ComputationalRadiationPhysics / cuda_memtest

Fork of CUDA GPU memtest :eyeglasses:
http://sourceforge.net/projects/cudagpumemtest
110 stars 31 forks source link

Fix segfaults on >8 GPU systems #23

Closed grische closed 4 years ago

grische commented 4 years ago

If a system has more than 8 GPUs, the program segfaults on some of the tests without showing an obvious problem.

This PR adds an early exit for such a scenario to avoid random segfaults. It also increases the limit to a more reasonable number for current systems. The impact on the memory footprint should be minimal.

ax3l commented 4 years ago

Cool, I want a system with more than 8 GPUs per host, too :) Thank you for the fix!

grische commented 4 years ago

@ax3l if you want, we can arrange something for you there. Just sign up in the following link and drop me an email! https://account.genesiscloud.com/signup

ax3l commented 4 years ago

Thanks for the offer, I appreciate it but cannot accept this and have access to HPC machines :) Let's get your PR merged now, as soon as CI passes. We just have some more jobs queued up in Travis right now, will revisit this later today.