<div>Dear Jonas,</div><div>AFAICT, CUDA should not be the issue. I am more worried about the NVIDIA toolkit as it is AFAIK not yet possible to build CP2K with NVIDIA HPC SDK and the module files created by GCC are not necessarily compatible with the ones from the NVIDIA compilers. I am not sure how much of CP2K can be compiled and runs flawlessly with the current or even the next version.</div><div>Best,</div><div>Frederick</div><br /><div class="gmail_quote"><div dir="auto" class="gmail_attr">jona...@hotmail.com schrieb am Montag, 28. Juli 2025 um 20:57:39 UTC+2:<br/></div><blockquote class="gmail_quote" style="margin: 0 0 0 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">Or maybe my assumption are wrong. <div>Is it possible to build Cp2k with standard cuda-12.8 and afterwards build and link Gromacs with NVIDIA HPC SDK?<br><br></div><div class="gmail_quote"><div dir="auto" class="gmail_attr">mandag 28. juli 2025 kl. 20:52:36 UTC+2 skrev <a href data-email-masked rel="nofollow">jona...@hotmail.com</a>:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">I'm trying to build cp2k with NVIDIA HPC SDK using this toolchain command:" <div>./install_cp2k_toolchain.sh --install-all --with-tblite=no --with-ace=no --with-gcc=no --math-mode=openblas --mpi-mode=openmpi --enable-cuda --gpu-ver=A100 --target-cpu=znver2 -j 32". </div><div>Everything installs well except ELPA. ELPA's configure script has an internal flag:"with_cuda_sdk_path" which should detect the SDK from my build environment:" export MODULEPATH=/opt/nvidia/hpc_sdk/modulefiles:$MODULEPATH<br>module load nvhpc/25.3" via the toolchain --enable-cuda. </div><div><br></div><div>However, it does not find cublas as shown in the following error:" checking for library containing cublasDgemm... no<br>configure: error: Could not link cublas; try to set the cuda-path or disable Nvidia GPU support". <div><br></div><div>The only reason for this as I can see is that ELPA does not know to look inside math_libs/lib64, which is wired as it should for SDK. I have confirmed that libcublas.so are present there and not in cuda/lib64, and have confirmed that my build environment are pointing to both the lib64 files. </div><div><br></div><div>Would appreciate all the help I can get to solve this! :) </div><div><br></div><div>Best regards,</div><div>Jonas Båtnes</div><div><br></div></div></blockquote></div></blockquote></div>
<p></p>
-- <br />
You received this message because you are subscribed to the Google Groups "cp2k" group.<br />
To unsubscribe from this group and stop receiving emails from it, send an email to <a href="mailto:cp2k+unsubscribe@googlegroups.com">cp2k+unsubscribe@googlegroups.com</a>.<br />
To view this discussion visit <a href="https://groups.google.com/d/msgid/cp2k/17ed4356-7db7-4fc9-89c7-79589fc1bae0n%40googlegroups.com?utm_medium=email&utm_source=footer">https://groups.google.com/d/msgid/cp2k/17ed4356-7db7-4fc9-89c7-79589fc1bae0n%40googlegroups.com</a>.<br />