Re: [AMBER] Fixed the problem of AMBER20 run with pmemd.cuda with CUDA11.1 and 3080 GPU card

From: David A Case <david.case.rutgers.edu>
Date: Thu, 31 Dec 2020 21:41:02 -0500

On Thu, Dec 31, 2020, liu kai wrote:
>
>Solution:
>The modification in step2 is not enough. Because the gencode is only support to sm75.
>While The 3080 card requires sm86.
>So, I added two lines, highlighted below.
>
> if [ "$cudaversion" = "11.0" -o "$cudaversion" = "11.1" ]; then
> # Implement the standard, not warp synchronous, compilation
> sm86flags='-gencode arch=compute_86,code=sm_86' <======
> sm80flags='-gencode arch=compute_80,code=sm_80' <======
> sm75flags='-gencode arch=compute_75,code=sm_75'
> sm70flags='-gencode arch=compute_70,code=sm_70'
> echo "CUDA Version $cudaversion detected"
> echo "Configuring for SM5.0, SM5.2, SM5.3, SM6.0, SM6.1, SM7.0, SM7.5 and SM8.0"
> nvccflags="$sm50flags $sm52flags $sm53flags $sm60flags $sm61flags $sm70flags $sm75flags $sm80flags $sm86flags -Wno-deprecated-declarations"
>

Thanks for pointing out this omission. I'm preparing an update that will
add sm80flags to the legacy build. But users should really be trying their
best to use CMake -- we can't keep up the effort to support two separate
build systems for very long.

Per Ross, I'm inclined to leave out the sm86flags. Do you have definitive
evidence for your statement that "The 3080 card requires sm86"? Or will
sm80 also work for you?

[Aside: above has the very common problem that the comments (in the echo
command) don't match the code itself. But it's not like we all haven't done
that....]

Thanks again for your help on this.....dac


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Dec 31 2020 - 19:00:02 PST
Custom Search