[CP2K-user] [CP2K:22455] Re: Unable to achieve speed boost on GPU nodes, exhibit same performance as CPU nodes.

Frederick Stein f.stein at hzdr.de
Tue Nov 3 13:54:36 UTC 2026


I would try different run configurations, especially less OpenMP threads or 
more MPI ranks. CP2K's OpenMP-parallelization is usually not very good and 
I am not sure whether CP2K scales up to 24 threads per rank. You should 
also double-check whether you have a cuda-version of COSMA and link against 
that.
Rohit Manjunath Kudtarkar schrieb am Dienstag, 3. November 2026 um 12:56:39 
UTC+1:

> Dear Frederick,
>
> Thanks for the prompt reply. Please find the attached files for your 
> perusal. Suggest a suitable header setting for my system.
> GPU: NVIDIA A100 80GB PCIe.
>
>
> On Tuesday, November 3, 2026 at 3:47:27 PM UTC+5:30 Frederick Stein wrote:
>
> Dear Rohit,
> Can you provide more details on your hardware (kind of GPU), how you 
> compiled CP2K (incl. log files), how you run CP2K and some example output 
> file, especially the header with information on the compilation and the 
> last sections on statistics of the different libraries.
> Best,
> Frederick
>
> Rohit Manjunath Kudtarkar schrieb am Dienstag, 3. November 2026 um 
> 11:13:51 UTC+1:
>
> Dear all,
>
> I am working on a sulfuric acid-water mixture with 700+ atoms for testing 
> on the CPU and GPU; however, I am not able to achieve the speed boost on 
> the GPU node. The GPU node running on 48 cores works the same way as the 
> pure CPU run on 48 cores. I have executed these tests with several solver 
> packages for GPU and CPU. And found no improvement as such. Can anyone help 
> me? What's the issue? Or is the density matrix sparse? The basis set is 
> DZVP, and the functional is PBE.
>
> Rohit Kudtarkar
> PhD IITB.
>
>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/851f9f4e-2a37-4c2e-b458-7b00f1f70f73n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20261103/425a50e9/attachment.htm>


More information about the CP2K-user mailing list