Tech Lab forum - General Discussions!

[jade scaffold]

Oh no it lost that crown

The 4090 actually took the generational uplift crown

1080ti still balling value at release

But not quite as good value as people remember it to be

[mossy gull]

I still remember the 1080 days, not the Ti mind you.
When Resident Evil Remake released, everyone was struggling to run it at 2160p on highest settings on their 1080s only for the Vega FE to be able to play it on highest settings.

That's when for me the lack of VRAM on Nvidia cards started playing a pain in my butt😅

[jade scaffold]

I'm really excited to try DLSS 4

The fact I can force it on in any DLSS supporting game is great

I really wanna try the new reflex too but I don't think you can force enable that in any game

[mossy gull]

I thought you could, just like Anti-Lag

[jade scaffold]

There's a general latency reduction then a game specific one that acts further

[mossy gull]

Ah, fair enough

[grizzled oracle]

I have a cheap Chinese rgb 2x8 pin extension cable in my GPU, it’s connected to an RGB header and obviously directly to the power supply by the 2x8 pin GPU cable. The RGB functions after a system restart, but as soon as I attempt to address any rgb in the machine with open rgb, it turns off all of the LEDs in the cable except the ones at the very base of the GPU cable connection.

It came with a SATA controller that has a male rgb connector that would not allow me to use a header, naturally because the link would end in a male.

Have I connected it wrong? Is the NIB unit chinesium garbage and just faulty off the rip? Do I need to somehow find a magic way to connect the sata power cable to a header at the same time?

[jade scaffold]

Could be faulty the software could just be total asscheeks as RGB software always is

[grizzled oracle]

I would figure the circuit is broken based on how its lighting after addressed but the fact that its fine before being addressed would indicate it’s fine

[jade scaffold]

The controller on it could be shagged I guess

[jade scaffold]

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

[mossy gull]

If I get it I will replace the LM with PTM

[jade scaffold]

Why

[jade scaffold]

https://www.youtube.com/watch?v=lA8DphutMsY

So DLSS 4 not completely perfect but aparently shockingly good

i mean frankly AI gen or otherwise 200 fps in Cyberpunk at utterly cranked settings is insane

i want to see how much fps people get without Pathtracing that number must seriously start nearing the 600fps mark with dlss

429fps in 2077 at 4k max settings with dlss on the 5090 without path tracing

also the 5090 is extremely quiet

thats insane

[mossy gull]

I wonder what the rest of the system specs are(didn't watch the vid).

[jade scaffold]

its either a 9800x3D or a 285k

[mossy gull]

I think the 9800X3D, or another chip, but not the 285k, that CPU ain't even that fast.

Especially not in games

[jade scaffold]

the 285k with the right setup can outperform the 9950x

the 285k is faster now than it was at launch thanks to intels optimisations

if you run it with fast ram it can keep up with the latest non x3D chips

[mossy gull]

Still makes the 9800X3D run at least 10% faster though 😅

[jade scaffold]

yea but CPU speed doesnt matter when you are GPU constrained

unless you got like super mega driver overhead

[mossy gull]

Yes and no, depends on the game really.
If a game really loves X3D then the speed of the CPU will matter.

[jade scaffold]

no it doesnt

i have had a cpu with more cache than ANY consumer cpu and far more memory bandwidth

CPU speed doesnt matter

Cache only effects .1% lows

[mossy gull]

So do I, but my 7970X ain't faster than a 9800X3D in games, it's actually slower because of the lower clockspeed.

[jade scaffold]

thats because your 7970x is a TR

has no memory bandwidth

has no GMI wide

[mossy gull]

It has 128MB cache

[jade scaffold]

so its ununified cache

[mossy gull]

Still, speed of the CPU does matter, not as much as on lower settings, but it still matters, of it didn't then the 285k and 9800X3D would be on-par, but the 9800X3D is significantly faster

[jade scaffold]

the Epyc i had has 256mb

Unified cache and would actually boost right up to 4.8ghz still lower than the 9800x3d but i did not get any performance disadvantage outside of .1% lows on a 7700x

CPU speed DOES NOT MATTER if you are heavily gpu constrained

unless your driver overhead is insane like intels currently is

which is why their gpus perform massivly better with super fast CPUs

i mean the 9800x3D literally pulls like 20w in cyberpunk 2077 at 4k its almost literally sat idle

all its doing is passing data

and not alot of it because you arent pulling 800 fps in cyberpunk

[mossy gull]

But what about the 285k?

[jade scaffold]

probably not much worse

it'll be more because it IS less efficient

but it'll still be borderline nothing

[mossy gull]

I am still curious how much impact it truly has

[jade scaffold]

none its like 3fps tops

where the 9800x3D will win out is likely minimums

[mossy gull]

Well the weird thing is 1 thing I heard that the 14900K was definitely faster in Cyberpunk than the 7800X3D in 4K Ultra Settings, path tracing wasn't a thing yet when I heard about this.
Nor do I have a 14900K to test it with😅

[jade scaffold]

if you use DLSS it will be

because you arent running 4k

at native 4k well you wouldnt be able to tell the difference between the 2

because Uh

21fps avrg

[mossy gull]

I don't know if people use DLSS, they most likely do because "it increases fps"

[jade scaffold]

yea because 21fps is fucking unplayable

Pathtracing is hard

maybe Intel CPUs lower latency gives them an advantage in path tracing though

AMD do have pretty high core to core latency comparitivly

[mossy gull]

Path tracing wasn't a thing when I heard about it😅
This was also around the time it wasn't known the 14900K committed Harakiri

So way way back

[jade scaffold]

that'll be like 30 serise then

but it cant have been

[mossy gull]

40 series were released

[jade scaffold]

path tracing released with the 40 serise the 14900k came out the same year

[mossy gull]

Just not path tracing

[jade scaffold]

no path tracing was a release "feature" for cyberpunk of 40 serise

im saying feature because its still not nativly runnable

mind you if you arent at 4k the 4090 can actually do path tracing native at 1080p and get like 80fps

[mossy gull]

It struggles in 3440x1440 though, oddly enough😅

[jade scaffold]

which considering pathtracing is what they use when rendering movies just 8 years ago it would literally have taken hours to render a frame with it 21 fps in real time is

[jade scaffold]

that is 3x the resolution

so you would only get like 25fps or something

[mossy gull]

Something like that, but the FPS is only slightly faster than 4K

Yep

[jade scaffold]

yea

like i said though

1440p ultrawide doesnt seem hugely bigger

but thats a higher resolution than the valve index has

[mossy gull]

Curiosity from my part, is the bigger L1 and L2 cache on the 285k gonna matter for games?

[jade scaffold]

my monitors resolution is pretty crazy tbh
5120x1440p its like just shy of 4k resolution in terms of pixles

[jade scaffold]

kinda

its a factor

[mossy gull]

I want LG's 5120x2160 panel😅

[jade scaffold]

i like my wiiiiide boi

[mossy gull]

Because X3D is I thought primarily L3?

[mossy gull]

I like the idea, but I dislike the incompatibility with Asian games😅

[jade scaffold]

but saying that people say oh modern games run like shit but i can get 60fps in what is basically 4k high in any game on whats now equivilent to a 4070ti / 7900xt/gre

[jade scaffold]

i just split the screen

can turn it into 2 virtual screens

run game on one side run what ever i want on the other

[mossy gull]

That will feel odd for me as I sit 1.5 meters away from the display in a simrig😅

[jade scaffold]

yea its equivilent to 2 27" monitors stuck together

there is one equivilent to 2 32" ones as well

but its not oled i beleive

[mossy gull]

It's just QD-MiniLED I believe

I could be wrong

[mossy gull]

I think mine are fine with their 45".

But there's also a chance I am just going to get a big LG C4 instead of 2 45" UW monitor

[cinder lagoon]

Btw:
if anyone ever wanna fiddle with iVentoy or unattend.xml powershell scripts, ask me. I now know how to work with them

no, it is not trivial.

no, it does not "just work"

[jade scaffold]

theres an QD OLED of the 49" G9 but im not sure about the 57" they might have made an oled one now tbh they make new versions of the fucking thing every few months for what ever reason

[mossy gull]

I meant the 57" to be a QD-MiniLED, or whatever what they call it😅

[jade scaffold]

just mini LED

just means its an IPS display basically with full array local dimming

[mossy gull]

Ah, the worst kinda Mini-LED 😅

[jade scaffold]

no thats the best kind

QDs dont apply to mini led displays

[mossy gull]

I seen today a TV with the marketing of QD-MiniLED, it was a TLC in a shop that isn't always trustworthy😅

[jade scaffold]

that was micro led

which is different

kinda unhelpfull nameing schemes very similar

[mossy gull]

Not micro-led, it was just 8000.-

[jade scaffold]

wait no QD mini LEDs are a thing

how the fuck does that work

a QD needs to be powered

[mossy gull]

For us a question and from them the knowhow

[mossy gull]

TCL is Chinese, they lately make all kinds of weird things work

[jade scaffold]

the issue is QD is a very

uhh

meaningless term

A quantum dot translates to "light Dot"

[mossy gull]

To be honest, QD panels often look too blue for me, usually because their blue backlight and I am sensitive to blue light

[jade scaffold]

they COULD be using quantum dots to block out the backlight

the backlight colour is entierly down to the vendor

entierly the literall backlight if the backlight isnt perfect white then it will end up looking colour

[mossy gull]

Samsungs are very blue, which to me makes them always look cold and blue for me, or well their QD-OLED do.

[jade scaffold]

Quantum dots OLEDs have a blue emitting layer

thats then passed through a QD layer per pixel

[mossy gull]

That might be why then😅

[jade scaffold]

also samsungs displays are nearly over 100% accurate

atleast the OLEDs are

you shouldnt see any blue leakage as the QDs arent transparent unless powered in a certain way

seems QD mini LEDs work the same way QD oleds work but with a actual backlight array

which means you maybe might get some blue light leakage from thoes

because the backlight isnt pixle per pixle

[mossy gull]

I doubt it's leakage, it's the same I still see UV with majority of "99%" UV blocking glasses, just less than usual.

Some types of light are simply stronger than others that most people don't see, so hard to focus on something you don't see.

My 45GR95QE-B are also on 40% brightness or I get blinded

[sacred seal]

Wow, without the new dlss the 50 series is not looking so hot. 5070 is only about 15% faster then the 4070 with only a 50 dollar lower launch price

[mossy gull]

Every year the same story 😅

[sacred seal]

If it was a single year it would not be so bad, but the 4070 came out closer to two

[jade scaffold]

Based on?

[sonic meadow]

So cheaper and faster isn't a good thing?

[jade scaffold]

15% isn't a great generational uplift

[sacred seal]

Apparently Nvidia https://www.pcgamer.com/hardware/graphics-cards/heres-how-the-new-rtx-50-series-cards-perform-against-the-previous-generation-of-geforce-gpus/

[jade scaffold]

Nvidia website shows 30% ish

[sonic meadow]

The difference is that these are supposedly cheaper not the same price or more expensive

[sacred seal]

For the 4090vs the 5090

Which is also the only card that costs more this Gen (I think)

The other cards are more modest

[sonic meadow]

ie dlss+framegen

[sonic meadow]

this comment (which was yours btw) is talking about with all that stuff off

[jade scaffold]

Still a poor generational uplift

[jade scaffold]

It's not the first 2 results are rt or dlss3 so show a more direct comparison than some of the others

[sacred seal]

Yeah, only the first two results are relevant. Dlss4 looks to be amazing, but I use my gpu for more than just gaming

[jade scaffold]

Both of them show about 20% not 15

But I guess this is what happens when AMD fumbled 4 out of 5 of the last generations

[sacred seal]

According to the article they said 15, but even at 20 that is not great for two years

[jade scaffold]

Yea it's not great

35% is about where you want it to be at

Ideally higher

This is what you get though when there's no competition

Makes the 5070 about as fast as the 4070ti

[sacred seal]

Yep. Going to be like Intel 6th Gen core parts. Going to be drip drip drip until amd drops something disruptive like ryzen

[mossy gull]

Nvidia focuses so hard on their DLSS and AI features, I think they forget to actually make a better card😅

Jokes aside

I think it simply becomes harder to make a faster processor, so I don't blame them for it'

[jade scaffold]

The 5070ti about as fast as the 4080

And the 5080 still slower than the 4090 though you are asking for a near 1.5x increase on a node that isn't actually smaller than last gen

Which is probably part of it being cheaper

The 5090 is also extremely power limited

[jade scaffold]

Just done some quick maths the 5090 could be 2x faster than the 4080 based on the cyberpunk bench marks 35% improvement over the 4090

[mossy gull]

But Cyberpunk seen some serious performance improvements though, so also keep that in mind

[jade scaffold]

Yea that's irrelevant

The 4090 is still 1.5x faster than it and the 5090 is still 1.35x faster than the 4090

[mossy gull]

That's still true yes

[jade scaffold]

~2x the raster power of a 4080 is kinda insane

[jade scaffold]

https://youtu.be/itpcsQQvgAQ?si=y8-ozSl60hs-Shk5

[mossy gull]

It finally support backwards compatibility 😅

Sort of

[timber barn]

A benefit is the better cooling I guess

[sacred seal]

Well the 4080 super launched last year at 1000, so if Nvidia is charging 2000 for the 5090 I would hope that it is at least 2x more powerful

[mossy gull]

That is a very valid point 🤔

[mossy gull]

For everyone who says 8GB VRAM is enough.
https://videocardz.com/newz/pcgh-demonstrates-why-8gb-gpus-are-simply-not-good-enough-for-2025

[sacred seal]

The article does not really prove anything. A GPU with more vram will preform better, and that has been the case forever. The question is if cards are going to be unable to play certain games at desireable settings because of their lack of vram. If increasing the vram gets you from 10fps to 20fps, you are still running the game at too high of settings.

You are probably not getting great framerates to begin with if you are trying to do raytracing in cyberpunk on a 7600xt

[mossy gull]

The idea behind it is because people still say "you don't need more than 8GB VRAM".
And the 7600 and 7600XT are the closest in performance between the 8GB and 16GB variants.

Last time I spoke someone who literally said games don't use more than 8GB VRAM so you don't need more, their evidence was their 1080 that didn't use more than 7GB VRAM.

[sacred seal]

It depends on the res and setting you are playing at.

[mossy gull]

I mean, 12% already on 1080p, that says something about how heavy games are becoming😅

[sacred seal]

Pretty well every game is playable with 8gb of vram (or at least I dont know any that won't) Actually most games will run with just 4

The 1650 is still the 4th most popular GPU according to steams hardware survey

By the looks of things about 20% of gamers have 6gb of vram or less

[mossy gull]

Because it's popular doesn't mean it's good.
For example Samsung Galaxy S24 is 1 of the most popular phones and seen as a phone with 1 of the best cameras, but it often falls short compared to comparable devices.

[sacred seal]

Yes, but we are debating what you need. A 4090 will outpreform any other GPU, but there is no game that needs it to run

A gtx 1080 is still a very capable 1080p card in 2025

[sonic meadow]

that has definitely not been the case forever. as long as you havent filled the vram dont need more vram than is available performance is going to be exactly the same.
historically this basically meant stuff either ran, or ran weirdly.
in modern stuff however... engines do all kinds of scaling based upon how much vram you have, independent of any settings you might have set

[sacred seal]

Fair. I was more refering to that you could run games at higher settings with more vram, not actual fps

If you had less then the required amount of vram your performance would take a significant hit as it was constantly swapping stuff in and out, or you would get wonky stuff

I was wrong, 1-6gb of vram account for about 30% of gamers on steam. 8Gbs account for another 35%. No mainstream dev is making a game where a full 65% of the potential market is unable to run the game.

[mossy gull]

And it quickly is falling short, it's capable yes, but if you have a 1080 with double the VRAM, it will perform better.

It's the same debate with a ton of stuff, 3 cilinder turbo-charged engines, phones that apparently make pictures the quality of a professional camera, GPUs with too little VRAM etc etc.

I seen the 6700XT beat the 3070 in Cyberpunk Raytracing because the 3070 ran out of VRAM in 1440p while in 1080p the 3070 took the crown back again.
Just the little notch of memory can make a big difference at times.

[mossy gull]

If we keep pleasing people who have little to no VRAM then we can't make our games look better, or else we still would be stuck on 256KB VRAM.

[sacred seal]

We have had the debate about your 6700xt and 3070 half a dozen times now. We have well established at this points that the test you preformed were flawed to the point of irrelevance.

[mossy gull]

Still, amount of VRAM is important you can't keep hanging onto just 8GB VRAM.

And yes, the 1660 and older GPU already can't play some newer gales because the lack of Raytracing cores

[sacred seal]

That is a bit of a nonsensical argument. The amount of vram you get per dollar goes up over time, therefore systems will gradually increase in vram capacity. Developers follow that trend and build games that preform on the hardware that is widely accessable

[mossy gull]

Tell that to Nvidia who still will sell GPUs with 8GB VRAM while Intel and AMD will probably provide at least 12GB VRAM

[sacred seal]

I am aware of like one game that requrire RT to be able to run, and it was some indie title.

That's nothing new. AMD has always beeten out Nvidia on the ram front

Well mabey not always, but for the last 12 years at least

Part of the reason why AMD cards remained relevant longer. They could keep playing on higher settings even when the nvidia cards that initially beat them started needing to drop down a few tiers.

[mossy gull]

Things could change overtime, and I hope Nvidia will learn that 8GB simply ain't enough, not even on their GDDR7 with AI optimisations.

But I guess time will tell, and we do need to get more VRAM, I wish upgradable VRAM was possible without technical knowhow.

[sacred seal]

10 years ago 2gb was mainstream, and 4gb was high end

The 390 increased it to 8gbs, the 1080ti brought it up to 11, and I think the 6800 was the first card with 16

3090 upped it to 24, and the 5090 will go to 32.

But just because we have a 32gb card now does not mean that the 8gb card is no longer useful

Even last gen I think the only GPUs that had 8gb of vram were the 4060 and 7600.

[mossy gull]

It was actually the Vega FE, which had 16GB HBM2, and 1 of the fewer cards that was capable of running Resident Evil Remake on highest settings on 4K at the time, but what idiot would pay 1K for a GPU at that time🥲

[mossy gull]

And the 5060 is rumoured to have it once again

[sacred seal]

5060 is marketed at a 1080p class card, so pretty reasonable for it to have 8

[mossy gull]

And the B580 I think it was, the Battlemage, has 12GB, and is a 1080P card

[sacred seal]

580 is marketed as a 1440p card as far as I am aware

Its intels best card

[mossy gull]

At RTX 4060-ish performance

[jade scaffold]

It can hit 4060ti performance in some cases

[sacred seal]

Official intel marketing
a GPU designed for 1440p Ultra Gaming, and will deliver performance that exceeds that of NVIDIA's GeForce RTX 4060

Not really true, but that is what its aimed at, and that is why it has more vram

[jade scaffold]

8gb is enough for 1080p at lower settings

[mossy gull]

If Intel is good at 1 thing it is making bad marketing😅
I remember the days of Dual-core i7

[sacred seal]

Don't think there was ever a dual core i7 outside of the laptop space

And that was when dual cores were still mainstream

[jade scaffold]

If you max settings in 1080p you can quite easily eat 8gb at 1080p especially if you use raytracing

Rt is enormously memory intensive

Probably why the 5090 has a 512 but bus

[sacred seal]

Yes, but are you getting 30+FPS on a card with only 8gbs of vram in that case?

[mossy gull]

People these days expect you do

[jade scaffold]

On a 4060 easy

The 4060 is still really quite powerful

[sacred seal]

Looking at some benchmarks a 4060 will just barely do 30fps with RT ultra and dlss when it can do about 60fps without rt and dlss. I don't think just going up to 12gb of vram would substantially improve its performance.

[sonic meadow]

omg, rt hurts performance?

[sacred seal]

Obviously yes. The point I am trying to make is the 4060 (which I think is the most powerful 8gb card out there?) is not really powerful enough to be making use of full RT, and therefore the cost of adding an extra 4gb of vram does not make a tone of sense of that use case

[mossy gull]

RT Ultra, aka something the card wasnt build for.
When a GPU runs out of VRAM it will use RAM as VRAM, making it slow down.

[sacred seal]

Yes, that is the point I am making. These cards are not built to run games where the 8gb of vram becomes the bottleneck

If you are hitting that 8gb limit, you are likely exceeding the reasonable settings for the card

[mossy gull]

Star Citizen uses 10GB VRAM in 1080P on a 7600XT, Forza Horizon 5 does the same thing.
Flight Simulator and Jedi Survivor also love VRAM.
STALKER 2 also loves some extra VRAM and so does Indiana Jones.

The fact is, 8GB is falling out of fashion, yet Nvidia keeps putting it on their cards.

[sacred seal]

And again, reported vram usage does not equal actual vram usage. Getting actual proper vram readings is a time consuming endevor that basically no one does.

Just looked up stalker 2 benchmarks for the 7600xt vs 7600. At 1080p you go from the 55-60 range down to the 48-53 range. A noticeable drop to be sure, but its not crazy

[jade scaffold]

Rt ultra isn't for the 4060

[sacred seal]

I know. I literally just said that
Yes, that is the point I am making. These cards are not built to run games where the 8gb of vram becomes the bottleneck

[jade scaffold]

Except there are several games where it does become the issue

GTA 5 even gets dangerously close at cranked settings to 8gb

The main issue imo is the 4060 is a 350us card with entry levels of vram

But just having more vram on each chip doesn't solve the 4060s issues

It's the bus width that's the main issue more memory would just be a nice side effect of Nvidia not turbo cheaping on bus width

Also vram does somewhat effect performance GPUs and CPUs even dynamically compression RAM depending on usage

More compression = slightly higher processor to memory latency not gonna make much of a difference really but still it's a thing

[mossy gull]

For 50 less, sometimes even less than 300 you can get the 7600XT and the B580.
Which in performance are pretty close to the 4060 and have more VRAM, so for non raytraced VRAM heavy games, both AMD and Intel are a better choice, unless you want to go older and get the 3060.

[sonic meadow]

both of which are rated at 75w more (190w vs 115w). which if you are somewhere with expensive power, or care in general...

[mossy gull]

Around 10 minutes, an RTX 3070 and a 3070 modded to 16GB.
https://youtu.be/T5mHQ3z6j2g

[sonic meadow]

so you are getting... less than 10% more performance (according to the techpowerup relative performance chart) for 65% higher power consumption

[mossy gull]

If you have a 7600XT from AsRock it can even pull from my readings 260w, for a 7600XT, which makes no sense, I guess that's why AsRock gave it 2 8-pin connectors😅

So if power is expensive, DO NOT get an AsRock card😅

[jade scaffold]

Yea NVidia hold an INSANE efficiency advantage with the 4060 being their second most efficient consumer card

[sonic meadow]

the entire 40 series had insane efficiency. except maybe the 4090, but you lost basically no performance by cutting the power target drastically

[jade scaffold]

The 4090s efficiency though is insane even compared to the 4060

[sonic meadow]

the architecture is insane

[jade scaffold]

No the 4090 by default is the most efficient

But you can cut it's power to like 350w and it's almost not slower

Making it 1.5x more efficient than AMDs 7900xtx or 1.8x more efficient than the rtx 3090

It's genuinely silly

[mossy gull]

Just wait till you shunt mod the 7900XTX and make it run as fast as the 4090😅
Then you can see the 7900XTX using twice the power of a 4090

https://youtube.com/shorts/i8mgIhq9j2U

Not related but funny

[jade scaffold]

You know what 2 things I'm exited for next gen?

5090 with a shunt mod

Because we know at a minimum 10% of it's performance is locked behind it's low power limit

So shunt mods on it gonna go hard

And then I'm actually excited for nvlink on the quadros

[mossy gull]

I was surprised when the 5090 got a single 12-pin, I thought rumoured it would have not 1 but 2 12-pin.

[jade scaffold]

Blackwell's nvlink on the ai accelerators makes all programmes view the nvlinked GPUs as 1 singular instance GPU meaning no program based support is needed to use 2 GPUs

IF that applies to the quadros

The fastest gaming system is going to be a pair of nvlinked quadros totaling 44,000-48,000 cuda cores with 128gb of combined vram

And I would love to see that

[mossy gull]

Now you're speaking my language

[jade scaffold]

Also I might need to upgrade my CPU for the 5090 I have a 7700x currently

9800x3D here I come ig

[mossy gull]

It can go worse, I am gonna use the 285k, just to annoy myself once again with Intel bullcrap

[jade scaffold]

You know the I gpus in Intel's new desktop chips are apparently really good

I want to try one

But I'm waiting for the I3 to release which has 4p 4e cores and the same IGPU might be a super awesome Uber budget gaming chip

[mossy gull]

Eh, I don't really care about their iGPUs, never really did, they can say how good they are and still getting beaten by an APU half the price

[jade scaffold]

They aren't surprisingly

They are genuinely kinda amazing

It's a tiny lil battle mage GPU

Overclockable too iirc

[sonic meadow]

and intels video hardware is really good

[jade scaffold]

That mixed with a CCAM 12ghz ram module might actually be an awesome sff machine

[sonic meadow]

amds apus are solid, but at their top end make for kinda lousy cpus

at least historically, unsure about their am5 ones

[mossy gull]

I have been waiting for CAMM2 for over a year now, when are we actually gonna get it😐

[mossy gull]

Ryzen Ultra

[mossy gull]

To be honest, I can't wait to have a Surface Book alternative with more "VRAM" than the bloody x050 series GPUs

[sonic meadow]

The joys of a hard drive

[mossy gull]

Who still use a hard drive for the OS?

[sonic meadow]

Friends old laptop

It's been running non-descriptly badly

I'll be throwing an SSD in, and hopefully cramming some more ram in as well

But it wanted to update, and had 112 days of current uptime

[mossy gull]

Oh no, at least not the record I seen few years ago with an up time of 3 years.

Where I told the owner a reinstall is faster

[sonic meadow]

This laptop sees no use, so that's probably been since she purchased it.. about nine years ago

[mossy gull]

The other 1 I worked on, the owner thought, to turn it off, you just close the lid.

And after 3 years it was too slow for her to do anything

[cerulean helm]

Started building my "AI" PC. Even with this mid tower Fractal North case the 2 MSI Gaming X Trio 2080 Tis just barely fit. Add in a Dark Rock 5 cpu cooler and this case is completely full, lol.

[cerulean helm]

New PC posted. And Linux OS installed. Installed Pop OS and man did I run into snags. Tried installing the iso that had the Nvidia drivers and every time my OS install just went blank. Took me an hour of messing around till I tried the iso that didn't have the drivers, and the install worked first time...

[cinder lagoon]

Surprise surprise - the nvidia linux drivers don't work 😝

[cerulean helm]

They do, just had to install them after I got Pop OS working. Install via terminal worked fine.

[cinder lagoon]

Huh

[cerulean helm]

Yeah, not sure what the issue was but the iso with the nvidia drivers didn't work, so I assume those drivers were either out of date or just the wrong ones. Cause I am now running the latest drivers with the 2080 tis.

[mossy gull]

AI PC, all you need is a brand new CPU or GPU from AMD, Nvidia or Intel😂

[cerulean helm]

The "AI" PC is for a locally hosted LLM to feed Home Assistant Voice Preview Edition to make my own personal smart assistant that is not cloud based.

[mossy gull]

It was meant as a joke as they mentioned AI at least once every few seconds

[cinder lagoon]

Have you done the HA AI setup yet?
A friend of mine is also currently experimenting and has asked me for input but I've never done it before

[cerulean helm]

Not yet, but I have several YT videos that go through the install when I get to that point.

Its way easier now the HA has their own voice assistant though.

[cinder lagoon]

Aight thanks ^^

[cerulean helm]

A good video for demo:
https://www.youtube.com/watch?v=w9BbjUowmnE

[sacred seal]

Super interesting video, expecially since I was working on similar tech about 3 years ago.
2 2080tis is an interesting choice, if you are local hosting your LLM I am guessing you are doing something like llama 3.1 8b

[cinder lagoon]

Nice, thx

[vestal stirrup]

To the mob of the tech lab forum, could someone identify this.
It was handed in at my work but no idea what it is.

[cinder lagoon]

More angles?

[cinder lagoon]

Looks like some kind of stainless steel hose adapter

[cinder lagoon]

Ohh, interesting

[sonic meadow]

well thats an odd one 🤔

[vestal stirrup]

At first I though it was a tv antenna connection

But it is riveted on the inside

[cinder lagoon]

What is the context? I.E. What do you work as?

[sonic meadow]

is it antenna plug sized?

ie tv antenna, rather than wifi antenna

[vestal stirrup]

Yes, I would say it is the right size for a cable connection

[cinder lagoon]

The cup lid gives a pretty good idea how big it is 🤔

[vestal stirrup]

I work in housing, so could be anything xD

[sonic meadow]

I didnt place it as a cup lid 😛

[cinder lagoon]

The closed end confuses me

[sonic meadow]

this would you say? https://www.matchmaster.co.nz/connectors-and-tools/08mm-t1/

[vestal stirrup]

I think you have it

The end is thinner on that example but it is near exact

[cinder lagoon]

Lol wtf is that used for

[sonic meadow]

a terminator does.. signal stuff at the end of cable

[cinder lagoon]

Wibbly wobbly timey wimey?

[sonic meadow]

I never quite understood myself, just that you needed them for 10base2 networking (ie coax)

[vestal stirrup]

Thank you all the help, it just increases the mystery of how it ended up at my work

[cinder lagoon]

...interesting wording 😂

[sonic meadow]

if you work in housing, it could have come attached to a bit of tv antenna equipment, and just.. dropped

[cerulean helm]

I had 2 2080 tis lying around not being used. And they work for LLMs. At least they can go back into a rig instead of sitting in boxes, lol. And yeah, using llama3.

[cinder lagoon]

Apperently they used for testing
https://www.youtube.com/watch?v=NfnsfwXoGo8

OHHH its an amateur radio testing "antenna" that is more reliable than a proper outdoor antenna because of weather etc.

[sacred seal]

Ah, you already had the GPUs. That explains it. Shame that FB did not release a 16 or 32b model of llama 3. 8b is fantastic for what it is and crazy fast, but its still a bit derpy

[west mirage]

hey quick question

I currently have 2x16GB CL16 3600mhz trident z-skill ram installed

I want to upgrade to 4x16

of I get 2 more of the same model of ram

is that fine?

or could there still be compatability issues, Ik ram is finnicy

[mossy gull]

It's fine, but RAM might not run as fast as the rated 3600 MT/s.

[west mirage]

well ik but they should be onpar with my current ram

considering i'd be buying the exact same thing I originally bought

[mossy gull]

Still, you're running 4 ram sticks over 2 channels, instead of 4 sticks over 4 channels.
So overall speed can go down depending on the CPU and mem controller, but also can stay the same.

Other than that you should be able to use all 4 without problems.

[jade scaffold]

which is fine at low speeds

like 3600mhz

its only an issue really if you are trying to hit the upper limits of ddr4 like 4000mhz or 4400mhz

[mossy gull]

Aye😅

This RAM stuff is confusing

[cinder lagoon]

Yeah it is..
still not entirely getting why DDR5 is specced in mega transfers and not in mega hertz anymore..
I assume it's mostly to mask the fact that the clock speeds mostly stayed the same while the latencies went up

[mossy gull]

Because the MegaHertz was wrong and it should've been Mega Transfers per Second all along.
Because the MegaHertz is only half of what MT/s is.

It doesn't really sell if you sell 3000 MHz RAM, it will sell much easier when it's 6000 MT/s

[cinder lagoon]

Thats what i said

The transfers do not represent the actual data rate tho

Also DOUBLE Data Rate has it in the name

[mossy gull]

People don't know what DDR means 😅

I often get that people have 1TB of RAM, while they actually mean 1TB SSD

[cinder lagoon]

Thats not the people configuring systems tho

[mossy gull]

What about 1st time builders who buy an i3 and pair it with a 4080

[cinder lagoon]

They dont needa worry

They got other issues

[mossy gull]

So true😅

[mossy gull]

I have known someone who paired the 2080 Ti with a Celeron, because he believed gaming only uses the GPU.

[cinder lagoon]

Never actually seen one.
Everyone who built and was above the age of 13 has done some kind of research because they didnt wanna throw their hard earned money out

[cinder lagoon]

Finding ressources that support that conclusion is becoming more and more difficult nowadays

[mossy gull]

I last time closed my Reddit in less than a second when I found someone who paired the 4770K with a 4090 and 8GB RAM, because the gigahertz of the CPU is faster than the Ryzen 5 or i5.

[cinder lagoon]

Could also be rage bait tho

[mossy gull]

I mean, I still see people who buy a 4080 with a too small case😅

So people don't think

[cinder lagoon]

Glad the 5090 is small tho

I'll be expecting a 5090 Ti some time tbh

[mossy gull]

I am happy too

[jade scaffold]

because mhz makes no sense

DDR4 was never in hz

because DDR4 "3200mhz" is actually 1600mhz

DDR5 is the same DDR5 4800 is 2400 mhz

but 4800 mega transfers

[cinder lagoon]

Wait so you're saying that the 3200 was already doubled 🤔
Hang on I think I get it now

[jade scaffold]

yes

DDR stands for "Double Data Rate"

[jade scaffold]

its not small but its reasonable

its still VERY tall for a GPU

big 115mm fans on it

but thats an improvement on the 4090

its pretty much identical in size to the 4090

but 2 slots instead of 3

JayzTwoCents has a vid unboxing it

https://www.youtube.com/watch?v=LcntTXZ4QTo

[jade scaffold]

i kinda doubt it

the 5090 itself has no competition

and will have no competiton this gen

hell even next gen AMD might not be able to close that gap

AMDs GPU rx10000 or what ever they decide to call it will have to be in excess of 2X of their current and last gen GPUs

[cerulean helm]

Okay so I'm running into a strange problem trying to link my Home Assistant (Its on a Home Assistant Green) And my "AI" PC using Ollama. There is an intigration in HA for Ollama but it won't connect. Here is the strange issue.

I can run Ollama on the AI server, and check that its running by going to the localhost ip address and port. I also can login using WebUI to the models I have running on Ollama. But, when I enter the actual ip address of the AI PC and the port for the Ollama API I can't get a connection. And this is while I am on the AI PC. When linking Ollama and HA you need the url and port its running on. Since HA is running on a separate machine I can't use the localhost, I have to use the ip address. Ollama is running via a docker using WebUI on the AI PC that runs Pop OS.

Steps I have taken:
Ollama runs on 127.0.0.1, and I have switched that to 0.0.0.0 for all interfaces to connect.

I am kinda at a loss of where to go from here.

For reference I used parts of this video guide.
https://www.youtube.com/watch?v=XvbVePuP7NY&t=1375s

[cerulean helm]

Okay, after failing using chatgpt for help and completely messing the entire thing up till webui was in a boot loop I just purged the whole thing to start over, lol.

[cerulean helm]

And an hour later if terminal use and an hour chat using chatgpt for assistance I finally got ollama to inigrate with Home Assistant, good grief.

And holy shit, I finally have the Home Assistant Voice Preview Edition synced with a LLM and HA to use voice commands on all my iot devices. I now have a locally hosted amazon echo or google home. I am so damn happy this works.

[cinder lagoon]

What was it?

PopOS Firewall?
Docker container IP not passed through properly?

[cerulean helm]

I had Ollama installed via docker using WebUI rather than just installed directly on the linux pc. Was fighting tooth and nail with ports and then discovered only today that I had to change the environmental variable from 127.0.0.1 to 0.0.0.0 to connect to all devices, and it refused to change or work.

Ended up just starting over and installing Ollama via curl command and changing the everything that way and it worked.

And now that I am done I am off to bed.

[jade scaffold]

Very interesting architectural change here

The entire SM is capable of both int 32 and fp 32

This probably explains why Blackwell is so insanely faster at AI stuffs

[jade scaffold]

4090 Vs 5090

[sonic meadow]

It will be nice having slimmer cards

The last few gens have been gigantic

I wonder how well these will work when stacked... Doesn't look like there would be much breathing room between, so a lot of second hand (ie warm) air

[jade scaffold]

Yea

Though from nvidias thermal graph it can probably manage it without issue

The thing is QUIET

Like shockingly so

[cinder lagoon]

Question:
Is it possible to share the same LLM between multiple HAs with multiple conversations?
Because it would be much easier in terms of cost lol.
Not like you're talking to that thing 100% of the time, might as well share to a friend or someone

[cinder lagoon]

The 5090 disperses like 0,39° per cubic centimeter whereas the 4090 only did 0,22° or something

Hang on lemme find the graph

Ehh close enough, still crazy tho

[jade scaffold]

thats just TDP vs volume though

[cinder lagoon]

Then again... took em long enough and the generational leap is not as big

[cinder lagoon]

Shows how efficient the cooler is

[jade scaffold]

the 5090 is also closer to 33% averaged from the slide

and in cyberpunk its 35%

not really sure how derbaur is getting 27%

[cinder lagoon]

27% from official slides and leaks

Iirc

[jade scaffold]

its not

[cinder lagoon]

Lemme double check rq

[jade scaffold]

the slides average out 33% between all 2 graphs nvidia gave us that arent DLSS 4

then we have a video of cyberpunk running

28fps vs the 4090s 21

which is also about 33%

[cinder lagoon]

Segment starts here
https://youtu.be/EAceREYg-Qc?t=647

Yea they were measuring pixels from an nvidia slide

[jade scaffold]

then hes got completely different results to everyone else

[cinder lagoon]

Like there was no number but they scaled the performance bars and got the number from that

[jade scaffold]

not sure again

how hes managed that

[cinder lagoon]

Not according to Toms hardware

[jade scaffold]

that was just geek bench

something natorious for either hyper inflating results or hyper deflating them

[cinder lagoon]

If you wanna hear exactly and why It's in the first 1.5 Minutes from this starting point

[jade scaffold]

yea no all he says is we got this result

which i dont know how they got because when people were doing the pixle measurments at the announcement people were getting between 30-35% for the 2 graphs we got plus 33%~ for cyberpunk 2077 which we had a video for

also this graph is wrong in one place

the 1080ti while yes launched later in pascals life time is not the apropriate GPU for that graph

that would be the original Pascal Titan

which was the first of the pascal cards

launching some 24 months again before the 2080ti did

though i agree this generations uplift isnt huge despite the time wait for it

though 30% over a 4090 is still over 2x as fast as anything that isnt a 4090

on interesting thing is Blackwell 2.0 doesnt use a new node while Blackwell 1.0 does use a new node

i dont really get too much why nvidia went that direction

[cinder lagoon]

Hmm didn't think of Titans, valid point 🤔

[cinder lagoon]

I'm sure there is a matching estimate somewhere, I will be holding my seat until we find it tho xD

Lmao tho

[jade scaffold]

yea i saw that lmao

AMD somehow mentioned AI more than NVidia

[cinder lagoon]

Not difficult if you put it in every new product name haha

[jade scaffold]

this is true

Ryzen AI max

common amd

the AI NPU in the normal Ryzen AI models isnt even good

the Ryzen AI MAX only beats the 4090 on technicality in a benchmark where the 4090 was out of ram

like instead of taling about your APUs AI capability which it uses for near enough nothing

talk about the fucking AWESOME APU you made

like atleast when NVidia talk about AI its almost allways in a way thats usefull

[cinder lagoon]

And then the curfuffle with the chipset and gpu naming schemes!?

[jade scaffold]

DLSS, Earth 2, Robotics, AI Vision, Manufacturing, Warehousing, self driving cars, ECT

when AMD talk about AI its almost allways fucking co-pilot of chat gpt

[jade scaffold]

yea no idea whats going on there

the naming scheme of RX 6000 and 7000 was almost perfect IMO

then they fumble the laptop APU naming too

Intel arent doing any beter with CPU/APU naming tbh

the fuck is a 265k

why not just 230k 250k 270k and 290k

like that would have been fine

oh also

on the topic of AI

did you see the NVidia Mediatek mini super computer

i fucking love it and i want one

look at him

and you see thoes HUGE netoworking ports on the back?

thats a ConnectX chip fed pair of ports

you can connect multiple of these things together like you would server racks and they act as one big computer

[cinder lagoon]

They gon be so expensive tho ;(

[jade scaffold]

20 Core Neoverse CPU
with a pretty beefy gpu i believe this is blackwell 1.0 based too so its a node thats 20% more dense than the rtx 5000 blackwell 2.0

yea 3k for one

but it is for super professional developement of super computer based ai systems

you can get a NVidia jetson tho now for $250

[surreal moss]

I thought that was the Project Digit and starts at $3000

[cinder lagoon]

Thats actually pretty reasonable considering the specs

[jade scaffold]

I also LOVE the design of NVidias DGX type super computer racks and desktops

[jade scaffold]

yes it is

[cinder lagoon]

Link pls?

[surreal moss]

The Jetson comment confused me 😂

[jade scaffold]

https://uk.rs-online.com/web/p/processor-development-tools/2647384?nvid=em-945-13766-0005-000

Out of stock currently

but £250 for a Ampere based Jetson

kinda based

nvidia just went lets cut the price of the jetson by 2/3rds

and make it faster

[jade scaffold]

yea

i MIGHT get one

that being said NVidia are working on a windows chip

might get that instead

[cinder lagoon]

Thanks a bunch!!

[cinder lagoon]

What's that supposed to mean?

[jade scaffold]

a ARM chip for the Windows OS

probably gonna be the only chip that can compete with apples efficiency

mind you Intel Kept up pretty well laptop efficiency wise with apple

AMD need to get idle power consumption under control their chips are really bad at dissabling portions of them selfs something Intel Chips are actually very good at aparently

[cinder lagoon]

Is compat better now?
Technically any arm chip is windows if you're brave enough

[jade scaffold]

not sure how compatability is on windows side

but alot of the issues with the snapdragon X is on Snapdragons drivers and microcode

[cinder lagoon]

True
40W while doing zilch is crazy

[jade scaffold]

NVidia already have experience making GPUs for windows

[cinder lagoon]

Probably way lower on linux tho

[jade scaffold]

oh no AMDs idel powr is just bad

the chips just cant shut down parts of them like intels can

like if you look at the performance per w of intels lunar lake and amds Ryzen AI 9s they are pretty similar

infact amd actually pulls very slightly ahead

BUT

if you look at the battery life of the intel lunar lake vs AMDs ryzen AI 9

Intel Smokes them

like almost 3x smokes them

[cinder lagoon]

Well,,, except on mobile - right?

[jade scaffold]

it is genuinely crazy how intel despite having a less efficient architecture for raw performance do so amazingly at low power

no

lunar lake is a mobile chip

its mobile exclusive infact

if you hit the ryzen AI 9 with max load and the intel Lunar lake with max load it would lead you to assume the AMD chip is actually more efficient

but in reality far from it weirdly

like genuinely Intel lunarlake IMO is hands down the best thing intel have made in the past 2 decades

look at this shit

Wheres this in your desktop CPUs intel WHERE

it beats the AI 9 365 even though the laptop thats in has a 10% bigger battery

another thing to note is how apple beats amd here despite having a battery thats like 40% smaller

Apple truly are insane

And here even though the intel And AMD chip in these laptops are configured to pull the same max tdp the intel chip still wins

and why you might ask?

Intels Lunar Lake architecture isnt a CPU it isnt an APU its a full blown SOC

It has all the wifi, bluetooth, USB. PCIE, RAM, right on the Chip itself this is far far more efficient than splitting out all thoes parts onto wifi cards or an aditional chipset it also lets them run the ram at half the power and a higher speed than AMD can, Lunar Lake is truly just awesome as a laptop architecture and IMO it is the IDEAL handheld gaming chip atm.

And in full load benchmarks even though the intel chip seems to loose at the same power the intel chip isnt actually running at the same power because its running all that aditional stuff on chip which does count towards its power budget

[cinder lagoon]

Someone surpassing Apple is actually pretty big dang

[jade scaffold]

its also a pretty speedy lil dude at single core Apple surprisingly beating even intel here in single core

Lunar lake is actually why im exited for some of intels Xeons this year

This year intel are dropping a ALL Ecore 288 CORE CPU

intels E Cores in lunar lake are genuinely insane they are 2x as fast as last gens E cores near enough which is crazy

that makes them like nearly ryzen 5000 fast

at like 2w per core

and you can put 2 of these silly 288 core CPUs onto a single mobo

576 cores in a single computer

thats more CPU cores than some GPUs have GPU cores

wonder how that would do rendering a game

[cinder lagoon]

Wait the jetson are good for training but not for running, right?

[jade scaffold]

Other way round

They RUN like ai vision is what they are most used for

But they aren't bad for ai dev

Especially not now they are so cheap

[cinder lagoon]

Ohh but not great for LLMs right?

[jade scaffold]

I would assume not

Could run a home assistant tho Id think

[cinder lagoon]

Bruh

[jade scaffold]

It's got 16gb of ram

If your llm will run on that it'll run on the Jetson

[cinder lagoon]

Are they better than a 7600xt for LLMs tho?

[jade scaffold]

To get more ram than that on a GPU for ai tho you are paying a fair bit these days

[jade scaffold]

I would presume?

I don't actually know

AMDs ai acceleration is really bad

[cinder lagoon]

Uhhhh oh man this stuff actually kinda complicated

[jade scaffold]

AMD don't seem to list a AI TOPs number for the 7600xt

Yea no the Jetsons better for ai

67 on the Jetson (@30w) Vs 45 on the 7600xt

Keep in mind the Jetsons CPU isnt great

And it's ram is shared between the CPU and GPU

[cinder lagoon]

So it's better and more energy efficient for LLMs?

[mossy gull]

It's shit anyways

[cinder lagoon]

Aha thanks for the insight, will relay

[jade scaffold]

For small ones yea

If you wanna run big thiccccc boi llms

You can get like 4 NVidia P100 Tesla's for like 400 quid

Or V100s for not much more

[cinder lagoon]

Hmm also compelling

Socan you recommend anything for 150-300€ in germany?
Used would be great if ceaper

[jade scaffold]

AI stuff requires you know what you are doing

i would make sure who ever you are recomending stuff too actually makesure they have played with ai before

before dedicating money to actual AI hardware

but yea the Jetson is great

Its also basically a rasberry pi on crack

has GPIO and all

[cinder lagoon]

I see... he wants a HA setup with proper assistant so there's that...
Never done AI before afaik but we've heard it's pretty much plug and play on the open source side

Except of the usual IT pit falls of course

[sacred seal]

The current best open source model is llama 3.1 8b which can run on pretty well any gpu depending on how quantized it is

The main question would be how fast it is

My AI rig at home is an old sever with a p40 shoved in there. cost about 500cad.

Doing some quick reading a 7600xt would do about 10 tokens per second

[cerulean helm]

Honestly I have no idea, but I believe so. You should be able to connect each HA instance to a centeralized LLM server. So, in my case as I run a seperate PC for ollama its just a ip address that HA connects to ollama.

Okay, so hilariously, the HA voice assistant mic picks up everything when the wake word is said. So while even standing next to it and telling it to turn off a light, it hears the tv thats on as well. And at the time it was talking about the dangerously cold temps and all the snow we have and the VA after turning off the light began telling me to be careful shoveling all the snow outside.

[jade scaffold]

i believe the Jetson orin nano super should boddy the P40

though it doesnt have nearly as much VRAM

[sacred seal]

8gb of ram total is not great for llms. You would have to run a super quantized version and your context window would be tiny

I guess it depends on how much ram your os uses though

6gb of vram is generally seen as the lower limit

[cinder lagoon]

Oh, doesn't it also have an AI filter that tries to listen just to your voice?

[cerulean helm]

No, though I should be able to find either a community integration in HA or perhaps use the LLM and create a filter.

[sacred seal]

From my understanding the tts happens on the home assistant voice (though it could happen on the pi/sever), and after that it all raw text. Probably not going to be able to get an llm to figure out what is your voice or not based just on a transcript

[sonic meadow]

Only if you think about it in percentages. In absolute terms it is bigger than any but the 30-40 jump.

[sonic meadow]

oh god, I should have cloned the drive on to the ssd before doing anything else. lefit takes like five minutes to boot

It only took until an uptime of 18 minutes for the HDD to calm down

[jade scaffold]

It's got 16

Which still isn't ideal its certainly not 24gb

But Vs a 7600xt it should still beat it out

[jade scaffold]

Wait yea it's 8

Which one has 16

I know the NX does and that's the same chip as the nano super

That's my b

[sacred seal]

16 would be perfect for something like llama 3.1 8b

[jade scaffold]

Yea the orin NX has 16gb

I thought they upped the nannos ram to match the NX in the super variant

Apparently not

[jade scaffold]

Apparently it can run that

Not sure how well

Nvidia arent good at providing actual benchmarks and very few people make ai benchmarks

[sacred seal]

It would have to be quantized and small context window, but I am not surprised that it would run it.

[mossy gull]

Nvidia and good benchmarks.
That's like saying the 5070 is as fast as a 4090

[sacred seal]

The nice thing about self hosting AI is it is infinity customizable

Or rather there are tones of variants and a fair few customizations

[jade scaffold]

Honestly imo if they can get DLSS perfect and get reflex 2 working on ai rendered frames I'm willing to let the 5070 = 4090 thing slide because it would be really hard to actually tell the difference at that point

[mossy gull]

The only AI I want are the ones helping me with 3D art stuff😅

[mossy gull]

12GB vs 24GB, still a massive difference in what it actually can render at the same quality though.

[jade scaffold]

The main issue with DLSS is latency if reflex 2 works on a DLSS rendered frame then as far as you are concerned that's a 16ms latency

This is true

But if again DLSS becomes perfect

The RAM limits mostly vanish

[mossy gull]

There are games that render a lower resolution and quality texture when a GPU doesn't have enough VRAM just to keep the performance up.

Just because of this, DLSS won't just fix that as it would upscaled the lower quality texture instead of loading in a higher quality texture.

[jade scaffold]

Honestly this is why I'm excited for DLSSs future if they can get it damn near perfect you can use it to massively augment performance I mean 240+ FPS In cyberpunk is actually insane and DLSS is only gonna get better because they have a completely new model now which is gonna keep training all the way to the next gen and gen after and so on

[jade scaffold]

Yea which again isn't an issue if you are using DLSS unless your game is eating stoopid amounts of ram

Like if you are going from 1080p to 1440p and your game is eating 12gb of ram at that point that's a developer being a dumbass

But yes Nvidia should still put more ram on their cards not disputing that

[sacred seal]

Dlss is a dead end technology. We will most likely go to AI rendering before too long. Games are just wiflreframes and AI draws the details on in real time

[mossy gull]

Cyberpunk 2077, Flight Simulator 2024, Star Citizen, Dragon's Dogma 2😅

[sacred seal]

The AI doom demo is a neat proof of concept, and far more stable than the AI Minecraft

[jade scaffold]

At 1080p are fine on 12gb

At 1440p though it's cutting close

[mossy gull]

3440x1440, and soon 5120x2160 😅

On 4:4:4 Chroma Subsampling mind you

In YCbCr

10-bit

[jade scaffold]

Yea then you are having issues but got normies 12gb would be fine for DLSS upscaling

Thats if they can get it near perfect

Which I hope with this new model and time they can

[mossy gull]

I noticed that a lot sadly enough 😐

[jade scaffold]

Yea like 2560x1440 would be fine at 12gb

Unless you run mega cranked settings maybe

[mossy gull]

Dragon's Dogma 2 would like a word then, that game hogs VRAM like no tomorrow, even worse than SC

[jade scaffold]

If you are running cranked settings sure

If you run normal person settings it's not that bad

[sonic meadow]

You do realise that some games will attempt to use as much of your vram as is available, regardless of how much they might actually "need"?

[jade scaffold]

They don't really

But they Reserve the VRAM they think they'll need

[sonic meadow]

Same thing from the outside

[jade scaffold]

Or just reserve 90% of your vram even if they are only gonna use 10% because reasons

[sonic meadow]

Let's you cache more textures/etc

[mossy gull]

Except some games run worse when they get less VRAM.

For example Cyberpunk 2077 is a good example with this, 8GB simply has worse performance than 16GB.

[jade scaffold]

Yes but no

[sonic meadow]

Your "except" is.. kinda irrelevant to what I was saying

[jade scaffold]

Most vram usage is cached textures often though a game will reserve a tonne of vram and do nothing with it

[mossy gull]

https://youtu.be/T5mHQ3z6j2g

[jade scaffold]

I really don't have time to watch that I'm at work

But I presume he runs the game at 4k and goes muh vram not vramming

[mossy gull]

He compares a 3070 8GB vs a modded 3070 with 16GB

[jade scaffold]

Duh

[mossy gull]

He tests from what I know 1440p and 2160p

[jade scaffold]

Yea exactly...

If you run high settings good luck running any modern game at 4k with less than 16gb

But medium 1440p? Probably fine

[mossy gull]

This is an interesting 1 at 4K ultra settings

[jade scaffold]

Hell wh40k darktide is fine at 1440p medium with 12gb but is just a bit too much for 8gb which is why I think 8gb should be exclusive to low end 1080p cards like a 50 class card

[jade scaffold]

Again ultra setting which people aren't actually ever running

But yes it shows the effect vram compression has

The chances are though these games are right on the edge of running at 8gb fine if you had 12 you'd probs get the same performance also iirc the modded 3070 is actually running faster higher end chips too which will make a few FPS difference

[sonic meadow]

And some of those differences could be attributed to.. different chips. Not 3070v3060 kind of thing. But ones with different boost bins. Or different cooling (and by this I include stuff like thermal paste).
The last one is however a not-rounding error difference.

[jade scaffold]

Yea that last one is the base 3070 actually completely running out of vram

The others are just the result of vram compression it's not running out but it's getting closish to full

[mossy gull]

https://youtu.be/6V_NrvRTopY

The 16GB sometimes has lower fps, but it's still smoother because the higher 1% lows

[sacred seal]

Clicked through that video, most of the games showed a 1% differences

8gb of vram is probably going to add an extra 40$ to the cost of a card, even at the 60 series I don't really feel it

[sonic meadow]

and 1% is 0 when doing comparisons like that

[mossy gull]

1% lows do a lot to me, but then I prefer smooth fps over high unstable fps.

[sacred seal]

Yes, but is it worth a 10-15% cost increase?

[mossy gull]

Yes😐

[jade scaffold]

Like I agree more vram is nice but it's just not worth the additional cost

And the difference in .1% lows isn't huge huge

[sonic meadow]

there was a time when 10-15% would have got you the next card in the series

[mossy gull]

The 8GB still looks more stuttery to me though

[jade scaffold]

I miss them days

[sonic meadow]

it didnt help that the "series" consisted of about 20 cards

[mossy gull]

I miss getting the Vega FE for 800.-, a top of the line card, now top of the line is almost 3K.

[jade scaffold]

The vega Fe was not top of the line

It was up there but it was not at the top

[mossy gull]

It was back in the day😅

[sacred seal]

You are watching a variable frame rate recording on a fixed frame rate device. You cant really see stutters

[jade scaffold]

Enormous power consumption poor performance for the cost

Slow memory somehow

[sacred seal]

Anything you see is more an artifact of the recording process

[jade scaffold]

But it did atleast have alot of Memory

But also adjusted for inflation the vega FE would cost more than a 5080

[mossy gull]

I mean, nothing a bit of Conductonaut couldn't fix, it did improve the cooling and therfore the performance.
But eh, my last 1 died in the RX 6000 era of GPU

[jade scaffold]

No it's performance was still shit

It was still getting shit stomped by the 700us 1080ti in performance

Not as much vram but the 1080ti was faster and had massively faster ram and larger cache

[mossy gull]

I mean, Vega GPUs were shit to begin with, I primarily had it for its VRAM 😅

[jade scaffold]

The entire Vega series were all crap

Cool on paper actually just inferior in practice

[sonic meadow]

their schtick was HBM, and it wasnt enough of a gimmick to matter for the most part

[jade scaffold]

It didn't matter atall

[mossy gull]

Weren't they repurposed workstation chips?

[jade scaffold]

The vega 64/FEs hbm was slower than the 1080tis gddr5x

[sacred seal]

I would argue it actually mattered a lot, its just that the underlying GPUs were not good enough to make effective use of it.

[jade scaffold]

It wasn't good hbm though

It was cheap slow ass hbm

[sacred seal]

You put GDDR5 on a vega and it probably would have preformed even worse

[jade scaffold]

Gddr5x though

It would have probs done better

Hey maybe it could get within 20% of the 1080ti lmao

[sacred seal]

I think the fury was the first card to use HBM, and as I remember it it was sinificantly more powerful then the 390

Moreso then a typical generational uplift of the time

[mossy gull]

Fury cards were tiny ass cards, which was a massive advantage usage with HBM, however the Vegas were huge cards.

[jade scaffold]

The Vegas were normal sized atleast all the founder versions were

The fury was very small tho

[mossy gull]

I have the Vega FE LC, a watercooled brass ingot

2 of them in fact

1 died when the 30 series were released, the 2nd shortly after.

Probably shorted out because of the Conductonaut

[jade scaffold]

Yea they aren't big they are just a normal 2 slot card about the same size as every reference card ever at that point

Need to get a Vega Fe LC for my collection

[mossy gull]

Just know, they're brass coloured😅
So they won't fit in with anything

The air cooled is blue I believe

Which isn't much better

The non FE is aluminium coloured

[jade scaffold]

They look really nice imo

And you can totally make them fit on with a build

[mossy gull]

That they do, just a simple design makes them nice to look at.

[jade scaffold]

Want a collection of gold/brass cards

[mossy gull]

RTX Titan is I believe gold

[jade scaffold]

So I want a Titan V a GV100 a V100 a Titan RTX and a Vega Fe LC

A bunch of shiny Bois

[mossy gull]

Shiny collection

[jade scaffold]

It is indeed

[mossy gull]

I was once considering a collection of GPUs that simply don't fit in any build, but I gave up as it's harder than it sounds😅

Also I love the era of plastic heatsinks

[jade scaffold]

You can make gold and brass work

They work really well is very Matt black

A gold card would look fine in my pc I think as long as it's not a mega tacky gold

[mossy gull]

I know, but cards like the Vega FE AC are blue, which fit into almost nothing.

You can make it work, but barely

[jade scaffold]

Again Matt black

Or a build that uses brushed steel in silver would work with blue

[mossy gull]

Exactly, but if you don't know it barely fits.

Heck the hardest colour might be puke yellow

[jade scaffold]

You could make nearly any colour work with some effort imo

[mossy gull]

Puke yellow, maybe if you want to put in a lot of efford

Hear me out, what about if a gold Blackwell Titan existed?

Piano black and brushed gold

I know something better, carbon grey with brushed gold

[jade scaffold]

Some alleged 5090 3D mark tests

[mossy gull]

That's surprising, the 7900XTX scoring above the 4080 Super

[jade scaffold]

Yea it depends on workload and how a game or program handles rendering

[sonic meadow]

some trading like that isnt unexpected

even when one card might be "better"

just the nature of the differences of design between the two architectures

[jade scaffold]

Though nice to see the 5090 pulling atleast 30% in these benchmarks over the 4090

The lowest it seems to get here is 1.29x faster than the 4090

Though there's one result there where the 5090 manages 1.5x

But generally results seem to be hovering around 1.3-1.4x

[sonic meadow]

Arghhh. That chart has both 2160p and 4k

[jade scaffold]

It does also appear to be roughly depending on test 2x as fast as the 4080 so I'm pretty happy with that.

Should make the 5090 between 2.25 on the very low end and 2.5 on the high end times faster than my 3090

Both results are well over the "atleast 2x as fast as my current card" requirement I have for upgrading

[mossy gull]

Both are 2160p,
But 2160p is usually 3840x2160, 4K is 4096x2160 😅
Just joking aside, I think they jumbled them around

[jade scaffold]

No

It's just rounding

4k is still 3840x2160

[sonic meadow]

4k is both

[jade scaffold]

In the same way 1920x1080 is 2k

[sonic meadow]

though in the comon usage it refers to 3840x2160

[sonic meadow]

waddya mean, 2k is obviously 2560x1440. you can tell because neither of the numbers is close to 2000

[jade scaffold]

The 2k 4k numbers are dumb

It's basically just for marketing

[sonic meadow]

it theoretically refers to the number of columns in the picture.
remember that 2k originally came from film/movies, and was used because it was what was consistent regardless of aspect ratio.
film, tv, and computer all came up with their resolutions and names more or less independently, and then HD happened and they sorta started doing all the same numbers. kept their own names though

and then marketing took over, and we have "4k" being not strictly 4k when used in the consumer space

[jade scaffold]

Yea 4k is just a very approximately 4000 pixel wide screen

[mossy gull]

DCI uses 4K for 4096x2160

1K = 1024x768
2K = 2048x1080
4K = 4096x2160
8K = 8192x4320

[jade scaffold]

Yea kinda irrelevant to the fact 3880x2160 is still 4k

[mossy gull]

I know, just playing around :p

https://youtu.be/xsM_iAMybXM?si=hCytZjyRxmXP-R9O

[jade scaffold]

someones kinda cool concept for a Titan blackwell using the 4 slot design seen on the prototype 4090

[cinder lagoon]

Dont think adding another fan is gonna do much but for some reason those pictores have a weirdly american vibe

Like it's giving 5 pound cheeseburger

[jade scaffold]

have you not seen this cooler in action

like this prototype was built

at 450w its so quiet its inaudable

this cooler is an insane monstosity

at 600w it runs cooler and quieter than the normal 4090 FE does at 426w

like 25dba it just doesnt care its so insne

[cerulean helm]

Yes thats the basic function of HA voice. However I have moved over the STT and TTS over to the LLM server and off the HA Green box.

[jade scaffold]

Proper 5090 reviews incoming today I think

[sacred seal]

5090 reviews are out and they ain't great without AI. 4k it's about 25-30 percent faster, and at lower res it's only about 20

[jade scaffold]

20-50 depending on title

Gamers nexus averaged the performance increase to a little over 30%

[mossy gull]

https://youtu.be/La4EdRPT_Mg?si=SKVzZXyboAjBOQ9T

https://youtu.be/ulUZ7bf_MXI?si=PIMWxUFHMaMXHFLO

[jade scaffold]

thats uh fucking insane

[jade scaffold]

yea the new DLSS is kinda fucking wild too

but ill still need to try that myself to see if the visual quality is any good

no increase in latency

thats impressive

what Nvidia need to do is get Reflex 2 working on Frame gen

if they can do that the only way you would ever be able to tell frame gen from a normal frame is the quality of the image and at 240fps you dont get an awefull long time to admire the artifacts

[mossy gull]

https://youtube.com/shorts/tEVlXjrMT34

[tender plank]

There was a video card that had some kind of switch that would detect if the cable was fully plugged in and would trigger a hardware level warning if the computer was powered on with the cable not fully plugged in.

[mossy gull]

I think they all have that now

[tender plank]

So MSI created a fix for an issue that's already been fixed?

[jade scaffold]

yea the sense pins are shorter now

they dont make contact unless sufficiently plugged in so dont let the gpu turn on

[sonic meadow]

Oof that power increase (dlss/etc off)

[hot berry]

So I built my first pc last August but then did some cable management on it recently, I have a couple questions, the cpu cooler has a 3 pin header that says to plug it into either the four pin cpu fan or the four pin pump fan on the motherboard, which one do I do? Also after doing some cable management when I turn on the pc there is a white light that will come on in the pc but then turn off? Lastly i realized that a connection in the case has some bent pins (see the pic) what does that connection do and is it important? Thanks I know that was a lot

https://pcpartpicker.com/user/AptMantis2278/saved/#view=LfKfmG

[cinder lagoon]

Given that the Mobo labels it "PUMP_FAN" you should choose CPU_FAN for your AiO Pump.

Regarding the bent pins: The component is a Fan + RGB Splitter / Controller.
They are used for connecting an Adressable RGB Strip or other device so the lighting is in sync on the whole system

[cinder lagoon]

Ahhh crap anybody seen the rossman video?

Sadly very valid points.

Hate to see it

Have begrugdingly unsubbed and will wait for outcome

Actually had wanted to watch WAN show today, stopped after 5 mins and went to other sources...

[sacred seal]

Way more rambly then I would like, but that is just rossmans style.

Honestly nothing he said supprised me, though how tight Linus is with money supprises me somewhat

[cerulean helm]

The amount of money that is demanded for a sponsor spot kinda astounded me. I mean, 30k for an unboxing video? Come on.

Otherwise no nothing surprised me and I had stopped watching pretty much any LTT video awhile ago as it wasn't really informative and felt like like I was watching a sleazy used car dealership commercial.

I should add, I was completely unaware that the Honey extension had a clause about retaining all your data for 10 years and watching everything you do. That, that alone is skummy and why would LTT defend that sponsorship let alone have taken it.

[tender plank]

This one? https://youtu.be/0Udn7WNOrvQ

[sacred seal]

So wierd issue, my laptop does not turn off when I close the lid.
Ideas?

To clarify, when I say that it does not turn off, it just logs me out. I don't even think it goes to sleep

[surreal moss]

Does it have the same behavior with sleep mode instead of hibernate?

[sacred seal]

Yes

[surreal moss]

Nothing in the BIOS has changed?

[sacred seal]

This is a laptop with a stock bios

It took me awhile to realize the problem, I don't open and close this laptop a lot so I was always confused when I would take it out of my bag and it would be totally dead every time

[surreal moss]

Oh, that's probably something running on it that's keeping it awake

I had the same problem with my old work laptop

It wasn't anything OS related that we found. My Tech Lead had the exact same model and configuration but I often ran some software that he didn't

[sacred seal]

Closing the lid should override that though, no? If I manually put it into hybernate it works fine, it just wont do that when I close the lid

[surreal moss]

You would think. When I configured it the same as yours and pressed the power button it behaved the same as closing the lid. That is to say it stayed on.

[cinder lagoon]

So wait - it locks but doesn't do the actions specified here?
Can you set it to "energy save" / sleep instead of hibernate and check with Teamviewer or similar if it does that?

[cinder lagoon]

Could be a program overriding sleep. Try closing all background apps

[cinder lagoon]

I understand the price.
They only have so much time, have like 10 employees work on that video in some way and have a pretty high production quality.
Time communicating with the sponsor, agreeing requirements etc etc.
Also stuff like equipment insurance etc, lotta hidden costs.

[jade scaffold]

For 3 pin headers it doesn't matter as any 3 pin pump or fan will always run at 100% speed

[cinder lagoon]

What I really abode is the fact that they knowingly sent texts to an old number that would put GN in a bad light and literally, as Rossman put it, didn't measure themselves by their own yard stick.
"Right to reply my" ass if you gon say shit in cry-voice on Livestream

[cinder lagoon]

Nooo
It does matter:

3pin still does include a TACH signal, which is used to detect fan/pump failure.
Only CPU_FAN Will have that warning by default, some mobos have the option to configure the warning for the pump header

[jade scaffold]

One issue

An os won't pass you that warning

They never do

[cinder lagoon]

The uefi will prevent to boot with the warning

That's why it still matters

[jade scaffold]

Not by default

[cinder lagoon]

Never seen one that doesn't beep or at least needs you to confirm that you (now) know

[jade scaffold]

I've dealt with loads of dead pumps on AIOs it's like the only thing that ever breaks on them

And I've had pcs just thermal throttle without giving any warning

[cinder lagoon]

Where they plugged into the correct header and the warning not manually disabled tho?

[jade scaffold]

Mobos have no speakers these days

A mobo with an error screen would probs tell you but boot anyway

[jade scaffold]

Probably

The CPU and pump headers are closest to the pump usually the most convenient to plug into

Also also had a Ryzen stock cooler fan fail and I know for a fact that was plugged into the CPU header and it didn't tell me shit

[cinder lagoon]

I've noticed the behavior is model-dependant and many act differently upon it but even all prebuilts I've seen have needed you to confirm the error or not let you boot at all

[jade scaffold]

The only mobo I've had that's ever told me the CPU fan isn't plugged in or is having issues is my £700 super micro server board

Which regardless auto boots anyway

Like plug the CPU cooler into the CPU or pump header but I guarantee unless your mobo is old or something kinda of professional pc the only way you are finding out your pump is dead is when the CPU is thermal throttling

My ASrock server board does also tell me if it thinks something's wrong with the fan too

But I don't think it prevents boot

Never had a fan fail on it though only it tell me the fan I'm using is running at a low rpm and may be stalled

[mossy gull]

https://youtu.be/iSfBWJlTdR8

[sacred seal]

Exactly. And changing it to sleep does nothing. Also no background apps that I am aware of.

[cinder lagoon]

Hmm you could try booting a vhdx of a fresh Windows install and see if it fixes.
Easybcd has the ability to add one to the Windows boot menu (but watch out for fake sites)
The original site requires you to enter email

I could also share the program if you want tho

[cerulean helm]

Oh good, windows broke my network connections. I now can't access my mapped nas drive. Somehow windows explorer has broken.

[surreal moss]

Something reboot and using dhcp?

[cerulean helm]

No, its windows exploerer breaking I think. Its really really slow, and/or not responding. So I did a sfc scannow, it found some corrupted files and repaired. I rebooted, windows explorer opens (sometimes) and showed one of my 2 mapped drives unavailable. If I click on say the c drive and back out it shows all the mapped drives connected. But when I try and access it, I just get spinning wheel till itshows an x again.

Tried accessing truenas via the ip address and it won't connect. At least on the PC.

And well, can't access it on my phone on my network either.

Well shit, hooked up spare monitor to the NAs to do a rebbot via the linux menue for Truenas and got a very bad message:

boot-pool has encountered an uncorrectable I/O failure and has been suspended.

Seriously did I just randomly lose by nvme boot drive? wtf...

Still doesn't explain why windows explorer is being shit but just come on...

Reboot of the nas seems to have fixed the I/O issue.

[surreal moss]

It didn’t fix the windows issue?

[cerulean helm]

Not right away. But did a reboot and for whatever reason explorer is responsive now. I'm going to hazzard a guess that explorer was hanging up because it couldn't access the mapped drives?

[surreal moss]

That's what I've observed in the past. It's like it keeps retrying and doesn't have good logic to just give up and notify it failed

[jade scaffold]

Only 3 days till new GPUs yey

[mossy gull]

I am too poor🥲

Also, I got the 245K for my 2nd system, and the core layout is weird.
2P-8E-4P

I know this because it parks the P cores when in idle

[jade scaffold]

Yea it's odd

The thing is the e cores are positioned under the p cores acting as kinda their 2nd thread

But the 245k is just unusual

[mossy gull]

From what I know the 285k is 2P-4E-2P-4E-2P-4E-2P-4E.
But I could be wrong, but yeah, the 245K is weird as hell😅

I do prefer the Ultra series, even though they're weird, games have less problems when running in W10

[cinder lagoon]

Recently tried using a mapped network drive from powershell.
Did ls V:\
"Path does not exist"

Open terminal from explorer on the drive
Displays V:\ on the command line

Type ls ./
Displays all files

So might just be network drives being broken since recently

[mossy gull]

That's an even wackier layout 😅

[jade scaffold]

That's the 285k

The blank spot is a empty tile that might have been used for an NPU if they wanted to add one to it

The layout isn't that weird tbh

You want weird look at lunar lake

The e cores and p cores aren't even connected lmao

But this is intentional because it allows them to totally kill power to part of the chip if it's not being used

Love the funky die arrangement of the 285k

[mossy gull]

It's still weird, why have the P-cores on top of the E-cores

[jade scaffold]

Yea it makes sense though

Each p core is meant to have its own little e cores

[mossy gull]

I am gonna test it with some games, and see how it fares against the 5800X

[jade scaffold]

Should be rather substantially faster than the 5800x

It should fall between the 9700x and 9800x3D a bit closer to the 9700x

Also be sure to overclock the cache and run it with some speeeedy ram as it scales really really well off that

[mossy gull]

The reason I went with the 245K is because I had some spare 5600MT/s XMP RAM laying around from an older build😅
It's also on a B860 board from Gigabyte.

It's in Star Citizen 15 fps higher than the 5800X

It ain't fast RAM, but currently does the job

[mossy gull]

And because I got the 245K for only 250.-

[jade scaffold]

thats pretty good price

[mossy gull]

It really is, especially because it's multi core is pretty fast and Windows 10 actually works amazing with it.

Like every game only played on the P-cores even though the scheduler isn't made for a big.LITTLE.

I am happy it works really well,

And I am happy all the SSD slots are toolles, and that for 250.- as well.

I am happy with this little upgrade.

[cerulean helm]

Oh fffffffff. Got home. Weekly backup to Truenas failed. Go chech the nas. Hook up monitor, its frozen, no response. Great. Cycle power, and looks like the boot drive is gone... In grub rescue atm.

In the bios for the system as it won't even boot into grub rescue now. And I can't see the nvme drive. So time to tear down the nas...

[cerulean helm]

Well, I was going to report back that I installed a spare 2.5 ssd and had done a zfs pool recovery but....

In the process of pulling the now dead nvme and installing an add in card for more sata plugs something has decided to short out somewhere. The psu immediately turns off as soon as I power up the system. So, no idea what the hell I managed to do here...

[sacred seal]

That's scary

[cerulean helm]

Uh yeah, as I have no clue wtf is shorting out. I at least assume its a short. Tipped the 4u chassis over to make sure there weren't any loose objects or something. Found nothing. Moved cables around, made sure everything is seated. Same issue. PSU immediantly turns off.

Ive got it poered off and unplugged. Going to let it sit for 20 min or so, maybe OCP kicked on or something. Its a Seasonic psu thats a year old, I doubt that just decided to kick the bucket.

[sacred seal]

Might be worth a check. Just unplug it and short out the start pins to see if it craps out without any load

[cerulean helm]

Well, I feel like an idiot. And thankful the psu did its job. The box of cables for the Seasonic psu had mixed cables in the box. To what I assume is my Evga psu since thats the only option. The sata cable I installed has to have been for the wrong psu, the cables look identical. As soon as I pulled that cable the nas started up.

[sacred seal]

That is a relief

Why psu cable pinnouts aren't standardized I will never know

[cerulean helm]

Yeah that be nice.

Also, the pcie sata card doesn't registor in the bios as I can't see the micron ssd thats installed. So, no go with that. I have no more sata ports available. Going to have to order a new nvme drive.

I do have a couple u.2 drives with pcie cards... Maybe that will work.

That worked. The intel enterprise ssd is listed in the bios.

[sacred seal]

Finally something is going right

[cerulean helm]

Yeah, finally. Now I "just" need to reinstall truenas and import my data pool and hope nothing goes wrong with that...

[sacred seal]

Just wasted a bunch of time. Apparently github urls are not case sensitive, but if you use it as a web host then they are...

Spent way too much time trying to figure out why there was nothing at the url

[cerulean helm]

Well, I have Trunas up and running again. And managed to reimport the pools. So far thats great. Now just trying to setup smb shares and get the drives mapped again.

[jade scaffold]

I am hoping for that 4e 4p core i3 want to make a suuuper mini pc with one I'm.hoping it'll have the full size IGPU and be a k part

[mossy gull]

You mean the U3?

[jade scaffold]

Yes

205/215 is apparently what the core ultra 3s will be called

[mossy gull]

Even they they ain't on top of the gaming charts, these Core Ultra look really promising

[jade scaffold]

They are a good starting point for intel to make improvements from