GpuDeviceCount fails to report the correct number of GPU devices (8, Tesla K80), how can I resolve this?

3 次查看（过去 30 天）

Daniel Moss 2017-12-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/374260-gpudevicecount-fails-to-report-the-correct-number-of-gpu-devices-8-tesla-k80-how-can-i-resolve-t

评论： Joss Knight 2017-12-26

I am having a problem with a 8 GPU Nvidia Tesla K80 system. It is a shared resource between multiple users which log on to the same machine. I have two other versions of Matlab successfully running on two of the gpus of the K80 system but when I run gpuDeviceCount in a freshly opened version of Matlab ...I am getting an error... looking inside gpuDeviceCount at the error that parallel.gpu.GPUDevice.count gives: parallel:gpu:device:UnknownCUDAError An unexpected error occurred during CUDA execution. The CUDA error was:CUDA_ERROR_OUT_OF_MEMORY

Does anyone know what could be occuring and how I can resolve this?

System: Windows Multipoint Server 2012 Nvidia driver ver: 23.21.13.8819 Intel Xeon 2698 Tesla K80 - 8 GPUS

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Joss Knight 2017-12-26

You should contact tech support about this issue since it sounds like your system is incorrectly configured and/or installed. They can give you some basic test code to run to make sure that the CUDA driver is functioning and can see your cards.

请先登录，再进行评论。

请先登录，再回答此问题。