E++ compilation in a cluster
almeida85 opened this issue · comments
Hi,
I trying to compile E++ in an HPC cluster. Inspecting the CMakeCache.txt file I saw all the flags and MPI variables that in principle can be used to customize the compilation. My cluster runs the SLURM queue system. I use mpirun for parallelization.
1 - Could anyone give an example of options to compile E++ for a cluster?
2 - Is E++ designed to run in GPUs?
Thanks.
Hi,
-
It mostly depends on the compiler and the cluster architecture, e.g. for Intel compiler I often used
-O2 -xAVX -axCORE-AVX2,CORE-AVX512 -
AFAIK E++ does not utilize GPU at all.
Best, Jakub
Hi,
Thank you for the reply.
I am using C++ compiler (GCC 4.8.5 20150623 (Red Hat 4.8.5-39)), in a cluster with CentOS Linux 7 x86_64.
My cluster has:
1 Frontend node:
– 56 TByte net storage (17 x 4 TByte SATA HDDs, RAID array, level 6)
– 2 CPUs / 24 cores per node (CPU: Intel Xeon Skylake Gold 6126 with 2.6 GHz, 12 cores/24 threads)
– 192 Gbyte RAM per node (8 Gbyte / core)
20 computation nodes:
– 16 nodes: Intel Xeon Gold 6126 with 2x12 Cores,192 GB RAM, 2 TB storage
– 4 nodes: Intel Xeon Gold 6130 with 2x16 Cores, 96 GB RAM, 6,5T storage
... plus 2 GPUs nodes. Which variables are the most appropriate for these settings? I am pretty new compiling customized codes in clusters.
Is CMAKE_CXX_FLAGS_RELEASE:STRING= the variable to insert the corresponding flags?
Best,
Yasser
... I also use mpicxx as MPI C++ wrapper.
I would try simply with -O2 or -O3 and compare the execution times; yes, you can put additional flags in that variable.
Is this resolved?
Reopen if problem still persists.