Question on GPU Computing

Question

Frankie Camacho 2020-9-20

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/596641-question-on-gpu-computing

评论： Frankie Camacho 2020-9-21

I'm running a low end ASUS GeForce GT 710 graphics card as a secondary card (I have a higher end AMD card that is not supported on MATLAB's platform). I wanted to know if there is any way to completely dedicate this card solely for GPU computing. My sense is that I get CUDA_ERROR_LAUNCH_FAILED and other TDR errors because the operating system wants to dedicate some of the card's resources for displays (although no display is connected to the card). I tried running the parallel computing demo and reached this error:

An unexpected error occurred during CUDA execution. The CUDA

error was:

unspecified launch failure

Error in parallel.internal.gpu.timeit

Error in gputimeit (line 66)

t = parallel.internal.gpu.timeit(f, num_outputs);

Error in paralleldemo_gpu_benchmark (line 170)

mmTimesGPU(ii) = gputimeit(@() A*B);

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Matt J 2020-9-20

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/596641-question-on-gpu-computing#answer_497365

编辑：Matt J 2020-9-20

Make sure you have updated drivers and make sure you restart your computer after their installation. Then use gpuDevice() to select the card you want Matlab to use.

https://www.mathworks.com/help/parallel-computing/parallel.gpu.gpudevice.html

3 个评论
显示 1更早的评论隐藏 1更早的评论

Matt J 2020-9-21

编辑：Matt J 2020-9-21

I'm not sure why you would need Nsight to do that. The TDR parameters can be set through the Windows registry.

Frankie Camacho 2020-9-21

I think you are right, but alas I am not sure how to do this. I suppose expediency played a role here. I feel a little more secure knowing that the NVIDIA options don't affect my AMD card, or at least that's what I tell myself...

请先登录，再进行评论。