I have no experience of using docker. From my understanding of CUDA, there
should be no problem of using GPUMD with any newer card. In the makefile,
the default flag is *-arch=sm_35*, which is still supported by CUDA 11.x
(although there is a warning about its deprecation soon). I still use this
default with my RTX 2070 (compute capability 7.5) card or V100 (compute
capability 7.50) card and there were no problems.
You could try:
1) change this flag to *-arch=sm_80*
2) make sure your CUDA version is CUDA 11.0 or newer
3) make sure the Nvidia driver version is not older than your CUDA version.
You can also compile and run any simple CUDA code to check if your docker
environment (I don't how it is called) works.
If other people have experience with using GPUMD or similar CUDA codes with
docker, please share the knowledge. Thanks!
On Sun, Feb 7, 2021 at 4:34 PM "소순성" <soonsung2001@xxxxxxxxxx> wrote:
Thank you for GPUMD and today, I have questions.
I have an opportunity to use a100 graphic cards so I want to apply GPUMD
With docker environment, I compiled it and tried to use it, but GPUMD did
(with 'top' linux command, 'gpumd' binary appeared but it stopped.)
Is there any restriction to use GPUMD on docker environment or
high-performance graphic cards?