Thanks for sharing these.
For point 2. Yes, Tesla A100 has a compute capability of 8.0 and you cannot
use a higher one.
On Thu, Apr 29, 2021 at 3:31 PM "소순성" <soonsung2001@xxxxxxxxxx> wrote:
I tested the GPUMD on A100 cards with docker environments
and I found that
1. the required version of nvidia-driver and cuda didn't be compatible
with A100 cards
So I made docker image for A100 cards (nvidia-driver:ver.450 and
cuda:ver.11.2.1) and it works now.
2. Choosing gencode for A100 card (sm_80 or sm_86), in my case, is sm_80
This is very important because with "sm_86", in my case, it caused error
with the phrase (no kernel image is available for execution on the device)
--------- Original Mail ---------Sender : "소순성" <soonsung2001@xxxxxxxxxx>
Recipient : <gpumd@xxxxxxxxxxxxx>
Received Date : 2021/Mar/8(Mon) 13:09:45
Subject : [gpumd] About compiling GPUMD
Have a good day and I have a question today.
I compiled GPUMD sucessfully in new environment without error but I faced
the failure to perform it when I tried.
It showed as the picture I attached.
(I added gpumd binary path to PATH)
Error text: Reading error for number of inputs.
I tried with command of both 'gpumd < input' and just 'gpumd' but it
showed same error and didn't work.
How can I deal with it?