Hi all,
My team and I got two nvidia GPUs recently so we think we might use them for program acceleration. We use the dealii container as our dev env and to keep its strength, we use nvidia-docker2 to provide a runtime cuda support within containers.
The command nvidia-smi shows the detection of GPUs while compiling for step-64 as a test, we occur the errors below:
dealii@d2409a2b5069:~/step-64$ cmake .
CMake Error at CMakeLists.txt:41 (MESSAGE):
Error! This tutorial requires a deal.II library that was configured with
the following options:
DEAL_II_WITH_MPI = ON
DEAL_II_WITH_P4EST = ON
DEAL_II_WITH_CUDA = ON
However, the deal.II library found at /usr/local was configured with these
options
DEAL_II_WITH_MPI = ON
DEAL_II_WITH_P4EST = ON
DEAL_II_WITH_CUDA = OFF
which conflict with the requirements.
-- Configuring incomplete, errors occurred!
which indicates the original dealii library didn't enable CUDA supported flag.
So is there a possible way to enable the flag without rebuild and install the whole library or does the dealii team has any plans to release an extra version based on nvidia/cuda image?
It is also possible to run cuda toolkit directly without installing it within the container for
singularity (sorry I didn't find better articles) and
nvidia itself, which is worth trying yet not fully matured.
In all,
do we have better and simpler solutions for GPU supported programming with dealii? Any comments are welcome!
Regards,
Chengjiang Yin