GPU stress testing

[Octopuss]

I decided to start playing with some undervolting and noticed the AMD driver has a stress test feature. Is it useless or is it good enough? Any ideas?

 

[The red spirit]

I decided to start playing with some undervolting and noticed the AMD driver has a stress test feature. Is it useless or is it good enough? Any ideas?

From my experience useless. It claims that card is stable, meanwhile after running Valley for few seconds card crashed. But to be fair, even testing card with Unigine tests is only so good. I had my RX 580 crash in Cyberpunk 2077, when it used FSR. Apparently FSR put more load to card and I needed to up the voltage, but either way I wouldn't trust that AMD tool at all. Hell, I would rather trust Furmark more than AMD stress test.

 

[Octopuss]

Furmark puts unrealistic stress on a GPU though, so it might even underclock and the result wouldn't be accurate anyway.

Really, there are so many programs out there and you just can't tell what's good and what's not.

 

[Sithaer]

From my experience useless. It claims that card is stable, meanwhile after running Valley for few seconds card crashed. But to be fair, even testing card with Unigine tests is only so good. I had my RX 580 crash in Cyberpunk 2077, when it used FSR. Apparently FSR put more load to card and I needed to up the voltage, but either way I wouldn't trust that AMD tool at all. Hell, I would rather trust Furmark more than AMD stress test.

Yup, this is why I also set up my undervolt in a demanding game and tinker with it till its stable while gaming.
Good thing is that you can have the game open while doing that, worst case scenario it will crash and reset to default voltages in the driver/Afterburner.

 

[The red spirit]

Yup, this is why I also set up my undervolt in a demanding game and tinker with it till its stable while gaming.
Good thing is that you can have the game open while doing that, worst case scenario it will crash and reset to default voltages in the driver/Afterburner.

Sometiems it's that easy, other times not so much. I used to own a weird GTX 650 Ti, which was stable in Unigine benches, also stable in games, but for some reason started to crash in GTA 5 for a a while, but then at same settings it ran GTA 5 perfectly fine for hours with even higher clocks and same volts. And back then cards like 650 Ti didn't even have any dynamic boost, so it just ran at fixed speed during gaming. I personally had the worst experience with RAM overclocking/undervolting. I frankly just gave up on it, because instability was random, no software can really detect it, all RAM testing software has been absolute garbage that purpose and even if you end up with "stable" settings after week your computer may still randomly crash. That is the absolute worst. CPUs were fine to overclock before boost, EIST and other things, but nowadays it's not worth doing that and now on top of various boost states, chips also have P and E cores, that's just makes overclocking dead to me. Our stress testing capabilities just aren't anywhere close to truly ensuring what will be stable anymore, hell we were struggling with CPUs without boost, EIST, Speedshift or even C states.

 

[Sithaer]

Sometiems it's that easy, other times not so much. I used to own a weird GTX 650 Ti, which was stable in Unigine benches, also stable in games, but for some reason started to crash in GTA 5 for a a while, but then at same settings it ran GTA 5 perfectly fine for hours with even higher clocks and same volts. And back then cards like 650 Ti didn't even have any dynamic boost, so it just ran at fixed speed during gaming. I personally had the worst experience with RAM overclocking/undervolting. I frankly just gave up on it, because instability was random, no software can really detect it, all RAM testing software has been absolute garbage that purpose and even if you end up with "stable" settings after week your computer may still randomly crash. That is the absolute worst. CPUs were fine to overclock before boost, EIST and other things, but nowadays it's not worth doing that and now on top of various boost states, chips also have P and E cores, that's just makes overclocking dead to me. Our stress testing capabilities just aren't anywhere close to truly ensuring what will be stable anymore, hell we were struggling with CPUs without boost, EIST, Speedshift or even C states.

So far I only undervolted the core voltage on my RX 570 using Adrenaline and Afterburner with my GTX 1070 and my current 3060 Ti.
This 3060 Ti was a bit tricky cause while it was rock solid under rasterization game load it wasn't with RT enabled on top. 'Cyberpunk'
Had to increase it a bit while lowering the boost clock a little and now it seems to be stable regardless of what I play/settings.

Probably I could further tinker with it but I don't want to risk stability and for me it runs well enough anyway. 'better than stock settings'
Can't say much about OC cause I was never interested in that.

 

[GerKNG]

there is no "true" test...
you can have a timespy stress test passing with 99,9%, play 100 hours of R6 Siege, BF2042 and Forza and then it insta crashes in Anno 1800.

it's trial and error.

 

[ThrashZone]

Hi,
Blender opendata can be switched to gpu testing easily and about as easy as it gets
Also blender in general can be switched to gpu in it's settings and just use classroom/... demo files if you already have blender installed
Blender Open Data — blender.org


Demo Files — blender.org

 

[Octopuss]

Guess I will just keep playing.
But trying to undervolt a card that's factory-pushed way over the stock frequencies (1340 vs 1411MHz) is mostly doomed to failure. I dropped it to 1390 and just cut the voltage in last 3 states by 50mV down from 1150 and will see if I get any driver reset in the next few days.

 

[Hofnaerrchen]

Just ran a - totally stable - 600 sec. Adrenaline stress test on the new 22.10.2 driver just to find out, that 3dMark (demo) did crash with those settings after just a few seconds. Same was true for the Furmark stress test, that ran for about 30 minutes stable but with just using 78%-85% GPU usage. By my experience - even if I would not pay 25-30 bucks for it - 3dMark is quite good in testing OC/UV stability. If you want to use it for stress testing you have to get the advanced edition.

 

[ThrashZone]

Hi,
Blender stuff is free and very good.

 

[Kapone33]

AMD stress test is defaulted to 30 seconds. If you run it longer it will give a truer representation but I have found that the newer the Game the more the GPU clock is maintained at it's highest. This test does the same thing but can't compare to Gaming for 3 hours.

 

[HTC]

Back when Zen appeared (or Zen 2: not sure), The Stilt (known AMD overclocker) used Prime95 + Heaven simultaneously to test his PC's stability. I've seen a video of him doing this with 4 mix matched sticks of RAM (total of 32 GBs) and i also remember him selecting more RAM to be used with Prime95 (as in not the default settings).

Can't seem to find the video

 

[ir_cow]

MSI Kombustor. Free and is does a great job at GPU stressing. https://geeks3d.com/furmark/kombustor/downloads/. Another good one, but not free is 3DMark Stress Loop test. It goes 20 loops on Time-Spy Extreme.