Wiki‎ > ‎

GPU stress test

posted May 25, 2016, 3:48 PM by Dong Xu   [ updated Aug 9, 2016, 10:52 AM ]
gpu_burn-0.6.tar.gz (compatible with nvidia-smi and nvcc as of 04-12-2015)

This is the best GPU test along with megadock.
Note: must run the program in the compilation directory. /garlic/apps/Downloads/gpu_burn-0.6/

http://wili.cc/blog/gpu-burn.html

http://cuda-z.sourceforge.net/

FAHBench

https://folding.stanford.edu/home/download-utilities/

FAHBench is the official Folding@home GPU benchmark that measures the compute performance of GPUs. Please see the FAHBench home page for download links and support information.

FAHBench 1.2 is obsolete but available for download here.

For More Information, Please See:

StressCPU v2.0 – Gromacs Based CPU Stability Tester

StressCPU is a CPU stress tester, based on the Gromacs code found in the Folding@home (FAH) fahcores that process work units. StressCPU stresses all of the processor cores on a computer to help verify system stability, and is one of the best testing tools available, pushing CPUs harder and hotter than old school favorites like Prime95.

v2.0 is an updated release, now supporting both ia32 (32 bit) as well as x86-64/em64t (64 bit) platforms. It is multithreaded (both pthreads and win32 threads) by default and automatically senses the number of CPUs on Linux, Mac OS X, and Windows. It runs slightly hotter, in particular for x86-64 systems, the checks are better, and you can now set it for a fixed excution time, e.g. 12 hours. The package includes pre-compiled binaries for Windows, 32 and 64 bit Linux, and 32 or 64 bit OS X.

StressCPU v2.0 is available to download directly from the Gromacs.org web page: stresscpu2.tgz

There is also a discussion thread and help topic for StressCPU v2.0 on the Folding Forum.

 

MemtestG80 and MemtestCL – Memory Testers for CUDA and OpenCL GPUs and CPUs

One of the members of the Folding@home team, Imran Haque, has developed a pair of memory testers for GPUs. Here’s a brief description:

MemtestG80 and MemtestCL are software-based testers to test for “soft errors” in GPU memory or logic for NVIDIA CUDA-enabled GPUs (MemtestG80) or OpenCL-enabled CPUs and GPUs by any manufacturer, including both ATI and NVIDIA (MemtestCL). They use a variety of proven test patterns (some custom and some based on Memtest86) to verify the correct operation of GPU memory and logic. They are useful tools to ensure that given GPUs do not produce “silent errors” which may corrupt the results of a computation without triggering an overt error.

Basically, the idea is that we wanted to put out a code to test GPU memory that’s roughly equivalent to Memtest86 on CPUs. If you run FAH heavily on a GPU, it’s a good idea to check out your GPU memory, just as one would run tests on CPU memory. MemtestG80 will run on any NVIDIA GPU with CUDA support; MemtestCL will run on both NVIDIA and ATI OpenCL-capable GPUs, as well as on CPUs with the AMD Stream SDK OpenCL runtime.

The source code for both programs is available under the LGPL license and is hosted at SimTK, the Stanford scientific software repository. It can be downloaded at https://simtk.org/home/memtest. Binaries are available both there and on this page.

If MemtestG80/MemtestCL detect memory errors on your GPU, we recommend taking the following steps:

  • If your card generates errors and is overclocked (this includes “factory overclocked” or “superclocked” cards – anything with higher-than-reference clock speeds), reset the clock frequencies to the NVIDIA/ATI reference frequencies and see if the problem persists.
    This is especially true for the memory clock. Errors in the Logic or Random Blocks tests are likely to be at least somewhat sensitive to the shader clock as well. The upshot of this guideline is that if your overclock is generating any errors above the stock frequencies, then it’s not a stable overclock.
  • If after this your card still generates errors in any test OTHER than the Modulo-20 test, these errors are likely indicative of a card that’s gone bad somehow. Such a card ought to be replaced.

We used MemtestG80 to collect data for paper #86, “Hard Data on Soft Errors: A Large-Scale Assessment of Real-World Error Rates in GPGPU”. However, we are no longer collecting MemtestG80′s information, so there’s no need to try to upload the results.


Comments