2

I'm looking for a tool that perform different operations on a graphic card in order to check for any problem or glitch.
I had a nasty freeze / crash problem with some games recently. At best I have a windows notification telling me that the nvidia driver crashed and was restarted, at worst the whole system freeze up and I have to pull the plug. So I tried to revert to the previous driver and to install the latest one (which is WHQL signed, no less), but to no avail. I tried to run FurMark and FluidMark while monitoring the temperature, and it reached 90° and stayed there without any problem. Since I've seen the game crash several seconds after it was started, I guess this isn't heat-related too. My current guess is that the game try to perform a specific operation or use a subsystem that the benchmarks didn't try, and that this cause the driver / hardware to fail for whatever reason. So is there any tool that exhaustively test a GPU for stability ? Thanks !

2

While they are meant to test memory primarily, folding at home's gpu stress tests apparently also find other issues with video cards.

I'd also update all drivers to the latest possible versions.

  • Thanks. I've been running it for 300 hundred iterations now (at a 800 MB load), and things seems to be doing well. I guess I should try to let it work overnight to be sure. – Timst Mar 6 '12 at 18:15
  • memtestG80 seems to have a 1024MB testing ceiling. Strangely memtestCL produced high GPU load, but zero load on the GPU's memory controller (memtestG80 loaded both). – Louis Mar 13 '13 at 3:38
2

Furmark will be very very harsh on the GPU chip, but does not check the video cards ram very well. GpuTool and AtiTray tool (mostly for ati/amd), will do what is called artifacts testing, it races wildly through the video memory, looking for places where the memory changed, and it heat the rams up more than most games would.

Here is a video ram tester recommended for the Nvida http://mikelab.kiev.ua/index_en.php?page=PROGRAMS/vmt_en

Here is some sort of video ram tester from major geeks, as long as you dont hit the wrong "download me" button there, they are also a reliable download location . http://majorgeeks.com/Video_Memory_Stress_Test__d5896.html

I dont know of something that would do Both at the same time (other than games). Because of the temps , which everyone including manufactures suggest is "ok" and I think is nuts. and Because of the type of fail. I would just crank up the Fan. there are tools that will make a new "fan graph" For some testing.

With a Gforce card, there is the "Msi-Afterburner" as one of the software utilities that can adjust the fan speed, and create a fan graph. With a fan graph you could Just tweak the temps down a little, without being on Roaring manuel. (Must be support for fan control on the card).

With a thermal probe, and some Other cards (not that one) I did some loose testing, and often the rams are hotter than the GPU, depending entirely on the cards cooling. The VRM, the voltage regulators which can have a thermal overload safety, they can be about as hot as the ram. that is why. It is very odd to me to observe that we try and keep the CPUs around say 60-70*C , then put the $$$$ video cards at 80-90*C, then often ignore completely the Power regulations, when all this stuff is silicon and transister gates.

If Overdoing the cooling, or even underclocking it a bit, does Not fix it, then it probably is some other issues for sure. If a bit of cooling helps, then you might also adjust the computer Cases Cool air intake to it.

  • I guess ATI Tray only works for ATI cards, and I have trouble finding GpuTool. The only one I find is the Community Technology Preview 1, and I can't find any mention of artifact testing on it. I'll try to underclock the card a bit later, it might do the trick. – Timst Mar 6 '12 at 18:18
  • @timst Yes that beta thing is the one they are using (community tech preview) . people were pretty happy with it, so it ended there? It worked good for me. – Psycogeek Mar 6 '12 at 18:31
0

For testing the GPUs memory I recently stumpled upon this: http://sourceforge.net/projects/cudagpumemtest

  • This seems like the best answer, but I can't figure out how to compile this for Windows. Have you seen a binary? – Louis Mar 13 '13 at 3:44
  • Nope, but you could try to use mingw, which has a similar toolchain as Linux. – math Mar 13 '13 at 11:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.