hardware

[median marsh]

i think 3060 is 170W

[wise elbow]

raw gaming performance gains have been minimal

dlss4 does run a lot better on 50 series though afaik

[orchid iron]

that's mostly because if nvidia makes something better, then they charge more for it, so price to performance hasn't improved much gen over gen because of that, they want bigger margins

[somber surge]

Runs basically the same on RTX 40

And DLSS Multi Frame Gen does not work well on 8GB gpus...so pretty much half of their lineup

• it's borderline on 12GB gpus

[wise elbow]

damn shame too because mfg actually looks really nice

[somber surge]

It's very situational though

You need a very good base framerate

Like 70fps+

And a high refresh monitor or it is just wasted

[wise elbow]

it's really for high refresh displays like oleds

pretty lame they market it as just as fps increaser

[somber surge]

Yeah that's the big problem

[wise elbow]

imo it's pretty ok if you even have a frame rate of like 30

[somber surge]

The tech itself is great even if not always usable in all scenarios, but their marketing is just insanely missleading with that shit everywhere

[wise elbow]

nvidia is just being a shit

[somber surge]

It's not, unless you're really really not sensitive to latency + playing on controller

And on a game that has a decent base latency

[median marsh]

may need more dedicated cores which they may have added to 50xx die's?!

[sleek grove]

Gonna do a clean install, should have done it ages ago tbh. The question is do I move to Win11 or stick with 10

[orchid iron]

Either stick with windows 10 or look into ways to debloat windows 11, there are also cheap enterprise windows 11 or 10 keys that could be quite interesting for you

[sleek grove]

I just went with win 11 since win 10 has only 4 months left support anyways

[orchid iron]

Don't go with the vanilla windows 11 though please

[sleek grove]

Win 11 Pro

I guess I have to run a few scripts to debloat

[orchid iron]

Good luck with that

[sleek grove]

idk so much, heard a lot of bad stuff but the biggest difference so far is the UI and a copilot shortcut

[orchid iron]

And significant performance reduction, along with sometimes awful drivers

[orchid iron]

I got copilot on my debugging windows 10 install somehow...the pc is 10 years old, wtf Microsoft

I found out today that I can use multiple keyboard

Finally, I was being bottleneck by my singular keyboard, time to make a multi-keyboard setup

[wise elbow]

https://tenor.com/view/ghost-in-the-shell-keyboard-gif-7519694

[delicate crystal]

Anyone know anything about pny cards? Thry seem really slow but they demand a premium

[ashen stream]

Debloating W11 in 2025, is pretty easy to do.

[orchid iron]

They used to be bad, then good, then with the 50 series not great

[wise elbow]

they're as good as anything else

card factory oc is meaningless

and from my experience they are the cheapest cards, though they're made in the US which is probably why

imo I trust them more than other manufacturers. have used dozens of cards manufactured by them in the past and never had any issues

[orchid iron]

EVGA and XFX are probably look at the 50 series launch and thinking to themselves, "that's a lotta stress I avoided right there"

https://m.youtube.com/watch?v=rqsZc1e8cAI

[delicate crystal]

Hmm ok. Been looking at options for plus 16gb nvidia cards and the aelection is limited.

[wise elbow]

it's really only the 4090, 5090, and high end workstation cards

[fallen oasis]

Idk. I have a 24gb 3090 card.

[wise elbow]

yeah though I wouldn't really buy a 3090 in 2025

[fallen oasis]

Depends if you want vram or $$

[median marsh]

i would buy it, but not for the stupid amount of 1500euro which some dealers here ask for it -_-

4090s are even listed more expensive than 5090s

[agile lodge]

LC versions are nice

[ashen stream]

best app for spring cleaning

[ashen stream]

thank goodness we never needed them for good support

Mfs give us like a new mob a year on vote, while some 11 year old drops an entire species pack, for like a decade now.

[wise elbow]

windirstat is amazing

[mighty patrol]

Space Sniffer good too

[delicate crystal]

Can only find 4090s on used market now, and who knows ifnits bed used as a mining rig

[median marsh]

yea at that price range you likely want a new card with warranty

[delicate crystal]

Indeed. Its basocally pegged me.into doing a 7900 xtx for mow

[fallen oasis]

Its basocally pegged me.
I see

[wise elbow]

hmm

plinko, hopefully?

[orchid iron]

@delicate crystal btw, did you fix your system's issue?

[orchid iron]

So true

[mighty patrol]

https://www.youtube.com/watch?v=FA6BCIWPJ30

[orchid iron]

All that for strawberries

[mighty patrol]

That is just the start

[orchid iron]

I know, but if they advertised crowing animal flesh for us to consume without the need to make animals, that'd be way more edgy

[mighty patrol]

Next real application is fungi I'm sure

[orchid iron]

Gotta start small

[delicate crystal]

Ya i was fprtunate that tge data drives were stable, just windows being windows. The drives that had windows on it for some reason werent set to raid 1 and being samsung evo 970s, finally shat themselves. I might have bbqed one ofnthem a while back so that explained a few things

Basically my plan for a syatem upgrade has been mpved up by a few months. Was waiting until september but now itll be july

Heres what I have so far

System specs

CPU: AMD Threadripper 7960x
Cooler: ???, probably Noctua something
Memory: Crucial 128GB 2x64 5600 DDR5
Motherboard: Asus Pro WS TRX-50 SAGE
Video Card XFX Merc 310 RX 7900 XTX 24gb
OS/App/Active Storage Drive: Samsung 9100 Pro w/ heatsink
Backup Drive: WD HGST UltraStar 12gb 512mb cache
Power Supply: Corsair HX1200i 80+ platnium
Case: Fractal design north XL mesh

Was going to switch ovdr to a 4080 or 4090 but theyre impossible to find here that havent bsen used in a mining farm.

[median marsh]

PSU seems a bit overkill, but gives you headroom if you decide to change the 7900 for a nvidia card 😄

and for cooler i would pick a noctua "clone", saves a bunch of bucks

[orchid iron]

If you plan on going with threadripper, look at Arctic too, they provide very strong competition against noctua in threadripper department

[orchid iron]

Also, you'll literally have better GPUs come out few months after you build that system, are you ready to go through the hassle of reselling that GPU in case you really want newer GPUs?

[delicate crystal]

Well roght now my syatem is being held togrther by hopes and dreams so sadly dont have a choice. That was the original plan was to wait until then. Maybe get something cheaper that osnt a beast, then upgrade later on? Thats what thos system is for the most part, if things go well, i can easily upgrade most.of it

[delicate crystal]

Thats the plan. Upgrade the vid card, memory, etc down the road if needed

[median marsh]

??? which GPUs are announced for that date?

i think we wont hear much until end of 2026 from nvidia/amd

[delicate crystal]

I thought it was the last quarter of 2025?

[median marsh]

both just released a new series, so i doubt it, especially for the high end market

maybe they do some AI cards and whatnot, but nothing that should affect gamedev too much

3090 September 17, 2020
4090 October 12, 2022
5090 January 30, 2025

so there has always been a 2years + gap

and i guess nvidia will likely switch to a whole new architecture for next gen, so that gonna take more development/testing time than what they did the last years 😄

[delicate crystal]

Wish they had a 24 gb version for thr 50xx series. All i found is 16s and one 32

[median marsh]

they do

for the chinese market 😄

https://videocardz.com/newz/geforce-rtx-5090-double-d-24gb-graphics-card-for-china-set-to-launch-in-august

[orchid iron]

https://m.youtube.com/watch?v=J2OuhP1ehbQ

6:30 BIG

[orchid iron]

Nvidia super series and AMD GPUs with GDDR7
No "new generations" but at leasts current one gets models with more Vram

And I think intel will want to eventually release B60 for consumers, maybe?

[median marsh]

so there's a chance that they will make a 5080 super with 24GB?

[orchid iron]

Oh yeah, partner leaks show they are ready for them

[delicate crystal]

That wpuld ve nicd

[orchid iron]

And overclocked 9070 XT with 32gb of Vram because why not xD

[median marsh]

ok, then i take back my statement from earlier

didn't expect a 5080 24GB to happen

[orchid iron]

Remember how they tried to "unlaunch" the 4080 12gb?

[delicate crystal]

I wonder what theyd retail for up here

[orchid iron]

No clue, prices have been wild lately, hopefully AI farms don't require them

[delicate crystal]

Hopefully. That added with the added charge that is everything sold in canada makes it fun

[orchid iron]

Also, I did a bit of research and if the AI's data has to be stored in ram, the performance of token output is nerfed so hard, that buying those insanely expensive 96gb Vram models actually make sense if companies really want several billion parameters models

[median marsh]

at least doing it the american way 😛

deepseek showed that necessity drives development

[orchid iron]

Yeah, I was very happy when I saw deepseek making Nvidia lose 17% stock price in 1 day

[delicate crystal]

Had a coworker lose 50k because of that

[orchid iron]

Skill issue, did he keep the stocks? Because they went back up after

... unfortunately

[delicate crystal]

Ya he did. He was a decent trader so he bought a ton more when it droppe

[orchid iron]

Smart

[delicate crystal]

Indeed. Wiah i had followed suit. I thought there stock woukd pulk a blavkberry from it

[orchid iron]

I wish too

[orchid iron]

@grave sundial an electrician checked grounding for plugs, and there was 3 plugs (2 of which are used by high power consumption appliances) and 1 available plug directly from the fuse box, do you think it is safe compared to the other ones?
And if it is, do you recommend a UPS since that grid is from the 80s to my knowledge

[median marsh]

couldn't he tell you?

plugin a lamp and turn off fuses and check if it's fused

then compare fuse type to the other ones

sometimes there are different ones, as there are slow and quick fuses and for different loads

[delicate crystal]

Trying to find a video card tgat handles multiple thibgs is a bloody pain

Tge 5090 woukd mert all, but 4k for a card is nuts

Tge pro cards arent great at gaming alledgly. Amd cards are weak on ray tracing. Anything under 24 gb of vram isnt great for photogrammy.

Half debating doing the rtx pro 4000 and just doing games on low settings :p

[grave sundial]

I always recommend a stabilizer and UPS for expensive appliances. One errant voltage fluctuation is all you need to be spending tons of money buying new components.

We get extension cords with fuse breakers in them, so that the cord itself breaks the circuit in case of a short circuit.

It's always good to have an intermediary between your mains and your devices.

[orchid iron]

Amd GPUs don't have dedicated ray tracing cores, however that means on non-raytracing workloads, they'll out perform their Nvidia counter parts because more silicon is dedicated to raster

Although rdna 4 showed AMD can do raytracing well finally
Udna will likely match 50 series raytracing performance since there wasn't much uplift in that department from 40 series to 50

[orchid iron]

So an extension cord with fuses and UPS
But isn't the computer's power supply 's job to stabilise voltage?

[grave sundial]

How much does a PSU cost and how much an extension cord?

It's a question of 💰 and how much you want to spend replacing busted hardware.

[orchid iron]

True

[median marsh]

also UPS usually have fuses, actually most likely all of them

[orchid iron]

Oooh yeah, I'll check the specs

[orchid iron]

And what would be the specifications of the power cord since I'll plug a 4k Watt UPS on it for a 1.6k watt PSU, for a pc that consumes probably 350 watts at most

[median marsh]

so it can run 10h on the UPS 😄

but most likely the 4k is what it can produce, not what it can store

rather get a trusted brand than some company which tries to bait with high numbers

APC/BestPower/...

[grave sundial]

Heavy-duty extension cords are cheap to buy, at least in India. You don't have to worry about the amperes. The concern is whether the large plugs fit. Usually, the extensions with fuses are meant for large plugs and heavy-duty stuff, so you should be okay.

Take a photo of your PSU's plug and you can compare it to the what the extension cord has, when you visit the store.

I had purchased this one. Goldmedal, made in India.

Saved my entire computer when the modem's adapter shorted out, once.

The fuse tripped, there was a small puff of smoke from the modem's adapter. The entire computer shut down abruptly, but safely.

All the hardware was saved. I only had to replace the inverter's fuse, because the damn thing was too old anyway by that time. 😄

This costs 8 EUR, here. Probably 50-60 over there.

[orchid iron]

Oooh, extension with surge protection, yeah there are a bunch ranging from 8€ to 20€ but they don't say if they use fuses most of the time

[grave sundial]

If it's got an automatic tripper, it's got a fuse in there, methinks.

[orchid iron]

Man, what a mess
While looking for 4k ups I realised most of them take 12-24 volt input (car battery generally) and output 230 volts
So then while searching for ones that actually have 230volt input and output, most of them are designed for server racks and cost 4k € xD
I've also seen that some UPS can handle an overcurrent double their wattage for a few consecutive seconds
Do you think it is a good idea to lower the wattage required from the UPS down to like 2000 watts or less ? Since I'm only finding server stuff

[grave sundial]

a 2kW UPS is useless for a 1.6kW PSU.

Except for giving you 5 mins to shut it down when you lose power.

Why do you need a 1.6kW PSU btw?

[orchid iron]

Yeah, I think I'll downgrade to a 500 watt psu

[grave sundial]

What GPU does your new system have?

[orchid iron]

7900 xtx

[grave sundial]

355W + CPU 125W?

Uh... you could go for a 700W PSU, I think.

Then a 1.4kW UPS will be fine.

[orchid iron]

Nvm, forgot tdp is 355w

[orchid iron]

Yeah

[grave sundial]

Can you exchange the 1.6kW PSU? I imagine it cost a bomb.

[orchid iron]

Idk, I opened it to check everything was in there when I received it, and i got it several months ago, so at this point maybe I'll look into selling it on eBay xD

[mighty patrol]

https://www.youtube.com/watch?v=LZa2dvscRdY

[wise elbow]

you don't need to downgrade

what PSU you use doesn't change how much power your PC consumes by a meaningful amount

just figure out how much power your PC actually uses and get a UPS with a rated steady state power of at least that much

you 100% need a UPS of equal or greater power to your true power consumption though or it'll probably trip

also pretty much regardless of what UPS you get (unless you get a massive one) it won't have enough stored energy to keep your PC running full blast for all that long

they're mostly just good for intermittent power failures and to give you time to shutdown when it's long term

[orchid iron]

I thought it was based on PSU wattage, not max pc power consumption

[wise elbow]

nope

a UPS power rating is the max amount of power it can output for an extended amount of time without shutting off

this is the same as your PSU. it's rated for a max power. anything above that max power and your PSU should shut off so it doesn't melt

your PC will draw as much power as it needs, limited by your components

also, make sure you account for monitors

most consume absurd amounts of power

i have a 4k ips and it consumes up to 210 watts

[orchid iron]

Oh, so a 2kw UPS is fine with a 1.6kw PSU as long as all components can't draw more than 1kw?

[wise elbow]

yes

a 2k UPS would be fine regardless even if you drew your max 1.6kW

you can probably get a smaller UPS

[orchid iron]

Awesome, this thing costs almost as much as a mid GPU, would of felt bad for underselling it

[wise elbow]

yeah they can get expensive

you have to replace the cells in them every few years as well im pretty sure

[orchid iron]

The cells?

Oh ups batteries, got it

Thank you

[agile lodge]

https://youtu.be/NO2QG_B5onc?si=p5EyfJEjfuH5dbNl

[delicate crystal]

Is this real.or one of those clickbait things

[median marsh]

jaytwocents, so most likely close to real 😄

[mighty patrol]

Better be

Or someones gotta give him the AI slop tuber talk.

[delicate crystal]

Guess i wont be buying a 5090 for a while 😄

[orchid iron]

Custom BIOS to make it accept 1000w

[grave sundial]

And after you achieve all this, you'll use that card to do nothing but generate a meme.

[mighty patrol]

Better just buying an Ada gpu

[orchid iron]

I'd want a Zeus GPU, too bad they won't come until 2027 at least

[mighty patrol]

Real Time path tracing

What's that going to cost 120k

[grave sundial]

It'll be the price of a thunder bolt.

[orchid iron]

I'm hoping if there is a GPU company out there that can put ram slots on their GPU, hopefully AMD will follow

[mighty patrol]

Ram slots and nvme slots

[orchid iron]

https://m.youtube.com/watch?v=u_Lxkt50xOg&pp=ygUJcGV3ZGllcGll

[orchid iron]

Nvme slots are fake, they don't exist

[gleaming stone]

hey all, im taking a jump and buying a 50series GPU, upgrading from my 6900xt. i have the opportunity to get either a 5080 now, or wait for stock on a 5090FE (not willing to pay the £600 premium for a gpu cooler and some AGB). obviously more Vram is gonna be better for UE5, but is waiting for the 5090FE to come back in stock worth it?

[wise elbow]

youll have to compete with bots to buy a 5090fe , assuming they ever actually come back in stock

5080 and 5090 also aren't even comparable. 5090 is almost 2x the performance iirc

decide what games and what resolution and if you want pathtracing and buy based on that

IMO a 5090 is overkill for any games

[gleaming stone]

but is it overkill for UE5?

[wise elbow]

you don't need that good of a gpu for unreal

unless you want to do unfathomably complex scenes. but be realistic with your use case

[gleaming stone]

well the project im working right now is about 14gb vram usage, and i get about 55fps in engine with my 6900xt

[wise elbow]

is 55 fps in the editor even bad?

[gleaming stone]

not really no. obviously there is more i want to add into the game but my knowledge lacks on where its stuff that would affect vram usage. things like windows with RT reflections etc

if stuff like that, or parallax snow doesnt really affect vram usage then 16gb should be fine right?

[wise elbow]

i feel like anything but a 5090 would be a really marginal upgrade coming from a 6900xt

[gleaming stone]

oh really? my friend has a 5080 but doesn't use unreal so we only have games to go on. but in games its like night and day. path traced indy jones game on 5080= 90 fps, on 6900xt = 7fps

[wise elbow]

that's because the upscaling and raytracing on 6900xt isn't great

but afaik RT in unreal doesn't even use rt hardware

and you won't be upscaling (nor should you design your game to require it) in editor

[fallen oasis]

Don't forget the fake frames.

[wise elbow]

yeah that too

like don't get me wrong the 5080 is a good card (I have one myself) it's just that the 6900xt is still really good, has the same amount of memory, and any upgrade is going to be at LEAST 1000 dollars for like, 30-50 % better raw performance

[gleaming stone]

so, consensus is 5090 or stick to 6900xt?

[wise elbow]

imo yes

there are rumored 5080 supers coming out this year that will have extra memory

[delicate crystal]

That bad? (Was looking at a pro 4000)

[orchid iron]

I'd recommend waiting to see the rumoured 50 super series, when those get released you can either buy them or watch as the price of non-super drop

[gleaming stone]

so i took the plunge and ordered one of these https://www.overclockers.co.uk/msi-geforce-rtx-5090-ventus-3x-oc-32gb-gddr7-pci-express-graphics-card-gra-msi-04859.html

[mighty cosmos]

A 5080 would be like 70-80% faster in raw performance than the 6900xt, 6900xt is a little bit slower than the 3090

[orchid iron]

@grave sundial is that why you said to never turn on EVO mode on UPS?

[grave sundial]

Yeah. Eco mode works really on anything but something that is a computer. 😄

Mainly because the bypass time is increased by a few milliseconds.

[orchid iron]

Congratulations, now also buy a 100$ load balancing checking device that you plug in between your PSU and your 5090 otherwise it may just melt and get damaged because Nvidia engineers wanted to justify their salary

[grave sundial]

WHAT?! We don't love 🔥 anymore?

[orchid iron]

True, free heater included

Don't buy that 100$ thing

[orchid iron]

Jeez, binging UPS videos at X2 speed for over an hour is quite something let me tell ya

[grave sundial]

At least you know you have a higher attention span than the modern Joe.

[orchid iron]

Idk, after 1 hour of full focus, my focus becomes so fuzzy and I'm almost forced into taking a nap

[grave sundial]

Which is why God gave us coffee, snortable drugs, and supplements.

[orchid iron]

And then there are a few who are coffee insensitive like me, and naps are basically only option

[grave sundial]

🥲

[orchid iron]

@grave sundial btw, some UPS have in warranty that using it with extra surge protection extension voids warranty, so check yours just in case

[grave sundial]

Thanks. The tech never breaks within the warranty period though. I believe they are robustly designed that way, so that they break outside of warranty and you have to pay the full price to get it fixed/replaced. 😄

[orchid iron]

True, but just in case, if they ask what it was connected to, don't mention the extra surge protectors

[wise elbow]

they would also probably never know unless you tell them

[grave sundial]

Oh, they know already. It's common procedure in India, because voltage fluctuations are common. In fact, I believe they aren't allowed to deny service because of that, here.

Of course, ACT OF GOD is still off the table.

[orchid iron]

Nice

https://m.youtube.com/watch?v=ZQX4FLrm9ac

Shame they started soldering things like virgins

[wise elbow]

did they?

I have one that's like 6 years old now and it's mostly sockets

i think their thinner ones are soldered

[orchid iron]

yeah, the change was recent

[delicate crystal]

Remember the laat thinkpad i had was like rocketship compared to the dell i had at the time. Rock solid too. Im sure thst it was running linux while dell was on windows didnt factor into it

[agile lodge]

seems more nvidia cards coming with more vram

[orchid iron]

Yup

[agile lodge]

https://youtu.be/61zTd3xctgE?si=9h7l1VtfK1riZm97

[mighty patrol]

What lol

Already

[somber surge]

Pretty much releasing the "real" new generation, how RTX 5000 should have been to begin with

Seems like their new way to go: release something half broken and obsolete for half of the lineup, then fix it with "Super" cards 6-12 months later

The Super "refresh" is just there to fix an obscene lack of VRAM or an absurd price nowadays

[median marsh]

5070 TI Super seems nice tho

350W and 24GB

[delicate crystal]

Sounds up.my.alley. hopefully its not a long wsit

[orchid iron]

If the 24gb models cost way more than the 7900xtx or AMD 's "super" 9070, then it is just going to be terrible value (to be fair, existing 50 series card already feel like terrible value)

[grave sundial]

What prices are these obviously-very-cheap-and-totally-not-flammable new GPUs from Nvidia?

[orchid iron]

I wanna go ahead and guess, +300$ over market price of normal parts, what do you think?

[grave sundial]

Oh. I was thinking more $5,000 per card, based on modern business practices.

[orchid iron]

Oh yeah, if they change the "super" trademark to "ai" I could see the Nvidia Rtx 5080 AI 24gb selling for 5k$

[grave sundial]

If this is true, the 5070S receives the largest buffs?

I guess the remaining party members received the second-hand buffs.

[orchid iron]

I remember how Nvidia mentioned to marketing that gamers only need 70 class cards, and everything above (80 and 90) are for professionals

[grave sundial]

Hehe. I wonder if that's because of those 7/70 million orders they received from the cloud companies, or something.

'Hey, hey, you don't need these cool cards. Those higher paying, wealthier companies need them!'

[orchid iron]

I misread and saw "those higher paying weather companies needs them! " XD

[grave sundial]

They probably do, too.

[orchid iron]

Yeah, they need AI for more accurate weather forcast

[grave sundial]

This makes me question the reason for the 5080's existence.

[orchid iron]

Yeah, I forgot if a 4080 12gb even exists tbh

[orchid iron]

Didn't super models have a slight increase in cuda cores over their normal variant?

[grave sundial]

Pretty much.

But here, they just add a bit more memory to the other models and draw more power.

I bet there's more than that that's not leaked yet, but still....

[orchid iron]

More content for the disappointment T-shirt wall of 2025

[grave sundial]

Hehe. I still remember that ASUS RTX-RX mixup from the other day.

I wonder if NVidia will 'accidentally' ship a buffed XT as an RTX super. 😄

[orchid iron]

I remember when things didn't upset hardware unbox much more than standing up, ever since 2025 he has been standing on roofs and on over 30 floors of height

[grave sundial]

https://videocardz.com/newz/nvidias-next-major-gpu-driver-branch-to-drop-support-for-geforce-gtx-700-900-and-10-series

Wokay! Important stuff to know.

If you are using GTX (or lower), basically don't update your drivers beyond 580 version.

[orchid iron]

That's me!

Heck ya, my drivers are 100% man made

[mighty patrol]

For now

Who's to say later they don't just get an AI to re-write the entire thing for an end of life update

[orchid iron]

Nah, ai has to find flaws in their drivers first, which is tough because their drivers are already perfect

[mighty patrol]

Soon ™

[agile lodge]

🪛

[wise elbow]

they'll still make security updates

[muted holly]

Looking for help / recommendations. I have been tasked with creating a desktop machine / gaming machine for unreal and running simulations. We have a budget of 10-20k US dollars

Any links or recommendations ?

[orchid iron]

https://m.youtube.com/watch?v=GpGOfidf9H4#

Oh wow looks great from the thumbnail
Oh he uses glue like a virgin

Wait, wtf at those "case feets" oh gosh no

[orchid iron]

How many GPUs and how many CPU cores and ram would you aim for?

Usually you'd have an idea of the workloads the machines should be able to do

[orchid iron]

If it is a gaming machine, what games on what settings would it run?
If it runs simulations, what kind? There are some that have been optimised for GPU, but quite a few also use the CPU

Budget is 10-20k but even if you spend it on high end threadripper stuff, there are tasks it won't be able to do as well as a cheaper build like latency is one of the draw backs (which generally doesn't matter for content creation)

[muted holly]

Two - four instances of unreal running simultaneously

[orchid iron]

https://m.youtube.com/watch?v=6v7Lv6SoJQE&pp=ygUQYXZhIDQgNTA5MCBidWlsZA%3D%3D

For professional workload, get something similar to this

For high end gaming performance, build something else, because gaming wise the CPU will be bottlenecking the GPUs in many scenarios (these CPUs have high latency, so FPS will be lower than a 9950x3d, but in terms of multithreading performance, the 7975WX just crushes it )

[muted holly]

Awesome. Appreciate the help I’ll dive in

[orchid iron]

Also, when using GPUs that use 400 watts underfull load and only has 1 single 12volt high power cable, get something like this:
https://www.thermal-grizzly.com/en/wireview-gpu-pro/s-tg-wv-p
Because due to the lack of load balancing enforced by the 12 volt high power cable standard, the cable may draw excessive power through 1 wire thus melting it and causing damage to both PSU, GPU and itself
The item I linked provides an alarm sound if something goes wrong like excessive temperature that would cause melting or simply the cable not being plugged in correctly

[muted holly]

Man so much has changed since I built my last pc

[median marsh]

also get fire insurance

i would guess that the newer cards have protection against that issue onboard

some do have warning LEDs, not sure why they didn't implement a solid detection on the GPU in the first place... but here we are

[muted holly]

i just called them and requested a quote on that

[orchid iron]

Good Luck with that, it is an issue with is quite censored by more official companies, since if they were to accept being quoted for that, they'll face the consequences and might get reduced or /and delayed shipments

https://m.youtube.com/watch?v=Ndmoi1s0ZaY

https://m.youtube.com/watch?v=oB75fEt7tH0

[mighty cosmos]

the design is flawed but the issue is a bit overblown
People are acting like all 5090s are burning up, the reality is that it's a very very small number

[orchid iron]

Yup, and for a product over 2k $, some are quite reasonably upset

At least several companies are starting to capitalise on it and make small gyzmos to monitor and make sure this design flaw doesn't damage things

[median marsh]

what should they do if their income is YT and drama sells better than boring nothing-news

[crystal goblet]

Hello to all of you
I'm thinking to get new GPU for my home PC
currently have 8GB RTX 3070 ti
and thinking about
12GB RTX 5070 or 16GB RX 9070xt
this two GPU are on same price level in my country
give me some advice ?

[orchid iron]

@muted holly Also optional, a UPS for the computer, would be quite helpful in case of power outage if the client doesn't have one

[orchid iron]

Don't get anything below 16gb of vram

[crystal goblet]

how about 16GB rtx 5060 ti ? is ram more important than speed ?

[fallen oasis]

For game dev, yes.

[muted holly]

yes we have some large ones already. i need to find a UPS for our puresstorage servers

[somber surge]

The 9070XT is far better, but also more expensive. As Lab said though I would not get anything below 16GB of vram if you wish to keep your gpu for a long time

12GB is already showing its limits in various scenarios

You can also often find 9070 non-XT much cheaper, it is just a bit slower so it's definitely worth it for a ~100$ discount

[orchid iron]

Speed doesn't matter if your VRAM overflows into RAM and tanks performance into making that speed worthless

[somber surge]

Yeah...best case you get about 60GB/s of bandwidth (not even considering the terrible latency) if you have a DDR5/PCIE 5.0 x16 system

Compared to most gpus that have at least 300-400GB/s of bandwidth nowadays

[crystal goblet]

is there a point to look at 24GB rx 7900 xtx to have more Ram ... or 16gb rtx 5070 ti . these two are somewhat above my price point ?

[somber surge]

Forget about RDNA 3 gpus (RX 7000)

16GB is fine, so get the much better gpu (either 5070Ti or 9070XT)

[median marsh]

that really depends on what you do

RDNA3 is apparently bad with RT performance, if you don't need that it's still a good card

so for gamedev (without RT focus) it may be the saner choice, depending on what kind of stuff you do, imho

[somber surge]

You loose on every single new software feature from AMD though

FSR 4 is a big deal

And Redstone will probably be even bigger than that

Both of which will never come to RDNA3, and both of which are already on Nvidia gpus (equivalent of those) in a more mature package

Unless you get it very cheap (like 300+ bucks less), there is just no incentive to buy a 7900XTX compared to a 9070XT/5070Ti for the vram

[agile lodge]

https://www.asrock.com/Graphics-Card/Intel/Intel Arc Pro B60 Passive 24GB/

[mighty patrol]

Business customers only

Someone buy me one

[delicate crystal]

Did a comparisoon with an rtx pro 4000 and its pretty far off performance wise

[mighty cosmos]

or even 1.7TB/s on the 5090

[mighty patrol]

Nvm then

I'll wait for next intel drop

[mighty patrol]

https://www.youtube.com/shorts/8SxtDCxt_sI

[orchid iron]

Oh right, and then he showcased some Asus GPUs xD

[mighty patrol]

But did he buy them?

[orchid iron]

I think

[mighty patrol]

or was it lent/ already owned.

[orchid iron]

Nah

[mighty patrol]

Shame Jay

[orchid iron]

You don't lent a 5090 astral xD

[mighty patrol]

I mean it's the motherboards that have the issue

So that's okay

[mighty patrol]

Someone might

Someone lent gamers nexus that pc gpu

You never know what people are willing to risk

[orchid iron]

Oof, fair

[orchid iron]

True, that means downgrading to MSI godlike mobos, ew

[mighty patrol]

Which he was even allowed to take apart

[mighty patrol]

Is that the only "decent" mobo left?

I got MSI and gigabyte mobos myself

[orchid iron]

I was joking, MSI godlike is by far most expensive class of consumer mobos

Costs a whole pc

[mighty patrol]

lol

Zotac ftw

Need more competition

[orchid iron]

Zotac makes mobos?

[mighty patrol]

Used to

Should still

[orchid iron]

Looks like it gave up

I can only find old chips sets in their line up

[mighty patrol]

Most recent I saw was a mining board

https://www.zotac.com/us/product/mainboards/zotac-b150-mining-atx-motherboard

[orchid iron]

Is that for zen1?

Nvm, it says intel 6th and 7th gen

https://m.youtube.com/shorts/hdKnp28HnQo

[gleaming stone]

Doggo picture

[fallen oasis]

Just after this picture: doggo marks his new territory.

[gleaming stone]

Old GPU (6900xt)

New 5090

[grave sundial]

Congratulations!

Now make something useful with that GPU and become rich.

💰

[somber surge]

Sounds like those crypto scammers that want to sell you their courses to become rich 🤣

[grave sundial]

Whatever helps you make that investment back and then some.

kidney sales, stem rendered cloning, whatever rakes in the 💸

[mighty patrol]

Heck set that thing up to fold some dna proteins should be powerful enough

[fallen oasis]

I should 3dmark my laptop.

[orchid iron]

Find out how you can use the power of AI to make your own GF, install this very fishy software and never be alone anymore

https://youtube.com/shorts/uTGZQdHg1yU?si=_WfvrJ3QMQLp5LpJ

[grave sundial]

Hehehehe.
Did you read about Love and Deepspace?

[wise elbow]

helping gf with a tr5 threadripper build. having issues with getting it to post. code is 0D which seems to be memory training related?

[median marsh]

that depends on the mainboard bios i guess

so check their manual

[wise elbow]

board is a gigabyte trx50 ai top, threadripper 7970

[median marsh]

if it's memory try with one modul, if it still fails try with one of the others

[wise elbow]

manual doesnt even have the code. but its what everyone says online

tried individual modules, bios update, blowing away cmos, and waiting a long time to see if training succeeded

[median marsh]

also try a bios reset, just in case

[wise elbow]

at a loss at this point

[orchid iron]

That's an annoying code because it might be anything

[wise elbow]

it seems like these boards are just memory sensitive

[median marsh]

did it turn off and on several times? my board does that during training

[wise elbow]

nope

[orchid iron]

Check if everything is connected properly

[wise elbow]

my only other guess is that the cpu just isnt installed right

[orchid iron]

You might have not connected something properly due to how expensive the stuff is

[median marsh]

Hi - 0d is defined as:
related to memory, VGA card or other devices.

did you try another GPU?

does the GPU have power?

[wise elbow]

no, but tried without gpu and same issue

[wise elbow]

yeah

[median marsh]

i would double check all power connectors on board/gpu if they sit correct

[wise elbow]

ughhh

i had this exact same issue im pretty sure with my ryzen build a few months ago

memory training took 2 damn hours

[median marsh]

wtf

[wise elbow]

i didnt even have a code

[median marsh]

it's baffling that they can't post without memory at least

[wise elbow]

my faith in gigabyte is weaning (it was minimal to begin with)

[median marsh]

it's not like the bios would need an incredible amount of memory

[wise elbow]

fr it probably fits in cache

[orchid iron]

@wise elbow have you tried with only 1 dimm of ram?

[wise elbow]

yeah

tried every stick by itself

[orchid iron]

Then either Mobo or CPU, have you checked pins?

One last thing if clearing bios didn't help, is looking if there is a bios compatibility update on their site, either update bios or downgrade it

[wise elbow]

just going to get a different memory kit

have exhausted everything at this point

[orchid iron]

If it isn't memory kit I'd try different Mobo after

[orchid iron]

Btw, is your memory kit from the mobo's QVL list?

[wise elbow]

yeah

[orchid iron]

Oh, that's rough

[wise elbow]

though in my experience the QVL is meaningless unless you're OCing

[orchid iron]

Time to overclock the ram then xD

[orchid iron]

https://m.youtube.com/shorts/vgXz_I6wB9c

Top comment is nice

[orchid iron]

Interesting, something I didn't account for is that even if tariffs get lowered on Q4, boats take several months to arrive, so right now, there isn't as many boats as usual that would be bringing Christmas holiday stock for the US

Meaning prices this Q4 will likely spike up until those boats post tariff settling arrive

[median marsh]

you can bet that the boats are just waiting to cross the border

ships*

[wise elbow]

i don't think it takes them THAT long

now going from the very start of a contract with someone new to manufacture all the way up to ship something... yeah

[delicate crystal]

Priced out the new syatem and yikes prices have gone up over the last six years

[wise elbow]

lol

[delicate crystal]

Old system was 4k in 2019. New one.fully specced will be almost 10k

[orchid iron]

I'm curious about specs of both

[delicate crystal]

The old.one was a 1920x threadripprr with 64gig of ram and two 250 gb samsung evo.970 drives and an 8gb video card

The new one is a 7960x tr, 128 ddr5, two 4tb samsung 9100, and a pny rtx 4000 24gb vid card

[orchid iron]

yeah, I feel like there is a slight difference in specs between the 2 xD

[delicate crystal]

Quite possibly. Ill freely admit my hardware knowledge is muddled

[hollow flame]

I can price that out for 5-6k with a microcenter, I'm curious on how you got close to 10k (With a 7970x)

[delicate crystal]

Its in canadian so thst might be causing some discrepencies.

[hollow flame]

AH yes that makes sense, that would be it

[delicate crystal]

Ya i should have mentuoned thst

Also noticed tge case im putting this in is over 100 mm shorter then my current case. Wonderingnifninshoukd upgrade the fans to avoid microwaving things

[wise elbow]

it's probably fine

case length isn't too much of a factor as long as your GPU has space to intake and exhaust

[delicate crystal]

Speaking of gpu foynd tge one i was goinf to order my shop has none available in canada. Might have to pick another

[delicate crystal]

Yep. Eta estimated to ve 6 to 20 weeks. Ugh.

[orchid iron]

it is a sign for you to wait :D

[delicate crystal]

Lol dont really have a choice. Only thing 24gb or so is either grossly delayed, wont release until late 2025, is 2k and is mined within an inchnof its lufe, or is 2k and 16gb vram...and three vdrsions behind. Or tgeres the 5090 but at 4k....ow

So ya....frustrating. sorry bout that

[median marsh]

https://www.asrock.com/Graphics-Card/AMD/Radeon AI PRO R9700 Creator 32GB/index.asp

[orchid iron]

all good ^^

[delicate crystal]

Is this o e of those ai cards ive heard about?

[median marsh]

idk what you heared about

[orchid iron]

Not the gddr7 of the leaks that are supposed to come out later this year

It is just the pro card with pro drivers for big companies

[delicate crystal]

Gotcha. I had heard some cards were using ai to help them draw graphics or something like that so the name had me wondering

[delicate crystal]

Decided to go with a 5060 thats on sale and hold until this fall for hopefully better gpu selection, and possibly going forntge 9960x threadrippers

[orchid iron]

Actually that's pretty good,you can re-use it as a testing GPU to see how well your built projects run on 8gb cards

[orchid iron]

https://youtube.com/shorts/j_M4ECuoQis?si=mdDWnL5EQqTcseDq

Man, this video shows really well that when we get 3d V-cache on GPUs, the world will change massively

Also explains why 5090DD exists even though it only has reduced VRAM and bandwidth compared to 5090 and 5090D since Nvidia GPUs are so dependent on that for its performance

[grave sundial]

how much is this card going to cost?

[orchid iron]

apparently sub 1k$ but you can put 128gb of ddr5 on it

[orchid iron]

oh you mean 3d v cache GPU from AMD? Oh that one gonna be expensive AF

[grave sundial]

No, I meant the Zeus one.

If it turns out to be as good for productivity as they say it might, why not...

[orchid iron]

yeah, sub 1k for the one with 1 gpu die (they have dual gpu die and quad gpu models)

but release for public will only be around 2027, so price and other stuff might change

people are calling this a scam, baiting investors and people that would want to buy the company though, so I just hope it is real

[grave sundial]

🤷 We lose nothing by waiting and watching.

[orchid iron]

yeah, as long as it doesn't push for stuff like AI and bitcoin, it may feel more real?
Because ray tracing for Movie studios and other specific workloads feel super niche, why would a scam aim those?

[grave sundial]

It ultimately depends on how much money they make and the profits they rake in. They'll change course accordingly. Heck, even the makers of the GPU might sell the company to someone else and that someone else might take it in another direction.

What matters to me is the price point and the performance.

If they can give me a nice one cheaply, I'll check it out.

[orchid iron]

https://m.youtube.com/watch?v=oMUEsz71_xQ#

[delicate crystal]

I thinkni understand where youre going on this but can you explain more? Woyldnt replacingnit with a better card down the road do tge same thing?

[orchid iron]

are you making a game (or projects that would get rendered on other people's machines) or are you working on types of projects which only get rendered on your machine?

[delicate crystal]

Tge eventual targrt would be games but was going to start with doing assets. Either way, itd be items that are rendered on other systems

[orchid iron]

This is perfect
You can test yourself the performance of your game on other people's hardware using this GPU, me for example I going to keep using my intel quad core cpu with a nvidia 960m laptop so that I can see how my game performs on the hardware

I'm even planning to buy an equivalent hardware amd conter part that is as old just so I get an idea of that too

Because of nvidia a lot of 60 class GPUs will be stuck at 8gb of vram, so it'll be a good opportunity for you to test "low settings" that way your project will be accessible to as many potential customers as possible

[delicate crystal]

Now thats pretty cool. I ne er thouvhg pf that being an option. Start up a vm, hook the gpu.to it, and test via there for example

[orchid iron]

never used VMs for testing hardware, how would that work?

[delicate crystal]

I think, and dont quote me.on this as ive never done it with a gpu, but if you can create a vm that mierors the hardware apecs of an averafe users aystem, and hook up a specific gpu for that vm to use, that would be how I think youd so it

[mighty patrol]

@orchid iron https://www.youtube.com/watch?v=HRfbQJ6FdF0

[orchid iron]

Now that's awesome

[orchid iron]

I see

[orchid iron]

https://m.youtube.com/watch?v=V2JT-E1ynIY

[mighty patrol]

https://www.youtube.com/watch?v=xdmxM-v4KQg

[orchid iron]

Anyone here tried Green Cell UPS ?

[orchid iron]

If consumers get 128gb per dimms, that means we could have up to 512gb of ram for am6 / nova lake refresh

First am6 CPU will be 32 cores, so that's 16gb of ram per core or 8gb per thread, perfect for compiling android xD

[wise elbow]

going to start measuring in giga transfers

[orchid iron]

true

[orchid iron]

https://youtube.com/shorts/p-FwsWQfJ0s?si=2r8AOeHONXlYtWdG

He didn't expect people to get salty at how he rated the 1080ti

Comments are very upset xD

[somber surge]

I mean, that was the first and last insane flagship Nvidia ever made in recent history, almost 75% more performances, double the vram and still very affordable, barely more expensive than previous gen, nowadays it would take like 3 generations to have the same spec/perf bump

Now you get like 20% more performance if you are lucky, for 20% more money, and barely more ram, and if it's not the flagship, you can cut all of those figures in half (except the price) and get the same pitiful ram capacity

[orchid iron]

@somber surge I know, but it still impresses me how many people agree, you'd think there exists people would would say otherwise, but apparently they are a very small minority

[grave sundial]

The old GPUs were good ones. Very reliable, too. There's also a bit of psychology involved. People like me upgraded from 128MB VRAM to 1GB-2GB GPUs. The sheer performance boost and stability we witnessed (in those good ol' days) made us quite grateful for what we had.

Of course, the fact that those GPUs were workhorses designed for durability, really helped. 😄

[orchid iron]

Yeah, we had it good back in the day

[grave sundial]

I think we'll have another round of that in the future.

When we fight for the 300 billion polygon scenes.

[orchid iron]

What I'm coping hoping for is that pcie 7 makes it possible for GPU to access system ram super duper quickly and then Vram becomes more like an L4 cache rather than ram

[grave sundial]

Gimme a CGRPU.

One block to rule them all.

[somber surge]

Bandwidth is not everything, latency would kill any benefits from that

When you need stuff <16.6ms or even lower, you can't afford to go back and forth to system ram

[orchid iron]

Alright, time to rebrand pcie to THICC pcie
And create new pcie called FAST pcie

[somber surge]

Not gonna change the fact that it is physically far from system ram x)

And still has to go through the cpu which add its own latency

[orchid iron]

Alright, understood, time to solder on ddr5 slots in my GPU

[wise elbow]

didn't older AMD cards have some cache?

did it even do anything?

[grave sundial]

Isn't the Zeus card you linked here previously, based on faster RAM access or something?

[somber surge]

All gpus have cache

But it's not a couple dozen MB that will solve a lack of GB of vram

[orchid iron]

Yeah, they have it at base, but someone posted a technology that could also be placed on AMD cards

The thing is not magic, not as performant as GDDR, but not crazy hard to pull off

[grave sundial]

Cool!

I am looking forward to Zeus.

May it have mighty thunder bolts.

[orchid iron]

https://m.youtube.com/watch?v=MEhYEEXGN3A#

Using dummy silicone to transfer heat away from the computing cores so it avoids the 3d memory on top is quite smart

[orchid iron]

"Why does UserBenchmark have negative trustpilot reviews?
The 200+ trustpilot reviews are mostly written by virgin marketing accounts. Real users don't give a monkey's about big brands. More » "

That website is the gift that keeps on giving

[devout pendant]

Does someone have problems with unreal 5.6, like PC shutdown out of nowhere?

I have never encountered problems in 5.5 or lower with my project. But in 5.6, my computer might just turn off.

and I have an 850W PSU.

[orchid iron]

Temps and % usage of CPU/GPU before shutdown?

[devout pendant]

currently, running in 5.5.

CPU load is low, like 10 % at best on a 12 core cpu

running a scene with 3times my character (full groom body with millions of curves) - runs fine. But in 5.6 it shut down in seconds.

fine = 3 fps, xD but no crash

[devout pendant]

This is weird. Now i could start the 5.6 scene with my character, and it still did not shut down. before that, it crashed always in a matter of seconds.

windows event manager complains about a "Unexpected shutdown" - but, that happened one month ago too, when the power actually went down.

[grave sundial]

If you have an integrated GPU, then shift to the iGPU (disconnect dedicated GPU) and start UE 5.6. It'll be SLOW SLOW SLOW, but at least you can rule out a hardware fault if it continues to shut down the entire computer.

[devout pendant]

integrated GPU ? It is a RTX3090 by MSI

[grave sundial]

That's the dedicated GPU.

Does your CPU/mobo have an integrated one?

[devout pendant]

in hardware manager i can only find the RTX 3090 under GPU

[grave sundial]

Urf.

[orchid iron]

@devout pendant what is your CPU?

[devout pendant]

ryzen 9 , 5900X - 12core

[grave sundial]

no, no integrated graphics on that one.

[orchid iron]

Yeah

[grave sundial]

Do you have another GPU lying around? 🤞

Perhaps something you can borrow from a friend?

[orchid iron]

Perhaps try clean installing older Nvidia drivers, recent drivers have caused issues

[devout pendant]

I dont think anyone I know has a comparable GPU (VRAM)...

[grave sundial]

You don't need a comparable one, just any will do. The idea is to see if a GPU fault is causing the shutdown.

UE certainly crashes, but it shouldn't shut the whole computer down.

[devout pendant]

but an avarage 8 GB VRAM graphics card would not handle that project.

[grave sundial]

that's okay. as long as the program shuts down and NOT the whole computer.

[devout pendant]

let me try the GPU drivers. I updated them recently

[orchid iron]

@devout pendant try to find and run a heavy GPU benchmark for a few hours and check temps etc

If it is really stressed, with fine temps and pc doesn't shut down, issue is very likely just software things, which is more annoying to fix, but at least you don't have to buy new hardware

[orchid iron]

Oh yeah, then it is driver issue if you updated them recently

Make sure to clean install to remove completely old ones

[devout pendant]

Basicly these things changed since the crashes:

• transfered project to 5.6,
• added a groom plugin
• updated drivers

the first time i noticed it after the groom plugin, but i think the driver updates came before

[orchid iron]

@grave sundial my two cents are on the AI generated drivers, what about you?

[grave sundial]

AI might hate his project, yes.

[orchid iron]

True

[grave sundial]

If you are using the NVidia App, you can roll back the drivers.

[devout pendant]

actually, i did not update the drivers in the past days... This must have been my laptop drivers ._.

[grave sundial]

Wait, this is a laptop we are talking about?

Not a desktop PC?

[devout pendant]

nono, desktop Computer, but I think i updated the drivers on the laptop for other reasons, not on the desktop computer, in the past days...

[grave sundial]

Regardless, a program shouldn't shut down a computer.

[devout pendant]

so, there was no update on the computer that has the shutdown problems.

[grave sundial]

Trigger a reboot, freeze the display, ok.... entire shut down, no no.

[devout pendant]

it turns of, but the small little fan i have, keeps running 😄

[orchid iron]

Try older and newer drivers then

[grave sundial]

and doesn't reboot automatically?

[devout pendant]

no rebooting. Actually i have to wait until i can manually reboot it.

[grave sundial]

🤔 Do what Lab said. Try older drivers.

I hope that fixes the issue.

But, this is looking like a hardware fault to me.

[devout pendant]

but the system itself must have power, if the small USB-fan for my table is still working after that.

[orchid iron]

So the software failed so bad that windows couldnt do anything and forced rebooted, because hardware looks fine from screenshot and info you gave us

[grave sundial]

No, the screenshot is from his laptop.

Different machine.

[devout pendant]

noono

the screenshot is from the computer.

Forget the laptop. I just forgot that i updated the laptop in the past days, not the computer.

[grave sundial]

Okay.

Well try the older drivers.

🤞 Hope that works.

AI must really hate the project to shut down the whole computer.

[devout pendant]

first i try the newest from june 17

[grave sundial]

Problem is, it doesn't reboot automatically if I understood correctly.

[devout pendant]

it does not reboot itself

[orchid iron]

Nvidia drivers are a virus, they must be purged!

[grave sundial]

They might be trying to get Eternal to buy a 4x or 5x card.

Burning his GPU on purpose.

Telling it to shut down.

[orchid iron]

We'd see 13/14th gen levels of burn if that was the case, maybe shutting down before saving to reduce time people can work on projects so they can buy Nvidia 6090 Ti prime super subscription for 1k a month

[grave sundial]

It might be a conspiracy to just hit Eternal's computer.

A test case.

If they successfully convert him to a 5x customer....

🔥

[devout pendant]

now somethings off, after the update. rightclick on desktop and going to the menu where i can change the wallpaper is broken

and the menu that opens the resolution settings loads for ages.

[median marsh]

enable blue screens

maybe windows just resets the machine

[devout pendant]

my background is all messed up, like it is confused that monitor 1 uses 1920x1200 and the second 3940....

wtf

[grave sundial]

Is this happening on just one monitor?

Or when both are connected and on the second one?

[median marsh]

is that a screenshot or a picture?

[devout pendant]

screenshot. This just happened for the first time, and it does not happen again. I tried to box select something. But now it works. But the desktop is now still confused about the resolutiion of the wallpaper.

since monitor A is 1920 and B is 3940

[grave sundial]

Could you disconnect one of the monitors?

Or disable one of them? And see if any other error shows up?

[devout pendant]

the resolution is fine now, but the "wallpaper settings window" still not opening. just the resolution settings.

and for some reason, the resolution settings window closes itself lol

Trying sfc /scannow O.o

ssomething was repaired.

[orchid iron]

┬─┬ノ( º _ ºノ)

[devout pendant]

So, i am surprised it does not crash, but it did not crash on my last attempt a few minutes ago, before the driver update and the system scan, so.... !?

[grave sundial]

Is your UE 5.6. project working?

[devout pendant]

it did work, for about 1 minute, hopefully it won't shutdown anymore.

i will see later

[grave sundial]

Cool.

[agile lodge]

https://youtu.be/yPFIj1VuYtc?si=iplPUK2DzQOY56e1

[wise elbow]

jesus christ your pc is fucked

[orchid iron]

@devout pendant news?

[devout pendant]

Hi, it seems updating the drivers / windows repair helped. the project runs faster and so far i did not have a crash.

thanks

this would be the first time updating drivers actually helped me lol

[devout pendant]

ok it jsut happened again. on low Viewport settings, without my character in the scene, just a landscape with trees.... ._.

[devout pendant]

someone mentioned, that updating bios helped. I just updated the drivers for the MB, hopefully that will be enough

[orchid iron]

your case looks more like gpu software issue, or software issue at least

Have you tried running benchmarks?

[devout pendant]

is occt useful for that?

wait no, it does not have a gpu benchmark.

but "stability tests" for GPU

[orchid iron]

gpu benchmarks makes 100% sure it is software issue, since if it is the gpu at fault, that should make your pc crash if gpu is under heavy load

but if it doesn't, then it is very likely just software issue that can be fixed

[devout pendant]

https://gpu.userbenchmark.com/ ? i could try that one

[orchid iron]

no xd

[devout pendant]

this?

[orchid iron]

is it free

[devout pendant]

yes

[orchid iron]

okay then

[devout pendant]

i can choice the graphics API, i guess, Directx12 is the best since Unreal uses that? (vulcan and opengl also showing up)

[orchid iron]

ye

[devout pendant]

their benchmark is just 20 seconds long and does not heavily use my gpu?!

[orchid iron]

lol

try furmark

[devout pendant]

no errors where reported, i finished furmark

[orchid iron]

can you loop it for like 30min?

[devout pendant]

furmark itself does not have that option?

[orchid iron]

idk, haven't used it, just saw it used quite a bit to debug gpus

oh btw, when you open your unreal engine project, does it crash at a different part? Like before it couldn't load it, and now it takes abou 1 min to actually crash?

[devout pendant]

It seems to not crash so often. on the very first day, it basicly always crashed when I went for play, on high settings in the viewport, while my character was there. Thats why i firt thought it has something to do with the groom addon I have. But later, it crashed just on the landscape, while i was working on PCG. for example, today, i let it run on low settings and edited nodes on the other monitor...

[orchid iron]

oh so even before you did anything to drivers it crashed the same way it does now, after a while randomly?

[devout pendant]

now it seems to be totally random. But the day, when i first mentioned crashes here, they where not so often, for some reason D:

not so often, as when i first started working in UE5.6

[orchid iron]

so either it is still a driver issue, and you just gotta change driver versions until it doesn't do it anymore

or something else

when UE crashes, the short log shows where the big more detailed log is located, find it and open it, then start looking from the end and search for when the engine starts throwing errors and warnings

[devout pendant]

furmark running for 8 minutes now.

[orchid iron]

might be a plugin that is causing issue and could be disbaled to fix it

[devout pendant]

Hmmm, we are still speaking of a shutdown of the entire computer? but still theres a log for that in unreal?

[orchid iron]

nvm

forgot about that small detail, then yeah still driver problems

[devout pendant]

So i changed the drivers, they where maybe 3-4 months old, now they are the newest, from june, but still crashing...

[orchid iron]

try the ones from over 6 months ago

[devout pendant]

furmark running for 12 minutes now O.o

VRAM temp = 82°C, GPU hotspot 75°C

[orchid iron]

Yeah, GPU is fine

[devout pendant]

ok i think i know understand how to tirgger the shutdown ASAP
what i tried:

• mainboard driver update - crash within 15 seconds of playtime
• Bios Update - same
• removing the "hair blueprint" addon from the scene - same
• removing my character frome the scene - crash after 1 minute.

notice: Whenever the fans work more, it seems to crash.

furmark, on the other hand did not cause a shutdown within 20 minutes on 100% workload.

next: trying to get old drivers.

GPU driver from dec 10 /2024... let's see

and after this, i try the game not studio.

[devout pendant]

Studio driver from december: no change
interesting: This time i was able to trigger the shutdown in a 5.5. Version of UE. Something that never happened before.

and the newest game drivers shutdown aswell. Hmmmmmmmm.

[orchid iron]

@devout pendant try making empty 5.6 project

[devout pendant]

Hmm. it almost never happens when i dont have a heavy scene, so I expect that this runs fine

i think i will swap my PSU.

if the mainboard/ memory check don't show anything

[devout pendant]

I THINK I FOUND THE SOLUTION

A mainboard has 2 power supply connectors, and one of them was not connected - on the side of the PSU.

I can't tell why. It was always connected. , the cable is behind the mainboard going to the other side of the mainboard.

but now i noticed, its not plugged into the PSU.

makes me wonder when this happened lol

unlike before, its not crashing now....

https://tenor.com/view/serious-reaction-ernie-sesame-street-gif-7737611

[mighty patrol]

At least you figured it out

[orchid iron]

lol, pc didn't crash on fur marks, but somehow crashed on UE, that's how hard the engine stresses the hardware xD

[devout pendant]

https://tenor.com/view/i-don't-know-idk-idk-about-that-gif-7336250873949261946

[devout pendant]

thats not so stupid, because furmark = GPU heavy and the GPU was correctly connected.

[orchid iron]

https://youtube.com/shorts/TTMyRpH0KZk?si=NVoPpHr_3kdrxiQm

[unreal ledge]

I have a question: what’s your internet speed? Mine averages around 4–5 Mbps.

[wise elbow]

like 500down and 20up

[unreal ledge]

500 Mbps !!!!

how

[wise elbow]

us

[unreal ledge]

How much did they charge per month for that speed? Is it around 100 USD?

man, my internet speed have been like this for 7 years or so, and I thought it is already really fast

500 Mbps is like 100 times of mine.

this is nut

[somber surge]

8Gbps down 1Gbps up (yes Gigabits)

[unreal ledge]

oh man.

[somber surge]

For 24€/month

For somethings France isn't too bad :p

[wise elbow]

about 70 for me

[orchid iron]

1.4 Megabyte per second (France)

Upload is like 1/10th of that

[mighty cosmos]

2.5GBit up/down

[somber surge]

Damn still having ADSL in France
And they say they want to fully shutdown ADSL network by 2030

[mighty cosmos]

in majority of EU you can get high speeds for pretty cheap because there is a lot of competition, everyone is trying to beat the competition with the speeds they can provide

2.5gbit up and down for me is around 40 euros

[grave sundial]

300 Mbps (27-28MB/sec download and around 20MB/sec upload) up to 1TB per month at $14/EUR 12 per month in India.

[wise elbow]

here in the US every ISP seems to compete to be the absolute worst

[agile lodge]

https://youtu.be/8-Rmv2rEHc0?si=SplZxMMDANmIK1jS

[orchid iron]

Yup

It is fastest speed possible due to network installation of building

Using mobile data is like 10 times higher speed, and not even using 5G xD

[orchid iron]

Really disappointed it happened with his wireframe pro on

[delicate crystal]

Here in canada we have three main isps thst are complete dogahit, but you pay big bucks

[mighty patrol]

Unless in Sask

[delicate crystal]

Give them time, i. Sure they could fuck that.up

[unreal ledge]

I can't still not visualize how fast is 8GBps download look like. Like how much time for it to download like 50 GBs of game ? jezz

like not even come to 1 mins and it already finished downloading ?

[somber surge]

Usually the limit is the download service (Steam for example), I can't use the full 8Gbps downlink on most services. Then the local hardware can be a limit too :p
But for Steam yeah it's around 5-10 minutes for a 50GB game

When I first got fiber I couldn't believe my eyes, I went from like 25 hours to 15 minutes to download a 100GB game

I wish that to everyone

[mighty patrol]

Speedy net for all!

[grave sundial]

To get the download rate, just divide the mbps/gbps by 8. So, 8gbps would be 1GB/sec.

And I second the motion!
SUPA COOL NET FOR ALL!!!!! 📻

[delicate crystal]

Its a solid start

[delicate crystal]

Just realizsd that i fodgot to get a power cord, and im not sure if the 5060 takes tge same power connector as this csrd

[mighty patrol]

If it needs it should come with it no

[median marsh]

it has one pciex8 connector only

so your psu should have a connector

[wise elbow]

^

[orchid iron]

Should have standard 8 pin connectors

[grave sundial]

We have a special sale here for the SAPPHIRE PULSE AMD Radeon RX 9060 XT 16GB Graphics Card - $430. 😄

[orchid iron]

Ah yes, considering an msrp of 500$ it makes sense it is an amazing deal

[grave sundial]

Here, the msrp is $800.

[orchid iron]

https://tenor.com/view/damn-shookt-shocked-gif-5580082

[grave sundial]

$815.62 to be exact, if you like counting cents.

So, basically, about 50% sale. 😄

[orchid iron]

Amazing, price is way over double, so 50% sale still has a profit! Smart

[grave sundial]

Actually, when they import it into India, they have to pay huge import duties.

Electronic import tax is high. So, a lot of that $800 goes into that.

[orchid iron]

OOOF, wtf

[grave sundial]

Conversely, if it's made in India, the price is extremely cheap. Which is why RAM sticks can be purchased for less than HALF of what they cost abroad. 😄

[orchid iron]

I think that medicine is also very cheap right?

[grave sundial]

Generics are, yes. Most 'curable' diseases' meds are cheap enough.

The higher-end stuff is more expensive, but we do have charitable hosptials (like in my town), where treatment is free up to a point. The poorer you are, the longer that point is, but yeah....

[orchid iron]

cool

[wise elbow]

should be 350 in us

[wise elbow]

India meds are so cheap. so much cheaper to buy direct from India than in US. everything gets marked up tens of thousands of percent

even though they are still made in india 🙃

[delicate crystal]

Huh. 550 cad here. So roughly 400 usd

[blazing solstice]

Hi guys, what do you think about this specs , I am a unreal engine artist who mainly works on Lighting and Cinematics.
INTEL CORE I9 14900K CPU
ASUS PROART Z790 CREATOR WIFI MB
TEAM T-CREATE EXPERT 64GB (32X2) 6000MHZ Ram
SAMSUNG 990 PRO 2TB NVME GEN4 SSD-NVME
RTX 5080 16gb

[fallen oasis]

Looks pretty good

[orchid iron]

When have you updated motherboard BIOS?

And when did you purchase it?

[blazing solstice]

didnt buy this yet. this is waht im thinking of getting

[wise elbow]

looks fine

[orchid iron]

Well, you might struggle quite a bit cooling that CPU, even with water cooling
Recently had a micro code issue where BIOS would put too much voltage in the CPU causing it to fry, and the symptoms were much more noticeable on UE
That's why I asked if you updated BIOS / when was it purchased

Now that motherboard seems very overkill, unless you need very high download / upload speed and a tone of drives connected to your pc
Although once you build it, you won't be able to upgrade the CPU since it is a socket that won't get anything new
Looking at that 64gb of ram, I'm assuming your scene doesn't have many assets or other things that might eat up all your ram

[wise elbow]

i would probably recommend a newer intel or amd though

and yeah might want to consider 2x48gb memory.

but I have no idea how much what you do consumes

[orchid iron]

2x64gb sticks of ddr5 memory were released recently too

[blazing solstice]

can u guys recomend me a cpu and a motherboard instead of this pls

[orchid iron]

Also, rumours say that RTX 5080 super 24gb is very likely to release end of this year, so if you don't mind waiting a few months, you could get it

[blazing solstice]

this is the website of the dealer https://www.nanotek.lk/category/processor

[orchid iron]

Newest intel cpu socket sadly seems on its way out already, and the intel 285k is basically almost same as 14900k
So if you really want intel those CPUs are the main options

On AMD 's socket: am5, the 9950x and 9950x3d are quite good picks with 16 zen5 cores, and rumours say it is very likely next CPU flagships will have 24 cores, so there is upgradability there

Motherboard wise, depends what do you want for specifications, do you want high download / upload speeds, do you want to have connections to how many m.2 or SATA drives?

[blazing solstice]

IM KOOL WITH amd AND MOTHERBAORD WISE I JUST NEED TO BE ABLE TO PLUG 2 SSD'S. AND BE ABLE TO PLUG A 590 LATER. sorry caps

[somber surge]

7950X3D is also a very good pick if you can still find it new, often cheaper than the 9950X while outperforming it in gaming and not being far behind in productivity (and it's the most efficient cpu on the market for heavy workloads)

[orchid iron]

True

[wise elbow]

285k is a lot better afaik?

[somber surge]

I would definitely go AM5 than anything Intel though for sure, just for the fact that you are not buying on a platform that is already dead

[orchid iron]

Is it finally better than 14900k?

[wise elbow]

than the 14900k

yeah

[orchid iron]

Noice

Took intel a while xD

[somber surge]

A bit better in productivity, a lot worse in gaming

It's getting destroyed by AMD X3D in both though

They might have fixed some of the issues so it's probably not as bad on gaming anymore, but still behind last gen which is not a good show

[wise elbow]

IMO in socket upgrades are overblown. by the time it's worth upgrading your PC a marginal generational uplift won't be worth it for the price of a whole new CPU

it's a nice option but I don't think it should be a reason to buy

[somber surge]

Depends on the upgrades you are getting

[orchid iron]

The engineer team that is making the 24 core zen6 CPU has made zen2 and zen4, remember those generational leaps?

[somber surge]

If you went from like a Ryzen 5 3600 to a 5800X3D that was a MASSIVE upgrade in all regards

[orchid iron]

Plus going from 16 cores to 24

[somber surge]

And Zen 6 does sounds like one of those massives upgrades for Zen 4/5 users

Though for sure Zen 5 was bad.

[wise elbow]

yeah but that's also going from a dog shit budget CPU to a basically flagship

[orchid iron]

Zen4 was so good it made zen5 non x3d look so bad xD

[somber surge]

It's not, the 3600 was very solid by its time, and the 5800X3D isn't that high end nor expensive

It's technically just 6c vs 8c + X3D and a gen newer

[wise elbow]

it was like 200 dollars more was it not

[somber surge]

And Zen 6 is like 3 node jumps, going from N5 to N2X

[somber surge]

Still much less than upgrading the whole system

[orchid iron]

Oh jeez, I thought it was N4, dayyym

[somber surge]

The marketing says 4nm but it's still an N5 class chip by TSMC standards

Everyone just skipped N3 aside from some Apple chips x)

[orchid iron]

Btw if you decide to pick am5, get a motherboard with x870e chip from one of the main vendors: A sus , Gigabite, MSI and ASRock
You don't need to get anything expensive, just maybe around half of CPU price will be fine

[orchid iron]

Ah gotcha

[blazing solstice]

@orchid iron Thank you so much.

[somber surge]

X670E is pretty much the same for cheaper most of the time

X870E just ads mandatory USB 4

[orchid iron]

Yeah but some can't handle 8000 MT/s speeds

[wise elbow]

who cares lmao get under 6400 MT

it's down to lottery that your CPU likes fast memory anyway

to my knowledge the memory controller on my 9950x3d is generationally bad and hates even a stable 6400 cl32

[somber surge]

Agreed, anything above 6000 is risky, 6400 is already a stretch for most Zen 4/5

[orchid iron]

Oh yeah, zen5 couldn't get a new memory controller so it is same one as zen4, but zen6 will have a new one, so maybe 8000 MT/s will be new sweet stable Spot

[somber surge]

I doubt the memory controller will get a +1Ghz boost to maintain a 1:1 ratio with 8000 DDR5

More likely the "sweet spot" for 1:1 will be around 7400 or something or won't even move

[orchid iron]

Why would they push for 8000MT/s so much on x870e then? Zen4 and 5 couldn't handle it effectively, would be very odd if zen6 couldn't aswell

[somber surge]

They can, but not at a 1:1 ratio

So depending on the workload it might be worth or not

Generally it's not

But maybe who knows, though even if that's the new sweet spot, it would need testing to see if it actually matters and is worth it

[orchid iron]

Yup

[somber surge]

If it just add 5% more performances while the generationnal gain already gives you +50% perfs...who cares 🤣

[orchid iron]

Imagine it adds 6% more perf while generational gain is like 80% xD

[wise elbow]

it's like 5% in some workloads

high memory bandwidth iirc

[orchid iron]

Com'on we can't be calling it zen5%, that's already taken xD

[somber surge]

We will have to see, Zen 5 was definitely very memory bottlenecked, so maybe faster ram will actually matter on Zen 6 with the new memory controller

[wise elbow]

zen5 seems to take PBO a lot better tho

[orchid iron]

https://m.youtube.com/watch?v=jP0zZwIDy_M

[somber surge]

That's because they moved the temp sensor to make it cooler

[somber surge]

F--- Nvidia

[wise elbow]

zen5 also eats considerably less power I'm pretty sure

enough to let you boost higher

[somber surge]

Not that much really

It's pretty much the same, just a bit more efficient

Zen 5 on Desktop is a small refresh rather than a new generation

[orchid iron]

Wise words

[orchid iron]

https://m.youtube.com/watch?v=EiOVOnMY5jI#

8gb of Vram is so bad, that in some scenarios, turning on frame gen can result in LOWER total FPS over native

[wise elbow]

shocker that an already insufficient amount of memory becomes even more insufficient when you use more

it's a shame they marketed framegen like that because it's pretty magical otherwise

[orchid iron]

marketing is the equivalent of frame generation

It generates more money from nothing

[somber surge]

I always called Frame Gen a complete BS on 8GB GPUs

Not enough performances for a good minimum fps, not enough VRAM to run the tech

And multi frame gen is even worse since you need an even higher baseline fps for it to be usable

Good tech in the right condition but the marketing is catastrophic and 8GB GPUs well...they're 8GB GPUs

[violet spade]

Frame gen is complete bs on any gpu. It’s fake frames, the performance does not actually
improve, it just makes people feel cooler cause the number is higher lol

[wise elbow]

i can 100% say that it improves motion clarity in what its applied to (assuming you have a decent base framerate)

even the "smooth motion" feature on tvs improves motion clarity; it just makes it really uncanny

this avoids that

no, it obviously wont improve latency (it slightly harms it) and wont make a game feel more responsive to your inputs

[mighty cosmos]

it does improve the experiencec quite a bit, it's awesome tech

but you need at least 60fps base frame rate for it to feel good

[orchid iron]

Frame gen on videos would be amazing

On real time rendering like video games, it the answer becomes quite blurry, so perhaps we should ask the AI generating the frames, how it feels, perhaps Jenson 's jacket was actually silk all along

[wise elbow]

you can use it on videos afaik

[somber surge]

It's not though. It has been overly used for bs marketing claims and it does not improve performances (no, using MFG X4 to go from 30 to 120 fps IS NOT usable or enjoyable Nvidia), but with the right conditions it really improved greatly the experience imo

If you have a good minimum framerate, unless the game already has a bad input latency, it really is free "smoothness"

I use it often in games to lower my power draw as well, going from like 70 to 120-130 fps instead of drawing much more power to reach those 130 "raw", and to me it feels the same

But it's definitely not performances and can't absolutely be compared to a native framerate, especially if you have a low base framerate, 50-60 fps can still feel ok if the game has a good base latency, but anything lower is terrible

I would be interested in seeing how they can merge FG with Reflex 2 though (if that comes out some day), that would be really game changer

[violet spade]

Iirc GN was also saying it makes up shit that’s not actually there when generating those frames, tho tbh that’s not much unlike what our brain does to fill in the gaps that our eyes aren’t actually seeing

[somber surge]

Frame Gen x3-4 does definitely have a lot more artefacts that FG x2, and their new "Transformer Model" is arguably worse quality wise (though a bit faster and lower vram cost)

But with a high enough base framerate those issues are not very noticeable

That does explain why FG x3-4 require an even higher base framerate though, like 70-90

[wise elbow]

its interpolation; its making its best guess as to what the in-between frame would be

historically the interpolation looked smooth but weird because it was basically linear

using ML it can make something more natural looking

[orchid iron]

It is also impossible because MFG X4 is pretty resource intensive, so it'll lower FPS to maybe 15-17 FPS, and then it'll multiply by 4
So you'd get above 60 FPS with about 1/4 being real frames for the game like tick, collision detection, physics etc

[somber surge]

Yeah I meant more like 50 fps to 120, which would make internally a 30 -> 120

In any case, MFG X4 is not usable below at least 70-90fps

Way too many artifacts and input latency would become terrible

I'm eager to see the RTX 6000 with MFG x10: look the 6060 (6GB vram) is better than a 5090!!!

[orchid iron]

Oh yeah, more frames!!!

[agile lodge]

https://youtu.be/AzZWd0MiHR4?si=TH4VXoAyCdZtY1Rx

[median marsh]

it was likely a cap if it was loud 😄

but damn... watching him handle high voltage parts...

that video could have ended sooner 😄

[mighty cosmos]

It's the opposite, transformer model is much higher quality because it allows the model to attend to the entire image at the same time while generating unlike the CNN model (what dlss used previously) that has a sliding convolution kernel (CNN models don't "see" the entire image at once)
But the transformer model is a bit slower

For example the transformer model can much more effectively resolve thin objects, fences, etc... because it can attend to the entire image so it has more context
In the old CNN versions those would often shimmer or disappear completely

it's also smarter with resolving incorrect motion vectors, where previously it would produce ghosting, now it doesn't

The biggest problem with frame gen is if you run the game at 30fps and frame gen to 120 or whatever
The engine is still ticking and processing input and physics at 30Hz (physics can be async and fixed tick so not a problem for all games, UE supports async fixed physics tick)

The ideal solution would be something that allows the engine to tick without rendering during generated frames... but you can't do that with an interpolative system
You need a predictive frame gen (predicting ahead) but that has its own set of issues, but I think that is going to be the direction frame gen will move towards

[somber surge]

DLSS 4 Upscaling with Transform is far better yes, but the Frame Gen isn't

GN did a good deep dive and the new model has a lot more issues (around 21 minutes) https://youtu.be/3nfEkuqNX4k

That's probably because the Optical Flow Accelerator (which was the entire reason why FG was supposedly not compatible with RTX 3000) has been replaced by a ML solution, it is cheaper but also worse

[mighty cosmos]

For whatever reason I assumed you were talking about super resolution, mb

Tbf the new one is better in some things and worse in others but the new one has much more potential because the tech behind it (model architecture) is just better
So I'm expecting the issues to get ironed out through additional training of the frame gen model
Replacing the Optical Flow Accelerator with an ML solution is not necessarily a bad thing because it opens up doors for things a purely hardware solution can't do

[somber surge]

Yeah for sure it's not necessarily bad, though for the moment FG is somewhat worse

[mighty cosmos]

I like Nvidia's research team, but passionately hate their marketing and business side 😆
The tech is awesome, but the problem is when the marketing side throws around the "5070 is faster than the 4090" BS

[somber surge]

And so it's even worse with x3-x4 FG

Yeah that's really a shame, the tech itself is wonderful, heck even the hardware is good, it's just that they have a shitty marketing team and want way too much margin on low end product

[mighty cosmos]

you know it's bad when even AI is roasting them for it 😆

[grave sundial]

Ironically, that side fuels their research. 😄

[somber surge]

Give that 5060 a good 12GB of vram and it would be the best low end gpu in a while

[somber surge]

AI fuels their research

Like they don't have anything to gain from doing BS on gaming, why the hell they still do, it's not like it will increase their AI margins or something

[mighty cosmos]

nvidia is rolling in money, they could be giving their gpus for free for the next 10 years and I'm pretty sure they would still be in buisness lol

[grave sundial]

Perhaps, but they can't. There are shareholder obligations. Must make more moolah.

Perennially.

Ceaselessly.

Perpetually.

🤑

[somber surge]

Gaming hardware is like what 10% for them now? For sure they would be fine

At this point just stop releasing anything if it's just to release garbage

Though again...the gpus aren't terrible, the real problem is the insane lack of vram

You keep the prices, the performance but give the 5060s 12GB, the 5070s 18GB and 5080 24GB, it's a really solid lineup

[mighty cosmos]

I would be surprised if it's even that much tbh
Since AI accelerators are going for like 30k usd per unit and they have a backlog of orders to still deliver

[somber surge]

Though they probably didn't because they need the 3GB chips for AI and they have limited supply of them

[grave sundial]

I wonder if they finished delivering those, what - 30 million something - cloud GPU orders.

[mighty cosmos]

yeah selling 8GB GPUs in 2025 should be a crime

[somber surge]

Though to be honest...RTX 5000 were in development while AI boom wasn't even a thing yet

So we can't put all of that on AI, it's just pure greed.

[mighty cosmos]

88% datacenters lol

[somber surge]

The lack of vram probably can be related to AI though, not wanting to "waste" 3GB chips on peasant gpus

[grave sundial]

The peasants can just toil in the fields and save up for the Blackwells, damn it! 🧑‍🌾

[somber surge]

Even the 5090 is nothing for them, that's what probably 300% margin? Hold my beer those datacenter gpus are more like 3000% margin 🤣

[mighty cosmos]

oh the datacenter gpus are overpriced to hell and back
I don't think there is anything on those GPUs to cost 30k, memory is not that expensive

[somber surge]

Those gpus probably don't cost more than 5K to make

[grave sundial]

Most of that money is probably for guaranteed support, replacements, etc.

Which, if they don't happen, is just extra money in the piggy bank.

🐖

[somber surge]

Just look at the RTX PRO 6000, it's like 9K for the same gpu as the 5090, same cooler and a bit more ram

[grave sundial]

You're paying for those three letters.

P R O.

[somber surge]

At least you don't need to throw away your gpu in the next year because it has vram capacity of the last decade

🤣

[grave sundial]

I upgrade once in 10 years.

It's good enough for me.

[somber surge]

Upgrade once a decade...still stuck on 8GB that's insane 🤣

[grave sundial]

It makes the jumps feel worthwhile AND usually, whatever I buy lasts me that long.

Not me. I started with 128MB.

Went to 2GB, if I recall.

Then 8GB and now, 16GB. 😄

It's been SO MANY years.

[somber surge]

Just saying... most people will be stuck with 8GB well into the 2030s

While it was already a thing in the 2010s

[grave sundial]

🤷 Most people don't really have any heavy use for anything more than that, which is why NVidia can get away with it.

Even if they did have a use for more VRAM, they don't 'know' about it.

As long as things 'work', it's fine.

[mighty cosmos]

At work we have a couple of machines with the rtx 6000 ada cards, they were like 8k each pretty sure
They feel so cheap compared to my 5090 or 4090, it legit feels like you are getting a worse product for more money, and it's slower than a 4090
So it's basically super inflated pricing for that extra memory

[grave sundial]

If the cards ever break down, will NVidia replace them for you?

[somber surge]

It's not fine anymore, and definitely not fine for a brand new 400$ gpu that would need to last while a new console generation get released

[grave sundial]

True. The pricing has really caught up to the shenanigans.

[mighty cosmos]

I would put my bets on "absolutely not" lol

[somber surge]

Is it the "Workstation Edition"?

It should have the same cooler as the 5090

And a bit higher performances

[mighty cosmos]

it's the older ada gen, we got them before the new ones were out

this

[grave sundial]

Apparently, they do if the card is under warranty.

[somber surge]

Ah yeah that crap

[grave sundial]

How long is the warranty on the ada?

[somber surge]

Probably like 2 months so they can get away with a burning 12V High Failure connector

[grave sundial]

🤔 3-years according to Gemini.

Must break and replace cards for free before that. 🤣

[somber surge]

Anyway

[mighty cosmos]

The 5090 is actually a pretty insane card if you have the right workload for it
In games yeah it's like 40-45% faster than the 4090, but for memory bandwidth bound workloads like machine learning or VR, it's anywhere from 1.8x to 2x faster than the 4090 thanks to that insane memory bandwidth

[somber surge]

Let's just remember that AMD does the exact same BS while they have less than 10% market share, I will always be amazed at that

[mighty cosmos]

AMD has such a good chance to win over the market and they are doing a mighty good job and fumbling that chance

[somber surge]

More like 25-30% at most, the bandwidth from GDDR7 is wasted with that architecture (for gaming at least)

[grave sundial]

That's probably what they designed the card for, keeping ML and AI in mind.

[mighty cosmos]

From my tests comparing the 4090 and 5090 it's 40-45% on average in gaming

[orchid iron]

I started with integrated, then while looking at old laptops I owned, I found an HP laptop with intel CPU and Radeon GPU
And then the laptop I have currently (4 core intel and 960m GPU which has 4gb of Vram even though I was fooled into thinking it had 3gb)

[somber surge]

They wasted their chance already and they will be going back to the forgotten land when RTX 5000 Super drop and fix that stupid vram issue

RX 9070 XT still isn't available near MSRP is most regions..

[grave sundial]

Let's not forget the excitement of 'shared video memory'.

I remember how hyped I was when I read about the dGPU sharing the RAM in laptops way back when.

It didn't work quite the way I thought it would.