Hi,
On Fri, Dec 02, 2022 at 01:16:37PM +0100, Vlad Cojocaru via AMBER wrote:
> I am trying to build Amber 22 with cuda support at an HPC site and I an
> getting a cmake error that CUDA 11.7.0 is incompatible with GCC 11.3.0. I
> don't have any control over the local settings and library installations at
> the HPC site. The local support team told me that they have compiled many
> software packages (including GROMACS) with this combination of CUDA and GCC.
> So they suggested this is a specific AMBER issue.
>
> I also looked at the CUDA compatibility and it does seem that CUDA 11.7.0 is
> compatible with GCC 11 in general but I haven't found precisely GCC 11.3.0
> on that list
>
> I should also say that I have built Amber22 with CUDA 11.8.0 and GCC 11.2.1
> on different local Linux workstations without any issue whatsoever.
>
> Has anyone seen a similar issue with this CUDA/GCC combination and has any
> idea if it can be fixed somehow from within Amber (without the need to
> install a new CUDA or use another GCC) ?
The standard document
https://stackoverflow.com/questions/6622454/cuda-incompatible-with-my-gcc-version
shows that they are compatible, but this may be a precision issue as you suggest.
What happens if you change amber22_src/cmake/CudaConfig.cmake line 86 from
AND CUDA_VERSION VERSION_LESS_EQUAL 11.7 )
to
AND CUDA_VERSION VERSION_LESS_EQUAL 11.7.0 )
?
For above and for CUDA 11.8.0, the checking code was written before those
versions existed.
thanks,
scott
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Dec 02 2022 - 12:30:03 PST
