Personal computing discussed

Moderators: renee, morphine, SecretSquirrel

 
Yan
Gerbil XP
Topic Author
Posts: 349
Joined: Fri Dec 21, 2012 9:37 pm
Location: Ottawa

Program like Prime95, but for GPU?

Tue Jul 18, 2017 3:42 pm

Is there a program like Prime95, or Memtest86+, but for GPUs, that not only checks whether the graphic card crashes, but also whether it comes up with the correct results? Does such a thing make sense in the case of a GPU?

I know about Furmark, but I'm looking for a program that can say whether the GPU didn't crash, but is nevertheless somehow defective.
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Program like Prime95, but for GPU?

Tue Jul 18, 2017 3:52 pm

That's a tough one, since to verify that a GPU is producing correct output for every frame, you need something that can compute the same output. In other words, a known good GPU. I suppose you could run a GPU-intensive workload, then in the middle of all that feed it a "reference" frame to render, and see if it comes up with the correct result.

But different GPUs (and different driver versions) may produce slightly different results on a pixel-by-pixel basis, so comparing the output frames is problematic. How much of a discrepancy is meaningful?

Alternatively, you could stress the GPU with a GPU compute workload, and compare the results with known correct reference results. But this isn't going to precisely mimic the workload of a game, so again you're kind of back to comparing apples and oranges.

I suspect that the short answer is "no, that doesn't exist".
Nostalgia isn't what it used to be.
 
biffzinker
Gerbil Jedi
Posts: 1998
Joined: Tue Mar 21, 2006 3:53 pm
Location: AK, USA

Re: Program like Prime95, but for GPU?

Tue Jul 18, 2017 4:05 pm

Is OCCT what you might be looking for?

Haven't used it myself but it claims to offer detecting manufacture errors for the GPU memory, and has a stress test for the GPU.

Image
http://www.ocbase.com/
It would take you 2,363 continuous hours or 98 days,11 hours, and 35 minutes of gameplay to complete your Steam library.
In this time you could travel to Venus one time.
 
Yan
Gerbil XP
Topic Author
Posts: 349
Joined: Fri Dec 21, 2012 9:37 pm
Location: Ottawa

Re: Program like Prime95, but for GPU?

Tue Jul 18, 2017 4:06 pm

Suppose I wanted to buy a used GPU, but wanted to test it first. What would I do? Simply try playing a game, and perhaps running Furmark?

That's not the reason I'm asking, but it would be a practical example of why one might want such a program.
 
mikewinddale
Gerbil First Class
Posts: 173
Joined: Sat Jan 07, 2017 2:22 am

Re: Program like Prime95, but for GPU?

Tue Jul 18, 2017 4:06 pm

To build on to the previous answer:

Consumer GPU firmware tends to sacrifice accuracy for speed. Think about how JPEG and MP3 other lossy compression will sacrifice some fidelity in favor of storage space. With audio-visual data, exact accuracy is not required, because the human eye and ear won't notice anyway. And every graphics card will have a slightly different algorithm for how to achieve the best ratio of accuracy to performance. So like the previous answer, I'm not sure what it means to test for a "correct" output by a GPU.

Now, with a professional card like a Quadro or a FireGL, things are different. Those professional cards aim for perfect accuracy, which is why they get such low framerates. Perhaps for testing those cards, there is something like Prime95. But in that case, you'd have to pair the Prime95-equivalent GPU-tester with a suitable GPU firmware that was aiming for accuracy. (The difference between a Geforce and a Quadro is mostly in the firmware, not the hardware.)
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Program like Prime95, but for GPU?

Tue Jul 18, 2017 4:14 pm

biffzinker wrote:
Is OCCT what you might be looking for?

Haven't used it myself but it claims to offer detecting manufacture errors for the GPU memory, and has a stress test for the GPU.

If the majority of GPU errors tend to come from VRAM problems (I have no idea whether this is the case), that may be a reasonable approximation to what the OP is asking for. Otherwise, it is probably not much more useful than running Furmark and seeing if the GPU crashes.
Nostalgia isn't what it used to be.
 
Yan
Gerbil XP
Topic Author
Posts: 349
Joined: Fri Dec 21, 2012 9:37 pm
Location: Ottawa

Re: Program like Prime95, but for GPU?

Tue Jul 18, 2017 4:17 pm

biffzinker wrote:
Is OCCT what you might be looking for?

Interesting program, with a weird GUI. ;-) Thanks for mentioning it!
 
Kougar
Minister of Gerbil Affairs
Posts: 2306
Joined: Tue Dec 02, 2008 2:12 am
Location: Texas

Re: Program like Prime95, but for GPU?

Wed Jul 19, 2017 5:49 pm

just brew it! wrote:
That's a tough one, since to verify that a GPU is producing correct output for every frame, you need something that can compute the same output. In other words, a known good GPU. I suppose you could run a GPU-intensive workload, then in the middle of all that feed it a "reference" frame to render, and see if it comes up with the correct result.

But different GPUs (and different driver versions) may produce slightly different results on a pixel-by-pixel basis, so comparing the output frames is problematic. How much of a discrepancy is meaningful?


I don't see the problem?? Just compute the same frame again and if the raw data isn't identical, then obviously something isn't stable. If a GPU was unstable it wouldn't compute the same image data twice. I suppose the artifact scanners could be simply using the CPU to check for out-of-bounds data too as that would be less computationally intensive.

Now I'm curious, I'm not really sure how GPU artifact scanners work. I assumed it was just checksum comparisons on two frames. EVGA's OC Scanner X has a dozen various specific types of workloads, DX, OpenGL, Compute, Tessellation, Particles, and some others. It offers built-in artifact scanning for about half of the various offered workloads: https://www.evga.com/ocscanner/ This program is pretty old and is an amalgamation of Furmark and other apps. Some of the older tests no longer work, but most of them do.

In my experience if there is a single extremely rare artifact that shows up once every couple minutes, then if the same test is kept running eventually the corruption will keep growing until it becomes visible, then keep growing still until it eventually crashes the program or GPU driver. So I kill any test if the app finds a single artifact, regardless of whether it is visible or not.
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Program like Prime95, but for GPU?

Wed Jul 19, 2017 6:03 pm

Bad VRAM or busted shader could easily give you same wrong frame twice if bits are stuck. If we're just looking for failures which cause random intermittent corruption under load then I'd agree with you.
Nostalgia isn't what it used to be.
 
ClickClick5
Gerbil Elite
Posts: 581
Joined: Fri Aug 15, 2008 8:28 pm
Location: Alaska
Contact:

Re: Program like Prime95, but for GPU?

Wed Jul 19, 2017 6:10 pm

One that comes to mind for the GPU would be FurMark. Not really a test program, but a stress program. And if there are any problems with the GPU, that program will find it. Most likely in the form of a driver crash or an occasional BSOD.

This is how I diagnosed a bad VRM on an old card.
i7-5930k (4.1GHz), 32GB DDR4-2400, 2080Ti, GA-X99-UD3, Samsung 860 Pro 256GB (os/programs), Toshiba 5TB (games), WD VelociRaptor 500GB (games/scratch disk), Win 10 Pro x64
Check out my site for trance sets and other goodies! clickclick5.com!
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Program like Prime95, but for GPU?

Wed Jul 19, 2017 11:09 pm

ClickClick5 wrote:
One that comes to mind for the GPU would be FurMark. Not really a test program, but a stress program. And if there are any problems with the GPU, that program will find it. Most likely in the form of a driver crash or an occasional BSOD.

This is how I diagnosed a bad VRM on an old card.

Yes, Furmark was mentioned in the OP...
Nostalgia isn't what it used to be.

Who is online

Users browsing this forum: No registered users and 1 guest
GZIP: On