We have a few computer (i7-6700) with gtx 1070. Our goal is to create a cluster that can utilize the gpu for calculations, particularly VASP (material science software).
From what I have researched:
1. I would be using ROCKS cluster v6.2. v7 is available but it is still new and I'm a bit afraid of possible bugs.
2. I would be using either MVAPICH2 or OPENMPI for a cuda(gpu) -aware MPI to utilize the gpu. Currently I am not that much familiar with the ports to be used by the MVAPICH, but I do know that we only have a ethernet cable (no infiniband or whatever). I might be mistaken with this. Also I'm not that aware on which version of the MPI to choose.
3. I would need a gpu-aware job manager. From some forum I have seen them recommending the use of SLURM.
If you have any experience on a similar build please do tell if you have comments. Your experiences would be deeply appreciated.
Thank you for your time.
Check these 2 links :
https://www.ibm.com/developerworks/linux/library/l-cluster2/
https://help.ubuntu.com/community/MpichCluster
Dear Mr. Mohamed,
The links you sent were for a non-cuda aware MPI, however, would the flow of installation be similar to a cuda-aware MPI?
Thank you.
cuda-aware MPI recommendation from NVIDIA: https://devblogs.nvidia.com/introduction-cuda-aware-mpi/
try to read this first :
https://devblogs.nvidia.com/how-build-gpu-accelerated-research-cluster/
before reading what you have on the CUDA-aware MPI page.
hope it will help you!
this link is useful
https://help.ubuntu.com/community/MpichCluster regards
Check this link
https://devblogs.nvidia.com/how-build-gpu-accelerated-research-cluster/
regards
- Badges
- Science topic
- Similar topics
- Names