[QE-users] QE-GPU - cudafor not found

[R T Phillips]

Dear Pietro, and also Paolo

Many thanks for the swift response.

--with-cuda-runtime=12.6   ... this leads to success in config.

make all    seemed to complete successfully.

Running pw with mpirun ... so far no good.

I would be very grateful if you have any suggestion for the right 
options to select to run pw in the present configuration. I seem to have 
omitted specifying an hca device, and have hcoll errors.

Many thanks, I really appreciate your help.

Richard



On 30/01/2025 11:33, Pietro Davide Delugas wrote:
> Hello
> you need to specifu --with-cuda-runtime=<value>,  where value is a 
> version of the  runtime supported by your compiler, I guess 11.8 
> should do.
> Hope it helps
> Pietro
>
> ------------------------------------------------------------------------
> *From:* users <users-bounces at lists.quantum-espresso.org> on behalf of 
> R T Phillips via users <users at lists.quantum-espresso.org>
> *Sent:* Thursday, January 30, 2025 12:23
> *To:* users at lists.quantum-espresso.org <users at lists.quantum-espresso.org>
> *Subject:* [QE-users] QE-GPU - cudafor not found
> I have been trying to compile Quantum Espresso for use on this system:
> Intel i9-14900 (24 core)
> Linux Mint 22.1 Cinnamon version 6.4.6
> Nvidia GeForce RTX 4070
>
> with:
> nvhpc_2024_2411_Linux_x86_64_cuda_12.6
>
> openmpi5.0.6
>
> qe 7.3.1
>
> PATH and LD_LIBRARY_PATH point to all the recommended folders.
>
> Compilers set to nvc, nvc++, nvfortran from nvhpc; mpif90 from openmpi
> also tried.
>
> ./configure  --enable-parallel --with-cuda=$NVHPC_CUDA_HOME
> --with-cuda-cc=89 --with-cuda-mpi=yes --enable-openmp
>
> this stops with:
>
> checking whether Fortran compiler accepts -cuda -gpu=cuda10.1... no
> configure: error: You do not have the cudafor module. Are you using
> NVHPC compiler?
>
> ... well, yes. What then?
>
> I have searched for info about whether the nvhpc should include cudafor,
> and if so, how to make it visible. There are discussions related to
> distributions prior to hpc_sdk, but I can't find up-to-date advice ie
> after cuda was incorporated in the hpc_sdk. I switched to F90=mpif90
> from openmpi-5.0.6 and hit the same problem.
> Is there a solution to this? Clearly I want to use all the cores in the
> CPU and those in the GPU to optimise throughput.
> Thanks to anyone who can help!
>
> Richard Phillips, Cavendish Laboratory, UK
>
> _______________________________________________________________________________
> The Quantum ESPRESSO Foundation stands in solidarity with all 
> civilians worldwide who are victims of terrorism, military aggression, 
> and indiscriminate warfare.
> --------------------------------------------------------------------------------
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu 
> <http://www.max-centre.eu>)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users 
> <https://lists.quantum-espresso.org/mailman/listinfo/users>
>
> ------------------------------------------------------------------------
> *From:* users <users-bounces at lists.quantum-espresso.org> on behalf of 
> R T Phillips via users <users at lists.quantum-espresso.org>
> *Sent:* Thursday, January 30, 2025 12:23
> *To:* users at lists.quantum-espresso.org <users at lists.quantum-espresso.org>
> *Subject:* [QE-users] QE-GPU - cudafor not found
> I have been trying to compile Quantum Espresso for use on this system:
> Intel i9-14900 (24 core)
> Linux Mint 22.1 Cinnamon version 6.4.6
> Nvidia GeForce RTX 4070
>
> with:
> nvhpc_2024_2411_Linux_x86_64_cuda_12.6
>
> openmpi5.0.6
>
> qe 7.3.1
>
> PATH and LD_LIBRARY_PATH point to all the recommended folders.
>
> Compilers set to nvc, nvc++, nvfortran from nvhpc; mpif90 from openmpi
> also tried.
>
> ./configure  --enable-parallel --with-cuda=$NVHPC_CUDA_HOME
> --with-cuda-cc=89 --with-cuda-mpi=yes --enable-openmp
>
> this stops with:
>
> checking whether Fortran compiler accepts -cuda -gpu=cuda10.1... no
> configure: error: You do not have the cudafor module. Are you using
> NVHPC compiler?
>
> ... well, yes. What then?
>
> I have searched for info about whether the nvhpc should include cudafor,
> and if so, how to make it visible. There are discussions related to
> distributions prior to hpc_sdk, but I can't find up-to-date advice ie
> after cuda was incorporated in the hpc_sdk. I switched to F90=mpif90
> from openmpi-5.0.6 and hit the same problem.
> Is there a solution to this? Clearly I want to use all the cores in the
> CPU and those in the GPU to optimise throughput.
> Thanks to anyone who can help!
>
> Richard Phillips, Cavendish Laboratory, UK
>
> _______________________________________________________________________________
> The Quantum ESPRESSO Foundation stands in solidarity with all 
> civilians worldwide who are victims of terrorism, military aggression, 
> and indiscriminate warfare.
> --------------------------------------------------------------------------------
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu 
> <http://www.max-centre.eu>)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users 
> <https://lists.quantum-espresso.org/mailman/listinfo/users>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20250130/d704b7d5/attachment.html>