ahh yes my favorite 600W on a mobile gpu

[verbal raft]

on the 4050

what

bugged?

[cyan crescent]

How the fuck?

[twin dew]

Anything electronic that MagicZ touches starts to glitch.

[pure karma]

i told occt to run everythign at once and it broke the readings on hw info lol

[dusky oyster]

Average TDP of the next Nvidia GPU generation:

[verbal raft]

😂

[edgy hazel]

man...

Just... stop touching GPUs

[pure karma]

the worst part is im not even trying... i have been busy trying to figure out how to unshitify my RGB

[tough owl]

I love no SMT on new Intel chips

[verbal raft]

"unshitify" i like it

[verbal raft]

why is that so good? i thought SMT usually made performance better

[tough owl]

https://tenor.com/view/imagination-spongebob-meme-gif-1852791965001335176

[twin dew]

It gives more performance in programs that have more threads than the CPU has cores.
But can lower performance on single-thread limited cases if the scheduler fucks up, like Windows one often does.

[mental oriole]

Wandows scheduler =

[charred willow]

Phis smt4 makes windows apps that are very single threaded crash sometimes 😆

[visual tree]

Had to switch from Android to iPhone because the phone started dying after 6 years of use and my sister gave me her spare iPhone

May God have mercy on my soul....

[gilded helm]

Did any Arrow Lake reviews use Satisfactory? 😦

Kinda tired of seeing Plague Tale make reviews, even though that's a benchmark disguised as a game.

Timespy is more useful as a frame of reference because anyone can download and use it.

[twin dew]

Almost certainly no.
But seems thsoe are worse than 14th gen on the games that love the cache in X3D:s
In case of Star Wars: Jedi Survivor, because of bad coding:

[gilded helm]

If someone comes out with a really good 9800X3D laptop, I could be interesting in 2025. But I don't have my hopes up.

[night girder]

How good are $40 cooling pads for laptops if you want to overclock?

[charred willow]

how much % gain is your OC worth on your laptop tbh

[night girder]

Not asking for myself, just want to know what people think.

[gilded helm]

It also depends on the particular laptop you're going to use. If it's not working well with your laptop's bottom vent and built in fans, you might be just as well with a $10 stand.

https://www.amazon.com/Adjustable-Silicone-Foldable-Ventilated-Tablet-Black/dp/B08G84P8ZX/

[night girder]

So it can be effective?

Shit, I assumed they were kinda garbage 😛

and like stupid to use if you want to OC a laptop.

I didnt' ask about the laptop, so I have no clue.

but since they mentioned it was for school, I have my doubts.

[gilded helm]

Undervolting laptops is the better thing to do, go for stock performance at lower voltages.

Or, if the laptop stand's fans are quiet, maybe you get the same overall performance, but it'll be quieter overall than a laptop's fans roaring full blast.

[night girder]

My suggestion to them was to lower their graphics in SF 🙈 The complain was: "my laptop is running hot when playing satisfactory, so I want to overclock it." There are a few flaws in this logic.

[gilded helm]

There's... a lot to unpack there, yeah.

[night girder]

but they said that a cooling pad was the solution for that, and it just made me wonder about em.

[gilded helm]

Broadly speaking you can wring a little more performance out of laptops, but you're not going to get to a new performance tier.

[pure karma]

thats sounds like a laptopified version of the overclock it so high it underclocks itself automatically if necessary overclocking method

defedently not going to improve temps tho lol

[night girder]

How do you overclock so it underclocks?

I know some stuff throttles at certain temperatures, but if you go too high, doesn't the system just shut down?

[pure karma]

its like overclocking a cpu to something dumb like 8GHZ it will either insta crash or ignore it and run as fast as it can

[wanton orchid]

but it's patented ! 🤡

remember the productivity road map consist in removing ability for everyone to ever legally sell something actually useful

[languid gulch]

yes, i made that one myself cuz it was too funny to pass on

[dire igloo]

Detrimental because if they provide less airflow than the laptop fans, they'll act as a wall and block the air

Use the normal stand without fans

[gilded helm]

Pat has plenty of bad execution under his belt at this point. But employees would go to war for him if the board starts sharpening axes.

[twin dew]

And the new CPUs have been in works for 3+ years.

[willow pike]

it's less efficient than the 5800x3d, tsmc 7nm, despite using tsmc 3E

[gilded helm]

Graphics are Pat's baby, he owns all the terrible hype of Alchemist, aka DG2. Remember DG1? If not, that's understandable. DG2 should have been given the same treatment as DG1, and whatever the first generation is that was actually ready should have been called Alchemist.

[willow pike]

i remember larrabee

[gilded helm]

DG1 isn't Larrabee. It's the OEM only discrete GPU predecessor to Alchemist.

[willow pike]

yeah i know, just saying I've seen some shit

[gilded helm]

Unfortunately it doesn't look like real competition is coming to the GPU market within the next 2 years.

[night girder]

Competition to what brand?

[gilded helm]

Nvidia. They're just going to be able to charge whatever they want, and continue to be the standard for developers as a result. It's mostly concerning to me because I'm on laptop, which also means fewer design wins for AMD. So my options are going to be even worse.

5090 laptop prices will probably continue to be meme tier lime 4090 laptop. For 377mm of GPU silicon if the rumors hold up.

[night girder]

Mhm. Nvidia is pretty dominant for sure.

[gilded helm]

Probably little top end performance gain as well. Same size die, same node family.

Hopefully segmentation in the rest of the product stack improves.

[willow pike]

people will buy the $2000 5090

how the fuck does this happen

[gilded helm]

They paid for N3 too. Embarrassing.

[stray badger]

I love my 7950x3d

[rustic panther]

That doesn't look right at all

But if it's true, hope Intel managed to lay the groundwork for more exciting stuff

[gilded helm]

In the end desktop DIY is a niche product. They need Lunar Lake to do well more than anything else.

[verbal raft]

Intels buldozer

[willow pike]

https://mobile.twitter.com/rogoway/status/1620588153500479489

Mike Rogoway (@rogoway) on X

My sources at Intel confirm. They say
* Suspended merit increases, suspended quarterly profit bonus, 401k match cut by 1/2
* Base pay for Grade 7 employees and above by 5% (grades 7-11) to 25% (for CEO)

Twitter · Feb 1, 2023, 1:01 AM

this is why

all the big brain people fucking ran

[rustic panther]

Yeah, but more that it doesn't pass the sniff test and I wonder if there's a fault in testing methodology

[gilded helm]

I know some people taking early retirements from Intel, they're actually thrilled about it. Mostly because the packages are really good and they're just going to take other jobs and be semi-retired.

[twin dew]

Intel got rid of all the "expensive" people 10+ years ago.

[stray badger]

actually, thats not that bad for intel

[twin dew]

Via early retirement programs etc.

[willow pike]

yeah i hear things aren't great in there

[twin dew]

As in all the experienced engineers etc.

[stray badger]

the graph looks very weird tho

why is 7700x above all others from its gen?

granted, its 4 fps

[verbal raft]

Also zen 5 percent my favorite (insert zen 5 percent screenshot)

[willow pike]

probably some hell combination of lack of SMT, inferior memory latency and windows fuckery

[stray badger]

it looks like the game just needs more cache and thats all it needs

[willow pike]

i bet it's intel default profile on new tests

[stray badger]

everybody is basically tied except for x3D

[twin dew]

Highest amount of cores on non-dual-CCD.

[stray badger]

makes sense

[twin dew]

Game tries to use too many threads and the cross-CCD kills some of the gains.

[rustic panther]

No, not on Intels part, but on Hardware Unboxeds part

I'm trying to find other benchmarks where the difference between the 14900k and 285k are that stark in cp2077

[verbal raft]

GN?

[rustic panther]

Do they still do text articles?

[gilded helm]

It's because there are software problems with the 285K. All reviews that I've seen have shown very uneven performance.

[verbal raft]

Or have they stopped testing cyberpunk?

[willow pike]

yeah some games tank on 11 24H2 balanced profile

[verbal raft]

Sometimes

[gilded helm]

Plenty of people still play Cyberpunk. It's Plague Tale tale that should be axed from benchmarks. It's a random indie game with few sales that's less useful than timespy as a reference point. But Plague Tale tale was randomly demonstrated as part of the Lovelace release so we're stuck with it.

[verbal raft]

They convert their videos to articles

[rustic panther]

Tom's Hardware sort of looks the same on cp2077

[gilded helm]

Don't get me wrong, I played them both and I think both Plague Tale games are great.

[rustic panther]

Hm.

[night girder]

The game does look good tbh.

[verbal raft]

Indie?

[stray badger]

plague tail requiem is a pretty decent UE5 benchmark

[rustic panther]

Still, I suppose that's what you get when you decide to move off of "fry the fucking sand" as a cpu strategy

[verbal raft]

That looks better than the "AAAA"games

[willow pike]

intel still can't match SoIC

[rustic panther]

The chips are much more efficient though compared to previous gens from what I'm reading, so good job on actually building something you can improve on Intel

[gilded helm]

Any generic UE5 benchmark would be better than Plague Tale. Even Satisfactory would be more useful despite it being so CPU reliant.

[rustic panther]

Isnt that good for a CPU benchmark?

[night girder]

Eh ... Satisfactory might be a bit wonky 😛

some stuff is implemented ... not in the best way?

like global ilumination ...

[willow pike]

what game isn't

[gilded helm]

Plague Tale doesn't even have a built-in benchmark, so you're also relying on subjective testing

[verbal raft]

Translation:good job intel for not making something that will melt in a year

[night girder]

well, they stated global illumination is available, but not really optimized.

[willow pike]

we hope

[rustic panther]

Hey, it's a bar Intel doesn't always clear!

[verbal raft]

LOL

[night girder]

Does anyone know, what criteria a game must have to be a good bench mark game? Because I was thinking, cyberpunk was trash at the beginning.

Who/what decides this?

[willow pike]

well we now have an extremely rare case of intel and amd using identical processes, and uh

not super great

[verbal raft]

I think run to run variance should be low

[night girder]

Deterministic?

[gilded helm]

Popularity should be a big factor in my opinion. Otherwise it's not a useful frame of reference. If a game has a free benchmark tool, that's fine though.

[rustic panther]

Eh, AFAIK both companies decided to stay in place performance wise to work on efficiency and setting themselves up for later, which is fine, at least I'm not missing anything with not upgrading, which makes me happy

[verbal raft]

Isn't that just for physics?

[night girder]

Wasn't Crysis once used for benchmarks?

[stray badger]

ye

[gilded helm]

Yep

[night girder]

Aren't physics important for benchmarking?

And I mean deterministic, if you run the same test over and over again, it should give the same result.

[willow pike]

amd's bigger strategy is more efficiency means more cores on EPYC, even if that means we don't get blazing fast gaming products. Intel's strategy is oh fuck everything is on fire 20A is shit we need TSMC

[verbal raft]

Well if it has a big effect on performance

[rustic panther]

Yeah, as it turns out chips can only take so much power before they break

[night girder]

To me it looks like high demanding games (gpu/cpu) are preferred for bench marking.

[willow pike]

gelsinger has been the entire fucking company on 18A

[rustic panther]

They shoulda hired Magic as an advisor, then they would've learned that without subjecting people to it! :^)

[night girder]

Nobody would use Banana for it, which is 3rd popular game still on steam 😒

[verbal raft]

Your definatly not wrong there

[stray badger]

i hope it pays off. we need intel as much as we dis them

[willow pike]

you have any fuckin idea how much better that would need to be than N3 to make up the difference?

[willow pike]

alas, we are cooked

i have watched intel fuck processes up for almost ten years now

[stray badger]

if we lose them, its just AMD and qualcomm

[willow pike]

i got no reason to think 18A is any different

"they say it's good!" they always say that

[stray badger]

and then we are back to skylake for another decade

[night girder]

Btw, Apple will probably announce new models with M4-soc next week.

[rustic panther]

We from Toilet Duck...

[willow pike]

apple's competition is no competition, they'll just hole up in their mac bunker and bilk prosumers of their money

[gilded helm]

The IP/assets will go up for sale if intel goes bankrupt, and we'll probably come out with Intel again.

[willow pike]

even if the M4 shits gold, which it do

also yeah pretty sure us government will intervene in intel

[night girder]

Yeah, you are not wrong. Apple is kinda in its own little bubble.

[stray badger]

hopefully still an American or at least western based company

[willow pike]

anyone reading this isn't gonna pay $1500 for an 8GB RAM laptop

[gilded helm]

The government would never let a foreigner buy Intel

[stray badger]

very true

[verbal raft]

Yeah get ready for the ryzen AI PLUS ++19985 AI XTXTX+++++ for the great low price of 2999 2percent faster than last ten BTW

[stray badger]

except.... november 5th

[willow pike]

yeah it's national security at this point

[gilded helm]

The house cleaning that comes with a bankruptcy can be good in the long term. But it's of course awful in the short term.

[willow pike]

speaking of apple, somehow the 4090 is more efficient than any apple gpu

it's 450W but that much faster

[verbal raft]

Yippe

[stray badger]

yeah, nvidia actually knows how to make a hella fast GPU

[willow pike]

so do apple, but not super big ones

[stray badger]

not really, they haven't been doing it for long

[gilded helm]

Does Nvidia break out desktop and mobile GeForce sales? I don't think Nvidia is ever going to abandon efficiency completely.

[willow pike]

still not sure trying to reinvent imagination was a great idea

[willow pike]

they been doing iPhone graphics for like 17 years

[night girder]

👀

[stray badger]

nah they been using powervr

[willow pike]

weird

[night girder]

Was looking up data between M2 and 4090 and came across an article. This is a AI benchmark.

[willow pike]

the max should not be beating any dgpu

okay uh

[stray badger]

probably poor optimization for nvidia

[willow pike]

no?

there's no way that's using CUDA

[stray badger]

yeah remove the https part

[night girder]

Ah, I forgot.

appleinsider.com/articles/23/12/13/apple-silicon-m3-pro-blows-away-nvidia-rtx-4090-gpu-in-ai-benchmark

[willow pike]

or the tensor cores

[stray badger]

not cuda anymore, OptiX

i think ai is its own thing

[willow pike]

optix is graphics

cuda is general api

ai uses cuda

[stray badger]

cuda cant access some of the featues of modern nvidia

i know it doesnt use rt cores

[willow pike]

that's what optix is for, but ai doesn't use rt

ai uses matrix multiplication, as in tensor cores, which is very much in cuda

[stray badger]

ah, so that is still cuda

seems like they could just make a unified API

[willow pike]

ppfffdfffgggggghhh

[night girder]

😂

[stray badger]

there we go

someone used cuda

[willow pike]

okay so apple can beat nvidia when you run something 25x slower

[night girder]

No, apple beats nvidia out of the box.

[willow pike]

[night girder]

But if you optimize, nvidia beats it.

but Nvidia cut its times by more than half when using appropriately optimized benchmark tools.

[willow pike]

fucking hilarious

[night girder]

but so far, I don't find much other data between the two.

You got sources?

[willow pike]

We're leaving the original story intact since it reflects how the non-optimized tool performed.

hack website

[night girder]

all I find are dodgy benchmarks.

[willow pike]

ye but I'll have to dig in chat history because as you've found, only apple users are really testing this, and they clearly don't know shit

The most interesting factor to note is the difference in power consumption. That result didn't change — Apple's chips performed well at a fraction of Nvidia's power draw.

they were TEN TIMES SLOWER

"performed well"

[night girder]

Well, this tells the same story it seems.

[willow pike]

that M2 Ultra is a 160W SoC

[night girder]

blown away, but it doesn't feel like a reliable source.

[willow pike]

that's just theoretical flops

here's an older one

4090 would be much faster still

they at least come closer than in ai, which nobody is beating nvidia at

(apart from maybe MI300)

[night girder]

Well, all apple has going is that it's consumes less power ...

[willow pike]

lmfao

a laptop benchmark

i cannot remember what the zephyrus used

this is also why apple outright killing dedicated GPU support means they do not care about anyone outside they bubble

[night girder]

I wonder how m4 will perform

[willow pike]

CPU wise, real good

[stray badger]

probably a 4060 or 4070

[willow pike]

curiously they never made an M3 Ultra

guess demand isn't there

and InFO_LSI ain't cheap

[night girder]

stuff is expensive, but so is a 4090 ...

apple prices are pretty brutal. Same as Nvidia's now.

pretty happy with AMD now 🙂

[willow pike]

you could get literally two 4090s

[night girder]

a 4090 isn't a PC.

[willow pike]

yeah i know but it's fun hyperbole

[night girder]

you can't deny that nvidia's expensive

[willow pike]

i got a 7900xtx because i was unwilling to pay £300 more for a 4080 so yes

apple take the piss

[languid gulch]

just spent 2 days cleaning my car inside & out for the first time since i bought it

[languid gulch]

https://www.youtube.com/watch?v=XXLY8kEdR1c

dumb question for the power draw section at 8:30. does that mean that intel's lying about its power draw?

[jagged snow]

Those chips are still much more effecient than last gen in some cases

So not lying, just the normal amount of skirting the truth

So that 12v power from the 24 pin does decrease the effeciency gains but not negate them completely

And it is possible(but unlikely) that that power really is just going to the board and not the cpu

There's no real way to validate that

[languid gulch]

good point

but man those power spikes are kinda suspicious

btw i just double checked my bios, turns out me risking it all to make the jump from 1st gen to 3rd gen also included the 5950X/5900XT, so at least i don't have to redo that

[jagged snow]

Ah nice

[languid gulch]

had that thought out of nowhere after i realized how much newer the 5900XT is

that mini panic of WAIT

have there been any hints of black friday sales?

[charred willow]

C/2023-A3 from the Takahashi Guidescope on the Calvin University Telescope

[stray badger]

we still have a month to go

[languid gulch]

oh i know but some companies like to advertise this far in advance

[sand ferry]

I do see that kind of thing 4-6 weeks in advance, depending on the store and/or product.

[languid gulch]

https://www.youtube.com/watch?v=WxXZlONu4Ig

ouchie

[edgy hazel]

My turn to post a boomer meme

I am currently fighting my browser with 3 windows each at least 50 tabs

[languid gulch]

yea, i'm at 40 tabs rn

i hate doing multiple windows so i do my best to keep it pared down

[edgy hazel]

Same but I keep accidentally doing that

[languid gulch]

i'd probably have double the tabs if firefox had edge-like tab groups

but my adblockers ended up failing on edge & chrome, & i'd rather have that be functional than tab groups

[gilded helm]

Firefox has tab group functionality built in but disabled, extensions can re-enable it.

[languid gulch]

yea but in a workaroundey way compared to edge, i like that kind better

[edgy hazel]

https://tenor.com/view/shocked-oh-for-real-gif-10349185

[languid gulch]

i tried looking for one that's identical to the edge method & couldn't find it

it's like 2-3 extra clicks for the firefox extension to get to a tab you're after

[willow pike]

Intel says we can expect a 15 to 20 ns increase in memory latency for Arrow Lake.

well then

[mental oriole]

Oof.

[soft bloom]

my tabs are my watch list for entertainment browser video.
the work window is easier to maintain... oh wait, these are not individual 10 tabs - these are 10 groups of tabs... shit

[pure karma]

always have been alho i did hear there changing it up and releasing a actualy decent phone sometime soon aparently

[willow pike]

there's also this

[pure karma]

bruh i have 2 and one is a new tab

[verbal raft]

if someone asks me if they should by intels new CPUs i show them this

[willow pike]

waste of sand

[verbal raft]

even better

intels best attempt to win efficency

[night girder]

AMD 5800X: 100.3W & 1216.9 miliseconds. Intel 285: 161.6W & 1194.9 miliseconds. Crazy.

AMD wins the zipping battle.

[verbal raft]

at least its better than last gen

[verbal raft]

millions of instuctions/joule

[night girder]

It's still just decompression with 7-zip 😛

[dire igloo]

Where's this from?

Looks like 3dcenter but I only know the German site

[willow pike]

it is 3dcenter

[soft bloom]

i turn over the calendar
and it's 3rd september yet again

(3dcenter sounds like 3rd december)

[dire igloo]

Can you link it pls?
Or did you use translate?

[willow pike]

am looking for it

https://twitter.com/3DCenter_org/status/1848980326489723207

3DCenter.org (@3DCenter_org) on X

According to updated benchmarks on gaming performance by @ComputerBase, Ryzen 9000 has now gained a whopping 7-10% compared to the launch results, roughly reaching the performance level of Raptor Lake.
https://t.co/LbOkLi7omA
1/3

Twitter · Oct 23, 2024, 6:50 AM

we like sourcing

[dire igloo]

Hm. gud nuf, fanks

[willow pike]

there's a link to the article in the tweet too

[dire igloo]

Yup, will check it out over the weekend

[dire igloo]

I live by Hitchens's Razor: what can be asserted without evidence can be dismissed without evidence

[willow pike]

the evidence says intel cook'd

[twin dew]

Lot will depend on Intel 18a

[night girder]

7 9800X3D is 8% faster than predecessor (15% for multithreaded)

probably already posted:

[willow pike]

i am not convinced 18a can save them

even if they literally spin transistors out of manna from heaven, they still have increased memory latency to contend with

and they made the absolutely insane decision to not increase L3 cache size when moving to a chiplet design

plus i haven't heard much about intel's SoIC (3D vcache tech) competitor for a while

and then there's the very real possibility that 18a is merely good

or broken

they have so much work to do

while amd are still using the absolute most basic kind of chiplet tech (apart from SoIC) and running absolute fucking rings around them

y'know in 2016/2017, i remember seeing a really early marketing slide from intel that's pretty much meteor lake / arrow lake, splitting their CPUs up into chiplets and stacking them on one base die, and thought "this will fucking suck when they cut the memory controller off"

AMD proved me wrong

and now, 7 years later, intel prove me right

oh and while i'm on one intel aren't even doing chiplets right cause they don't reuse anything

the CCD you get on any ryzen is what they use twelve of to make an epyc, which gives them huge flexibility in binning

that compute die in arrow lake is for arrow lake, xeons do not use it

you don't get the same flexibility

[safe trench]

im on intel also

[willow pike]

or are they using 16 of them these days iun fuckin know

they get to shit out one chip design and get dozens of products out of it

[rustic panther]

I just hope Intel can right this ship because I don't want to end up in a situation where there's only one viable choice again

[willow pike]

alas, we are cooked

we are doomed to have 8 core X3D parts for the next 5 years

gonna rename myself to doomerman

[rustic panther]

I mean, really depends on if Intel can build on this gen

[willow pike]

i got no hope

[safe trench]

285k was barley any diference

[willow pike]

and while amd dominate cpu, they dun give a shit about radeon no more

so nvidia will keep jacking up geforce cost

[rustic panther]

They're not stupid over there, so I assume that there's a roadmap here

[willow pike]

the smart people left when they bonuses got cut

[safe trench]

liny only cares about the epyc

afterall what is there to glaze other than a useless bragging right?

[willow pike]

liny?

[safe trench]

linus

[willow pike]

lmao it'll make for great ltt thumbnails

and they'll bilk big companies of all money

(you think this 285k is bad, it is so much worse in servers intel has been cooked for years)

[safe trench]

new thumbnail idea: LINUS KISSING THE NEWEST EPYC!

[safe trench]

ksp forums shutdown i think also

[edgy hazel]

I'm on Intel too.

On my macbook at least

my PC is another story

[maiden coyote]

Sucks to suck ig lol

The layoffs at intel are concerning, it seems they're shedding a lot of talent, otherwise I'd say for the next few gens amd is what you want, in 5 years it might shift back to intel

[night girder]

"I am on Intel" says herg (Radeon RX 7900XT) 😂

[edgy hazel]

my macbook is intel

[night girder]

Just funny.

[edgy hazel]

my PC is

[night girder]

and Apple is no longer a partner with Intel last I know

the older stuff sure have Intel. But those will get replaced.

[safe trench]

no way, really11!1!1!!1!1

[dire igloo]

Well, you gotta consider:
Fire the cheap workers that cost very little and don't hold you back or fire the expensive engineers with their stupidly high salaries that keep bringing up issues?

[maiden coyote]

Here's to hoping intel comes around, we need them to stay relevant, and continue with arc

[night girder]

Do we?

[safe trench]

we cant just all rely on tsmc

[edgy hazel]

you should know about this Mr. Fire Worker

[maiden coyote]

Would you like to pay 1500 for a amd cpu?

[safe trench]

they are literaly in taiwan

china can just do the funny move

[maiden coyote]

Or $2000 for the ryzen 9900x lol

[night girder]

Wait... you said* ARC which is a GPU. So what does that have to do with CPU? Intel can stay relevant for CPU.

[dire igloo]

Ayo what, no, that's not my name
(It's Firework and errrrrr ....)

[maiden coyote]

If intel sinks amd will do like nivida.

[safe trench]

we need arcs

[edgy hazel]

that'd 100% cause ww3

[night girder]

If Intel fails on GPU, we still got Nvidia and AMD. Keeping eachother in "balance".

[safe trench]

"mr president, theres a second temu b2 bomber heading towards tsmc!"

[safe trench]

amd is loosing to nvidia

[maiden coyote]

Having a 3rd player in graphics is nice.. intel offered av1 encoding on their cheapest card, no one else does that

[safe trench]

and intel challenges the 3050

which isnt much but its damm impressive for a first gen gpu

[night girder]

I am also rooting for intel, but I am not expecting much for the next few years.

[edgy hazel]

tbh nvidia is only good for the high high end. Everything else is outclassed by amd and intel.

[safe trench]

and creators

[maiden coyote]

I think they can carve out a nice niche in hardware video encoding.. they kind of already have

[safe trench]

you killed chat

[maiden coyote]

Ohwell

[dire igloo]

Quicksync is also nice as a decoder for media PCs

Three things Intel needs to work on:

1. Getting rid of the "bad drivers" reputation
2. Compete with upper midrange
3. Become viable for productivity

So far, number 3 is where they're doing the best

[night girder]

the "bad driver" reputation needs two things; good drivers and time.

[edgy hazel]

and good drivers require more than 2 junior devs working on them

[dire igloo]

Tho tbf, arc drivers did improve a lot since launch, especially for modern titles

Still struggling on old titles tho - but who cares about 250 vs 500 fps in siege

[safe trench]

why do i need 1000 fps in portal 1?

[night girder]

Nvidia users 😉

[fiery bane]

Does the Nvidia GeForce MX550 2GB graphics card work satisfactory?

[soft bloom]

huh?

[low scaffold]

Hello there fellow Pioneer!
You're currently chatting in off-topic channels.

Please note, the entire off-topic category is reserved for not Satisfactory related chattery. To comply with your FICSIT Inc. contract, please make use of the appropriate channels in the Satisfactory category. Thank you!

Command executed by: hergarian

[maiden coyote]

Probably not, but It makes a nice paperweight

[soft bloom]

btw...
expanding conclusion from interview with <insert name i forgot> about ARM competition
GPU field seems to have the same peroperty - it's not like current monopoly at <insert time period> is holding some secrets that let them dominate, because as shown from recent developments, new companies given enough investments can reach considerable heights in relatively quick time (AMD catching up to Nvidia, Intel slowly releasing their first models) (also in CPU space 6 years ago (OMG am i old?) no one believed in AMD)

[willow pike]

you are not old

not even i believed in amd, not after bulldozer

[soft bloom]

it's sad that everyone is like "AI is the future" and go use cloud services that bought Nvidia cards...

[twin dew]

In DirectX11 mode, yes.
DX11 downsides are that the story/progress videos will not play, just audio.
Far terrain will be spiky.
Other minor graphical issues.
But the video parts are just animated visual gags related to the audio.

[maiden coyote]

Exactly, and good drivers, and being competitive will come with time and presistance.

[willow pike]

zen was an astonishing ass pull

[soft bloom]

I believed in AMD when i was building my PC. 2200G brought me through

[willow pike]

lisa su is a goddang genius

[fiery bane]

How many frames do you expect it to run?

[willow pike]

jensen huang is a merciless genius

pat gelsinger

[maiden coyote]

Amd was good in the 2000's they released released x86-64... then bombed in the 2010's with bulldozer until zen

[twin dew]

Depends on settings and save size.
With CPU and RAM being more important in long run.

[maiden coyote]

I guess I'm old too

I remember reading about althon 64 in a magazine lol

[willow pike]

how's your back feeling my dude

[twin dew]

Playably for the GPU in low at least, in that DirectX 11 mode, that needs launch argument to enable.
As DirectX12 needs 4GB of VRAM minimum to work properly.

[fiery bane]

Thanks

You are the best

[maiden coyote]

hothardware.com/reviews/amd-athlon-64-x2-3800 lol...

[languid gulch]

i remember making the jump from 32 to 64 bit

and it involved going from 256mb to a whopping 2gb of ram for $200

on 4x512MB sticks

i can't remember what the gpu was in that one, but i'm certain it was a PCI slot, & when it died, i opened up the case, & 3 out of the 7 fan blades were gone. not just snapped off, but gone. nowhere in the case. still have no idea what happened to them

[verbal raft]

you can run satisfactory on anything heck i used to play it on the VEGA3 it ran like a dre.... A NIGHTMARE

also has anyone heard about this small very good UNBIASED reviewer?

the person that writes for this site is so nice............

so unbiased..................

they even include something that is fully unneserary..............

[verbal raft]

even less so.............

[edgy hazel]

most of us simply ignore that site

[glossy glacier]

https://youtu.be/RQSBj2LKkWg

sums it up pretty good

[verbal raft]

my problem is that if the avarage gamer that wants to build their own PC (and do some research) and they want to see.which is faster this or that? userbechshwarkshcmart comes up first so they click on it and then they say oh,clearly the 14900k squishes the 7800x3d with no problem so they waste a whole bunch of money for no reason

[verbal raft]

absolutly does

[languid gulch]

i've always heard that userbenchmark is biased against amd (& they very clearly are), but where did it come from?

[dire igloo]

It's basically a single curating the site and he just HATES AMD to the gut

[willow pike]

it's SEO'd to the top of google

even the intel subreddit banned it

[soft bloom]

i did but they are biased

[gilded helm]

I don't know that anyone has ever uncovered the "why?" behind the website, but it's basically just someone's pet project.

[night girder]

It's on their website?

Why we do it
It's tough to choose hardware. An army of marketers steamroll social media with anonymous accounts: reddit, forums, youtube etc. Incompetent "moar core" marketers would sell ice to Elsa.

Who we are
We are an** independent team of scientists and engineers**. We do not have time for HR, PR or marketing.

[gilded helm]

If you believe their own words 🤷

[night girder]

Our users
Our users are independent thinkers, they demand testable facts and they have no respect for the status quo or the latest fad.

Nothing to do with believing, that's their statement on the why.

But I mean, it sounds so unprofessional. "Incompetent "moar" core" marketers would sell ice to Elsa?"

[soft bloom]

i never read those blobs of text. without them it's alright to use to just compare AMD cpus between generations, hehe

[night girder]

Since UserBenchmark declines sponsorship, it has become the target of a smear campaign which intensified following improvements to the CPU effective speed index in July 2019. Billion-dollar brands can try to shut us down but they can’t change who we are, the clue is in our name. UserBenchmark serves users exclusively without corporate sponsorship or "free" samples.

The whole about-page feels like a reddit post 😕

I stopped using userbenchmark long ago.

[wanton orchid]

I really thought it was trolling at first, then I started to question some years ago

tbh it's so obvious userbenchmark may serve as amd community toilet flush

[sharp oasis]

Gcam on a moto edge 2022 slaps tbh

[dire igloo]

Was about to compliment you on the Frozen Edge pick, but why change away from B12 glory

[languid gulch]

yes but do you have enough side panel clearance 😛

[sharp oasis]

They are there

Just on a pull config

It's push pull

[dire igloo]

In an H6, madman

[sharp oasis]

It barely fit

Gcam does nicely with my Motorola's 50mp sensor

[sharp oasis]

I almost had a clearance issue with my mobo VRM heatsink

Its like sitting right on top of it

[safe trench]

funny

anyways is userbenchmark trustable?

[dire igloo]

As trustable as the Nigerian prince in your email inbox

[night girder]

What did you think we were talking about?

[safe trench]

erm mogerators!

[safe trench]

is it spyware?

but they are infact seoing the max

[maiden coyote]

They claim to be unbiased, yet call amd "advanced marketing devices"

I've been using passmark lately.

[wanton orchid]

I tend to use passmark too, also I reported some cpu data issues to passmark, and they personally responded and fixed the issue (there was a miss in how their system retrieved the data from their latest version hardware databases)

since then, the reading from databases as been clearly updated on the website without the issue I reported earlier

[vale finch]

But can it run satisfactory? Hehe

[pure karma]

windows... get some help

[narrow estuary]

Appears to be a Buddhist calender, try to minus 543?

[twin dew]

You got hit by long time Windows "feature" where it syncs time based on outside HTTP(S) timestamps and not NTP-server.
And all Linux servers have used completely random timestamps for long time already for security reasons.
So normally those are discarded as invalid by that stupid method, but if enough of those random stamps are close enough, your computer clock gets set to completely random time.

[pure karma]

the clock is fine i just found it funny because its just another one of thoes circumstances where everything breaks when i touch it lol

[twin dew]

In theory for times when NTP doesn't work, or when you need to get initial fix to get NTP working.
But has just been liability since those server timestamps were randomized long time ago to not allow use in timing attacks data source.

[twin dew]

Yeah, but point was that there is legit reason why it changed to insane value.
From old Windows part that should have been disabled by default long-long-long time ago.

And sorry, not HTTP(S), but TCP headers.

[pure karma]

so if im understanding right it just picks a random number for security?

[twin dew]

The servers fake that TCP timestamp option output with randomized time.
And Windows tries to use those TCP timestamp replies it gets to set the time.
When you get large enough number of replies with close enough random number by accident, Windows sets the time to average of that randomized number.

That option is supposed to give the servers unix timestamp.
But that is security hazard today, as there are many attacks that rely on knowing the servers accurate time.

https://arstechnica.com/security/2023/08/windows-feature-that-resets-system-clocks-based-on-random-data-is-wreaking-havoc/

Ah, I was misremembering

So it was HTTPS and like only, and pretty new.

And completely bonkers.

[languid gulch]

dumb question time

if i got something like the 5900XT/5950X, how worried should i be about blowing a cap with amps

[twin dew]

The limiting factor are the power stage/mosfet temperatures, and those have monitoring and throttling if needed.

So you just get less performance if the MB power delivery cannot keep up, as the CPU throttles down.

[languid gulch]

i've read mixed things about my mobo & its ability to handle the amps

[twin dew]

But it only leads to that throttle down on power delivery overheat lowering all-core performance.

And using top-down cooler instead of tower cooler, or extra fan giving the power delivery some airflow can help much on that.

[twin dew]

And you can just set ECO mode too for the CPU.

105W ECO for example.

[languid gulch]

i'm wondering if it's worth the extra $100 over the 5900X

[twin dew]

5900X isn't really for anyone.
Either you want single-CCD if you don't need the multicore, or go to the two full-CCDs if you do need it.

[languid gulch]

yea true

plus it'd give my cooler something to do

and with AM4 reaching EOL i might as well go for top of the line to keep it fairly relevant in the future

[twin dew]

But again, if you want 5700X3D or 5900XT depends on what you actually use the computer for.
Do you actually use it for work/number crunching that benefits from the extra cores or not.

Where the extra speed actually matters.
Because the 5700X3D can do those too, just slower.
So if you don't constantly use the multicore, then 5700X3D would probably be better.

[languid gulch]

it's mostly a gaming rig, with archive work on the side, but i've been thinking about doing more with it

5600X is starting to feel a bit choked off occasionally

& idk if i wanna burn $200 on 2 more cores like the 5700X3D, even if it has the fat cache

i like to do lots of things at the same time, which definitely seems to like cores

[twin dew]

Depends on if you want more game performance (5900XT doesn't give much)
Or more multi-core performance (5700X3D doesn't give that much)

[languid gulch]

i mean, 5600X gets me basically all the gaming performance the 5000 non-X3D can provide

it feels like more cores would be better

[twin dew]

Yeah, little higher single-core on the higher parts.

[languid gulch]

5900XT/5950X is 64mb L3 cache, which is double the 5600X L3, & with the optimizations they've done, i'd hope that'd be a noticeable difference

it's not the 96mb, but it's kinda close

[twin dew]

Not really.
Still same 32MB per CCD as the 5600X.
Cross CCD cache usage is slower than hitting RAM.

[languid gulch]

ah true

i keep forgetting about "the glue"

[twin dew]

Point of the X3D is that it has 3x the L3 for the single CCD.

[languid gulch]

ah crap right

so that explains that

why didn't they make dual socket AM4 so we could do 2x 5800X3D

[twin dew]

So if the stuff doesn't fit into the 32MB L3, but fits into 96MB L3, then the speed is much improved.

Or even if it doesn't fully fit, but keeps more of the accesses in L3, and not pure cache trashing.

[languid gulch]

i'm currently at 32gb 3200MHz ram, should i increase capacity or speed?

if i upgrade to one of the 5900s

[twin dew]

Depends?
More RAM only matters if you run out.

[languid gulch]

most i've been seeing when i'm going all out is around 26ish

[twin dew]

Then 3600MT/s would give more performance, from the raised Infinity Fabric speed.

Not so much from the RAM frequency increase itself, but from that upping of the IF while still being linked.

[languid gulch]

ah crap, nvm, mobo doesn't support anything past 3200

[twin dew]

If you have that crappy MB, and was planning to possibly change RAM anyways, why not just update all three then?

And go to AM5 or similar.

If it doesn't have (OC) to 3600MT/s.

Anything above 3200MT/s is always OC on AM4.

[languid gulch]

oh i haven't checked OC

[twin dew]

As that is the max official RAM speed for the CPUs.

3600MT/s is just what AMD says that 99+% of the CPUs should be able to do.

And officially-unofficially recommends for 5000-series multidies

[languid gulch]

yea lemme give it a shot, gotta hop into the bios for it

[twin dew]

BIOS will have options.
Question is if those are expected to work or not.

[languid gulch]

right

[twin dew]

What MB?

[languid gulch]

B450 AORUS PRO WIFI

old, i know

[twin dew]

Support for DDR4 3600(O.C.) / 3466(O.C.) / 3200(O.C.)/2933/2667/2400/2133 MHz memory modules

From the Gigabytes Spec page for rev 1.x.

So 3600MT/s is expected to work from the MB side.

[languid gulch]

kk, let's see if it likes me

oh, my ram is 3600MHz rated

k, this better work then 🤣

[twin dew]

Try with the XMP on then.

Don't try to manually just set the frequency.
Needs the RAM voltage raising that the XMP does.

[languid gulch]

got it

Xmp is enabled, but stuck at 3200. Can't adjust it at all

[twin dew]

Start CPU-Z, show the SPD page.

[languid gulch]

Tried to fire it up too fast apparently, locked it up

[twin dew]

So just 3200MT/s XMP.
Not 3600.

[languid gulch]

hmmm

[twin dew]

And the part number also says it is 3200MT/s RAM.

[languid gulch]

i have a box that's 3600

probably in one of my parent's systems then

[twin dew]

Not for that RAM.

[languid gulch]

well damn

now how do i sneakily swap out ram without them noticing 🤣

k, well, guess that's off the table then

so i guess now it's best bang for buck changes for ~$400

ish

if i go am5 then that's new mobo & ram for sure

but the 5900XT seems to usually be cheaper than the 5950X, so i could put that into faster ram, maybe more capacity for shits & giggles

well, in any case, got some more thinking to do, thanks for the help 😄

[twin dew]

Remember that that 3600MT/s is only for two sticks.
If you go to four, official max is 2666 or 2933MT/s depending if they are 1R or 2R.

Down from that 3200MT/s official for 2 sticks of either.

[languid gulch]

ah right

so if i went 4x16gb i'd actually lose ram performance as opposed to 2x32gb

[twin dew]

Yes.

[languid gulch]

kk, will keep an eye out for that

[glossy glacier]

You can always try to it run the speeds with for dimms. Official support is lower but if the memory controller is good enough it'll work

[twin dew]

Unfortunately Corsair is very good at the binning and using the crappiest chips they can for their RAM at each bin.

[glossy glacier]

I have a 3700X and 5800X each running four dimms at 3200

[twin dew]

Ah, you meant that way.
I understood trying to go above the current RAMs XMP.

Yes, it can work, at least if you raise the VSOC that isn't done by default at that point, which would happen automatically for above 3200MT/s RAM speeds.

[visual tree]

When will EU finally ban daylight saving time, there is no point in using it today....

[twin dew]

When they can agree on how to banish it.
If everyone needs to always use old winter time, old summer time, or can choose on per country basis whichever once.

It was agreed to stop switching, but no agreement on how to implement, and then COVID hit.

[visual tree]

In short: it will take 10 years until we reach an agreement 🫤

I don't really care if it's winter or summer time, just get rid of the stupid daylight saving time

[narrow folio]

They could just meet in the middle, a one time 0,5h adjustment and call it done for

[visual tree]

There are suggestions like this

[twin dew]

That is the current status.

With Winter time.

But I would for example personally rather have Finland at +3 all year around.

Basically that pic is just "Winter time all the time, everywhere"

Summer time everywhere would be +1 more to those.
And then if each country could pick, you might get more patchwork, but what would that really matter?

So I would say to make it "Every country decides once" at EU level.

[visual tree]

Maybe they could make something like in the picture by making Portugal use GMT+1, region from Spain to eastern Europe (+ Sweden) GMT+2 and Finland, baltic states, Romania, Bulgaria and Greece use GMT+3

Just to avoid a patchwork like you mentioned

Or 4 regions

Found another proposal

[wanton orchid]

that one have more respect for natural timezones

per country random choice would be very bad for inter country work
especially for Shenzhen

[twin dew]

How so?
But that is coming from a country where you have always had to be aware of timezones when crossing borders.

[soft bloom]

seasonal timezones talks open?

https://youtu.be/z2EUPDSabY0?si=CvVRyXuTurqhyPDa

[willow pike]

lmao ryzen x3d was a fucking afterthought

"we got all these server x3d chips, what if we put them in am4"

https://youtu.be/RTA3Ls-WAcw?t=789

amazing

[twin dew]

Yes, that came out already in previous videos.
Extra CCDs from first test batch for the server CPUs that then were used to make couple internal test chips for the OC team, who found out they were great for games.

As in too few CCDs to make another test server CPU.

[willow pike]

this is a previous video, i just missed it

[twin dew]

Ah, sorry.

[willow pike]

so lemme see if I get this right

AMD's strategy;

• use cheap chiplet technology to absolutely shit out top performing EPYCs
• increase cache to compensate for chiplet latency
• use one small CPU die for as many products as possible to get amazing yields and find a use for everything they make

Intel's strategy;

• use extremely expensive chiplet tech (foveros)
• do not increase cache
• use different dies and different packaging for every single market

and then, when the intel xeon sucks and nobody wants it, offer steep rebates and discounts to force through big sales numbers of their much more expensive products

i don't fucking get it

[twin dew]

Those extra cache chips were special product for some specific user AFAIK originally.

[willow pike]

that's even funnier

[twin dew]

Who had load they knew would benefit from it.

And was willing to pay part of the development costs.

[willow pike]

though really I just mean more cache without x3d

zen 2 doubled L3 versus zen 1 to compensate for moving the memory controller to an IO die

then zen 3 uses a ringbus to help even more

also like, we have meteor lake, arrow lake, lunar lake, sapphire rapids, emerald rapids, granite rapids, sierra forest, all completely different CPU die designs, completely different packages

it's gotta be so expensive

[twin dew]

Also even major internal architecture differences too.

So not even same core design in all.

[willow pike]

AMD just makes three kinds of CCD and combines them as needed

(regular, dense, x3d)

and this is how intel ends up with the same profit margin as costco

[twin dew]

Well, now they have Zen5 CCD, Zen5c CCD, X3D extra part, but that just is add-on to the normal Zen5 CCD.
And then two different IO dies (consumer, server)

[willow pike]

yeah I'm being generous to intel by saying x3d is a different chip

[twin dew]

The X3D CCD isn't different for the CCD itself, they all have the connection points for the extra cache.
The X3D CCDs just get thinned and the cache added on top in extra steps.

[willow pike]

oh by the way rumor has it zen5 x3d is flipped compared to zen4

CCD on top of cache

[jagged snow]

Oh that's very interesting

How are they doing power delivery then?

[willow pike]

helps with heat dissipation,l

[willow pike]

probably with a bunch of TSVs in the cache die

zen 5's L3 is much smaller than zen 4 so there's already not enough room to put cache on top of it without it overlapping the cores

though again, just a rumour

we'll see soon enough

[vital prairie]

Hello can i get some help i tried again to see if i can get my mic to work with discord which was a mistake because now its not working in general

[verbal raft]

what did you do?

to make it not work in general

[vital prairie]

all i did was unplug it and its not working

which was a mistake

[verbal raft]

uhh weird. did you plug it in to the right one?

[vital prairie]

back where it was yes

[verbal raft]

then i have no idea why it doesnt work in general

[verbal raft]

but what i do when my mic doent work in discord is that i reset the audio settings

[vital prairie]

my mic never worked with discord

[verbal raft]

neither did mine until i reset it

in discord audio settings

[vital prairie]

well i gotta get it to work in the first place

yea i got it working i can hear myself

had to do this bull

[verbal raft]

good to know,(ill save it for later)

just in case

[vital prairie]

now for discord idk

yea idk what im doing there

[verbal raft]

try this

it worked for me

[vital prairie]

did that twice it did not work

[verbal raft]

then the best i know is to follow this https://support.discord.com/hc/en-us/articles/360045138471-Discord-Voice-and-Video-Troubleshooting-Guide i know you've probably been using it but this is the best advice i can give you

[vital prairie]

welp as far as i know im never gonna get it to work

[gilded helm]

Open Windows Volume Mixer and see whether something weird happened there?

[soft bloom]

Why don't we use holograms to record and display? Is it just sensors problem?
I mean, that would be such an interesting media.

[dire igloo]

I recommend you watch the 3Blue1Brown video - it's a super delicate process that requires monochromatic light

[soft bloom]

that's exactly the video that made me ask question

but the limitation of having monochromatic light doesn't seem™️ fundamental

[dire igloo]

except it absolutely is fundamental

the whole process depends on interference

it also depends on the reference wave being projected in the exact same matter

also, it's questionable how mass-producable these holograms are - and whether you could do moving picture

[dire igloo]

the whole reason why holograms are scientifically important in the first place is intereferometry - Grant showed that when they nudged the one object only a tiny bit

[soft bloom]

i didn't understand the reference about nudging

[dire igloo]

from when they had the hologram made and the objects still behind the glass, they moved one object in the scene only a little bit

and it created an interference pattern with the image on the glass

[soft bloom]

i will say that most mind blowing thing is that every shard of initial film can be used as a window into whole scene. kinda reminds of those broken phantasy mirrors. I think Harry Potter had one?..

[soft bloom]

i need to rewatch this

[soft bloom]

that's when talking about physical object with recorded information. but digital photo/video industry solved it over film, so I think it is solvable here as well. there are already developments of way to store much more data than we used to on flash or magnetic disks

[dire igloo]

we don't have the sensor tech necessary to store interference information digitally

and then you still have the monochromatic effect and the incredibly delicate process

[soft bloom]

i disagree, and here's why:
fundamental theme about hologram is recreating state of wavefront (amplitude and phase)
the fact that figured way to do that with coherent light (single wavelength) is more of detail of known implementation.

[dire igloo]

just having two wavelengths introduces a whole variety of possible scenario - you have four waves hitting the screen, making four different types of interference, assuming object waves and reference waves don't interfere with each other (which they absolutely do)

[soft bloom]

F, i wish i could visit Crimea and go to museum of holograms in Chersonesus

Hmmm, i remember it was near old city, near large temple
and the second museum is in Kyiv, in lavra...
i just have one question - why are they near temples?

[dire igloo]

idk, my local university has a lab working on nonlinear photonics (including holograms)

[soft bloom]

this is probably naive thought: if you have combination of N1. N2... Ni wavelengths, you record scene with same film, and use the same combination to recreate it. isn't it kinda the same?

[soft bloom]

well, lab is different matter.

[dire igloo]

it's a delicate enough process already. now imagine what happened if a single speck of dust would refract a single wave of light to be just SLIGHTLY off

[edgy hazel]

Wait you can lowercase sql commands

[soft bloom]

i just getting reminded of interview of someone who studied st. Sofia cathedral in Kyiv (yep, another one is in Istanbul), and he was stating that before reconstruction in barocco style there was some visual phenomena with mosaic. and now observing that hologram museums are near temples it makes me question if he was up to somthing 🙂

[soft bloom]

beam splitting screen in front of film that both let's illumination light hit the scene and get reflected onto film...
ok i just realised that it's real hard to rely on recreating light sources when shooting with any possibility of nature interfering

[dire igloo]

there ya go, that's the whole thing

holograms are delicate enough as is, you'd be more than quadrupling the amount of precision needed

[soft bloom]

this discussion made me realise that interference is kinda...
we think of it as pattern, right? thin zones of destructive intereferences, peaks and valleys.
but that is already observation over some period of time, that is larger than wave oscilation period.
so when we have ,ultiple wavelengths, superposition principle remains, and interference as well stil lexists, just that it's period get's ... harder to observe patterns.

[dire igloo]

it's merely a pattern if we simplify it a lot - assuming monochromatic point source of light (or in this case, parallel waves)
we have a whole field of phyics and mathematics dedicated to splitting waveforms into its components - fourier transformations.
Now combine the mess that is monodirectional fourier transformation with the mess that is monochromatic interference into multi-directional multi-composite light interference

[soft bloom]

ok actually i realised that in some cases there will be no pattern, like here:
https://youtube.com/shorts/nNFDvW0h1_8?si=UoL7qccD38cvb7fZ

[soft bloom]

all digital photography does is capture amplitude in in localised rectangular area behind predifined color filters. half job done.

the other half is merely controlling sun to recreate scene's lighting...

[dire igloo]

this logic sounds like "you have a 50% chance of winning the lottery - you win or you don't"

[dire igloo]

also, I don't see how this is related to interference

[soft bloom]

you might have 2 wavelength that simply don't, like ever, even over long periods of time, establish a pattern of constructive/destructive areas

so question is - do we even need reference light for make awesome recreations of past?
there are those digitally created ones as well

and was it Apple that anounced some weird product with near-hologram capabilities for video calls?

(not apple glasses/vision/whatever)

https://youtu.be/EmKQsSDlaa4?si=sB73tDvAggaIRg4_&t=2159
timestamp to other examples of holograms

having studying linear optics since school, and having pretty good grasp over that simple subject i never really thought about how weird it is that you can simultaneously see objects that are "normal" and "inverted".
this video does good job of showcasing that, and in motion! https://www.youtube.com/shorts/xi5oUVhZG6I

[willow pike]

https://youtu.be/oEeQ2mbkTlM

[verbal raft]

👀

[willow pike]

parity

[stray badger]

it still draws 250w...

please let it not be on 18a

oh what, its on n3b

how did they fumble it so hard? a 14900k on n3b would even be an improvement

[willow pike]

• no hyperthreading
• clock regression
• increased memory latency
• windows scheduler weirdness
• no extra L3 to compensate for chiplet

cyberpunk and other super modern games are built with hyperthreading in mind, it was unthinkable that high end CPUs would ditch it

so instead of having 16 threads on 8 big cores with extra threads on little cores, you have 8 threads on big cores and more threads getting shifted to little cores

it took this fucking long for games to properly use it, this trend isn't reversing any time soon

[soft bloom]

it took new console generation for games to expand beyond 4 cores... but i don't think there will be different console anytime soon

golden age of gaming hardware is gone

[willow pike]

yeah so long as consoles have hyperthreading, games will be optimised for it

(not all games, there's still shit programming)

[soft bloom]

i have a feeling that games don't think of consoles as much anymore

are there any big games anounced that would need something like powerful cpu?

[wanton orchid]

if you take 2015 level of cpu performance any game need powerful cpu

it simply now the bottleneck is becoming obviously programming skills and tools and not hardware anymore

Hardware will no longer compensate for programming shit

[willow pike]

single threaded performance isn't going to rise like in the old days

[wanton orchid]

and the programming shit starts at os level
where now Linux is more performant in games
even though Linux is tuned for computing

[willow pike]

we get 15% after two years and we think that's great

[wanton orchid]

it is

400% upgrade spoiled boomers

[willow pike]

i 'member core 2 duo knocking my socks off

[gilded helm]

Unless stacking becomes way cheaper, consistent big gains are probably a thing of the past. Otherwise, performance per millimeter of silicon has hit a pretty tough wall.

[stray badger]

maybe moar cores at some point

[soft bloom]

well, yes. i am talking about powerful relative to right now, or even 5 years ago.

[twin dew]

Rambler in action:
General Ryzen 7000/9000 AM5 CPU DDR5 and infinity fabric OC information
https://www.youtube.com/watch?v=Xcn_nvWGj7U

[civic mural]

Dudes, your thoughts on Typst? I kinda hate LATEX so it seems great for me

https://github.com/typst/typst
Source. ||God, anything but LATEX, please...||

[willow pike]

AMD can already stack CPU on IO die with integrated vcache, they do it in MI300, hopefully it'll get cheap enough for consumer

intel also has the same tech, foveros direct, but fuck knows if they even want to use it

[gilded helm]

MI300 is gigantic though. So far AMD hasn't used chiplets to give consumers an unusual amount of overall silicon, it's just broken up in pieces.

[willow pike]

yeah it's a whole other world

servers get star trek and we get scraps

[gilded helm]

It's the same reason HBM isn't coming back to consumers, even the crappiest cutdown versions of anything with HBM will probably sell for thousands.

[twin dew]

That is AFAIK the reason next Radeon generation isn't coming to highend.
There was a design, but it used chiplets with substrate tech that didn't pan out at reasonable price and production capacity.
Above the current gen substrate, but below the one used in MI300 and Nvidia equivalents.

And that substrate tech is just way over capacity for those datacenter cards for next year or so until TSMC can bring more capacity online.

[willow pike]

there's also the matter of all this packaging tech being highly production constrained

[twin dew]

That highest end one.

[willow pike]

in this AI gold rush, everything is being sold

so they would much rather dedicate capacity to AI chips than hope people actually buy a radeon this time

[willow pike]

probably InFO_LSI

[twin dew]

CoWoS-R AFAIK, which would have been just organic interposer RDL InFO without the LSI chips for data that are in the CoWoS-L
Which didn't pan out.
Current low-end is on CoWoS-S and the highend is on CoWoS-L

Instead of the silicon interposers in the CoWoS-S

Or might have been made to work, but would have taken capacity from the CoWoS-L lines pumping out the datacenter chips.

[willow pike]

yeag

that does make more sense as 7900 uses InFO RDL

man that's a shame

decided to use a tech years ago that they had no idea would skyrocket in demand

[twin dew]

Yeah, it AFAIK uses that CoWoS-S, and 8000-series highest end was designed ot use that CoWoS-R.

And isn't being brought to market for manufacturing reasons, as they just cannot do another pass to not need it in time for this generation when the production side didn't pan out.

[willow pike]

there's no credit to amd "not wanting" to make a high end radeon 8000

"yeah we decided we're gonna make mid range chips this time, just as planned"

[twin dew]

Ah, no, even the MI300 uses just CoWos-S:

AMD’s MI300 was originally designed on CoWoS-R but we believe due to warpage and thermal stability concerns AMD has to instead use CoWoS-S.

[willow pike]

lmfao yeah no fucking way is AMD taking away from MI300

that's their golden goose

god that's tragic

imagine you're on the radeon team

you spent years bringing up navi 44/48

then you see the AI boom

you know MI300 uses your tech

you go home and cry

[twin dew]

Based on what I can find, Navi 31 might have used TSMC InFO_oS
Which is below all those CoWoS variants.

Yeah.

So integrated RDL layer on the basic substrate.
Without interposers.

That CoWoS-S has silicon interposer, CoWoS-R would have had organic interposer and CoWoS-L has combination to give the high speed data traces silicon and others organic.

[willow pike]

yeah that's what was needed for the memory controller dies

[twin dew]

Same idea, more expensive and performant implementations in those steps.

[willow pike]

I think the only user of InFO_LSI is apple

[twin dew]

Where that -R was supposed to be lower cost alternative to the -S

[willow pike]

but they seem to have given up on it, ain't seen an M3 ultra yet

[twin dew]

That InFO_LSI is probably alternative name for that CoWoS-L.

[willow pike]

they are flipped

one is chip first, the other is wafer first

similar end results but different manufacturing

[willow pike]

yeah that's the end result

[twin dew]

The TSMC given reference pics are basically the same.

CoWoS-L is LSI chips embedded in the organic interposer for the data lines.
Like the InFO_LSI too seems to be.

[twin dew]

Ah, that way.

[willow pike]

two different ways to manufacture a similar result

[languid gulch]

so when they say that next gen radeon isn't targeted for the high end, have they mentioned where it's supposed to land? because the 7900XTX isn't that far behind the 4090, at least not like how badly radeon's been beaten flagship-to-flagship in the past

[twin dew]

7000-series had three dies, the 7900 used chiplet one and replacement for that Navi 31 will not be coming in 8000-series, so replacement for 7800XT and down only.

[glossy glacier]

Three? They have two different types, GCD and MCD. Always one GCD and up to six MCDs

[twin dew]

Navi 31 is the only one to use chiplets, with two variants (6MCD, 4MCD).

As in that 7900-series models (XTX, XT, GRE).

The other two are monolithic

Navi 32 and Navi 33.

Ok, might be wrong on the Navi 32 being monolithic.

[willow pike]

navi 32 used MCD

the GRE is a navi 31 with 4 MCDs

[twin dew]

Yeah, I hadn't caught that it had 4x and 3x MCD variants, and that only Navi 33 was monolithic.
And misremembered 7900XT being 4 MCD and not 5.

[willow pike]

we're saying they had plans to make a radeon 8900XTX, and now they will not make it

because they'd rather use their factories to make AI stuff

[glossy glacier]

Ah, lower end 7000 are monolithic? Didn't know that, thought all are chiplet based

[willow pike]

7900, 7800, 7700 are chiplet, 7600 monolithic

[twin dew]

...

He can report all he wants, when there isn't much CSS can do for him.

[willow pike]

idjit

[maiden coyote]

can't help those who arn't willing to help themselves.

[soft bloom]

sometimes i think that culprit of modern cloud computing is Kaggle

#off-topic-media message

[wanton orchid]

who cares that water as been put in the fuel tank, it's the car which throw a red light at me when I turn the key !

[verbal raft]

still chiplet 7600*

[twin dew]

No, Navi 33 is monolithic:

[willow pike]

ye, one chip

navi 33 used in 7600

[twin dew]

All the various 7600 versions, and 7700S laptop GPU.

[verbal raft]

oh sorry im super dumb

i thought you were talking ryzen

for no reason whatsoever

[glossy glacier]

understandable confusion xD

[verbal raft]

that guy is malding

[stray badger]

yeah they really should change that, maybe make radeon only 3 digits long?

790 GRE would have made more sense imo since the last digit is never changed

[edgy hazel]

Gets even more confusing with the 5700xt being a GPU and CPU

[willow pike]

yeah I'm not a fan of them using xt on cpu

ryzen is already at 9000 so they'll probably change that to three digits next time, they already did on mobile

[glossy glacier]

If they finally settle on a naming scheme for mobile...

They had like 6 different shemes for the last 4 architectures

[willow pike]

hope they ditch the ai shit soon

[wanton orchid]

it's an oem shit

[verbal raft]

it would be good but there is no way they are actualy gonna do it

[ivory rampart]

xt on cpu us confusing, but what intel is pulling right now is a lot worse

[jagged snow]

Starting to have problems
this is why I normally stop at a week

[languid gulch]

think my record is 28 days

[maiden coyote]

debian linux says it's just warmed up at 17 days lol

I'm at almost 80 days on my pve server

[jagged snow]

Yeah, there's an inane number of wake/sleep cycles on my desktop though

Probably average of like 10/day

[maiden coyote]

I've had windows 10 run fine for like 2 months I think, but no wake/sleep cycles.

[jagged snow]

Windows really doesn't like those

[maiden coyote]

my desktop keep bsoding so I'd be happy to get a day of uptime out of it.

I thought it was my cpu overheating on satisfactory... which didn't make a lot of sense for a modern cpu, as I'd expect them to hang, or slow to a crawl.

[jagged snow]

For me random BSOD was a bios issue

[maiden coyote]

I think something is up my with nvme drive now, as I'm not getting any crash dumps, and it doesn't send crash reports, and yesterday the nvme was gone from the bios menu and it wouldn't boot

[jagged snow]

Oooh
That's no fun

[maiden coyote]

it started only when playing satisfactory, then with just chrome open, and now at the desktop. tonight or tomorrow I'm going to try to reseat the nvme and see if that fixes it.

[mental oriole]

I've had 116 days on my laptop earlier this year (with sleep wake cycles that is) Worked fine. Lost the time because windows update.

[jagged snow]

Ahh wow
I've never had anything over 50 days

My desktop is worse about getting buggy after a while than the laptops I've had

[mental oriole]

It's a struggle with wandows.
I usually hit like 30 max on main pc.

[dire igloo]

*5600XT

[languid gulch]

almost like it's confusing 🤣

[mental oriole]

It is...

[winged valley]

I sure love learning stat

Although this chapter has actually been pretty easy to follow

[dire igloo]

I read "if a radioactive intern" instead of "if a confidence interval" and was very confused about what the hell you were learning

[sand saddle]

Does anyone know anything about android TV boxes?
Would need one for an older TV.
Maybe some of the aliexpress shit would even work? I got no idea.
Wouldn’t want to spend too much. At first though I could just use a cable and stream from a phone but the apple adapter doesn’t support hdcp.

[glossy glacier]

Chromecast, raspi+libreelec or get a cheap mini PC.
Depends on what you want to use and how to interact

[soft bloom]

cheap tv box is less expensive than either option

[glossy glacier]

🙃

[winged valley]

Do you mean like cable? No, it definitely isn't cheaper lol

[verbal raft]

I love windows update:)

[tough owl]

this is true correct?

games no?
and optimized for nvidia is just them having long standing cuda support where as amd has rocm which still sucks to this day

[glossy glacier]

Cuda and rocm are more for General computing rather than gaming. For gaming its DirectX, Vulcan and OpenGL.

Most games are more optimzed for nvidia because it has the higer market share

[soft bloom]

No xable. I mean like a small tv box with android that cna connect over wifi , has some hdmi and usb port

[dire igloo]

Zluda

Nvidia has a long standing collaboration with OBS which they use to optimize the synergy between Nvenc and OBS

[tough owl]

amd doesn't like it anymore

[dire igloo]

Which is ironic given that it beats ROCm

[tough owl]

That's probably why they don't like it

[maiden coyote]

since we mentioned uptime yesterday.

[glossy glacier]

Nice

[edgy hazel]

baller

[willow pike]

https://x.com/kopite7kimi/status/1850012915069563168

kopite7kimi (@kopite7kimi) on X

@Kepler_L2 Although PTL will reintegrate IMC into the compute die, NVL will once again separate and optimize it.

Twitter · Oct 26, 2024, 3:13 AM

why the fu

[maiden coyote]

you trying to tell me I need to update my kernel? lol

[edgy hazel]

nah why u got a server with a ryzen

4th gen intel gang lmao

(also update your kernel)

[night girder]

so apple magic mouse now has USB C instead of lightning. But the port is still on the bottom of the mouse 😂

[dire igloo]

reminds of Brennan Lee Mulligan: "I will die on ANY hill!"

[glossy glacier]

update your kernel 🙂

[maiden coyote]

Eventually

Because ryzen 5x makes a decent home server cpu. I should've gone for the x3d variant but this works well enough

[verbal raft]

IM SO FED UP WITH THIS GARBAGE

[dire igloo]

questionable source

[verbal raft]

the guy doesnt even have these two CPUs

[dire igloo]

wait what

[verbal raft]

oh well i reported his channel (nothing is going to happen) and left a sweeeeeet comment for him:

[dire igloo]

🚫 Beware of sketchy comparison videos 🚫
Some YouTube channels upload hardware comparison videos at a suspicious & unrealistic rate, sometimes several per day. There is no assurance that these channels use reliable and fair testing methodology, so the results in the videos could be inaccurate due to negligence, or intentional manipulation in accordance with the channel owner's agenda.

Try to stick to more reliable and reputable sources like Gamers Nexus, Hardware Unboxed (Techspot), and TechPowerUp.

someone made this copypasta once

[soft bloom]

who the heck names CPU "ultra 7 265k"?

did they name it as april fools joke but someone forgot to re-draw ideas?

[verbal raft]

i diets

[soft bloom]

or is intel run by 12 year olds?

[verbal raft]

suspiciosly fast upload rate:check:

[soft bloom]

500 views
seems legit

[verbal raft]

its not even the views

[soft bloom]

i honestly think that side-by-side comparison is quite a waste of time. especially when it's not on a streamlined benchmark (script)

[verbal raft]

like there are many small things that point to the whole thing being fake

[soft bloom]

ok i just don't want to even think of hardware media world. i will let it be

[dire igloo]

even if it wasn't fake, it'd still be inaccurate

[verbal raft]

yup

[dire igloo]

I meant the methodology, not the results

[verbal raft]

yeah that too

[dire igloo]

the results are a direct consequence of the methodology being "I made it the fuck up"

[verbal raft]

👍 😂 😂 approved analasis

[dire igloo]

hehe, anal

[pure karma]

idk but the name alone is setting this generation straight to the garbage especially when they could have made it the 230k,250k,270k,290k

[verbal raft]

yeah why not go with 290k instead of "285K"

it would be so much simpler

[pure karma]

yup

unless they are saving it for the next one and using this generation as a public test but its still atrocious

[verbal raft]

yeah the 290k could be the KASS of the ultra series (with how intel decided to name them)

or maybe they would name it 285KS

[dire igloo]

295KYS

now with double the voltage

[verbal raft]

And Half the Waranty for NO reason whatsoever:)

oh,and we dont garanty that your house wont light on fire 🙂

[glossy glacier]

Intel: What a resonable name! now substract 5 from it

[verbal raft]

14nm ++ +++

[soft bloom]

math is my favorite kind of anal

[verbal raft]

https://www.youtube.com/shorts/cI-SdJAf0Ow

same thing

pretty much

[dire igloo]

I mean, it's a top of the line CPU, of course they'll have it come with a free Gigabyte P850GM PSU

[verbal raft]

amds new laptop naming sceme isn't much better