Will altair EDEM 2025 run in distributed mode on HPC?
Hello
I would like to run Altair EDIM on a Linux system in distributed mode.
I could not run 2022 in distributed mode. but
Now, in the updated version, can anyone please tell me if EDEM2025 can run in distributed mode?
(what i mean is when i submit a job should it be in one host only or not)
With the limitation of needing 1 host it has always been very hard to run edem models on HPC as i cannot get a host more than 32 easily. i need to wait for long time. If it were to run in distributed mode, it would make things easier.
Please let me know.
Find more posts tagged with
I am using gpu right now with CUDA on local machine. suppose i want to put it on hpc i dont know how you can select multi gpu do i have to do it on the gui?
i know in batch command you can select gpu. but again the wait time to run one simulation will be a lot.
Is it possible to run in distributed mode on hpc with gpu?
Altair EmployeeGPU like CPU is only shared memory not distributed, however if you have multiple GPU's on one node of a HPC you can choose to run using one of the GPU's or all of them.
For example if you have 4 GPU's on one node these will have device ID's of 0-3. When running command line you use -E 2 for GPU CUDA and -D for the GPU device.
edem -c -i mysim.dem -E 2 -D 0
The above runs on one GPU (device 0)
edem -c -i mysim.dem -E 2 -D 3
The above runs on one GPU (device 3)
edem -c -i mysim.dem -E 2 -D 0+1+2
The above runs on three GPU's (device 0, 1 and 2)
Hello Stephen, So are you trying to say that if i request one host on hpc then i request multiple gpu (not sure if i can request GPUs from other computer) you are saying i can run a distributed job of edem case file?
Hello Stephen,
I am able to initiate a simulation on hpc with single GPU but i am not sure how to call multiple GPU
is this a way to call multiple GPU for EDEM in linux(cent OS) this script should call 4 gpu (num=4) in 1 node (n=1) please let me know
#!/bin/bash
#BSUB -W 30
#BSUB -q gpu
#BSUB -gpu "num=4:mode=shared:mps=no"
#BSUB -o out.%J
#BSUB -e err.%J
#BSUB -n 1
#BSUB -J Edem1
edem -c -i simpleEDEMcase.dem -E 2 -D 0+1+2+3
Altair EmployeeHi Satri,
That looks correct to me, I'm not familiar with the BSUB commands but the EDEM command you submit at the end, with -D 0+1+2+3 is the correct command.
Just to be clear the GPU's have to all be run on the same computer, you can't have distributed GPU across different computers. Its multi-GPU in shared memory system same as multi-CPU in shared memory system.
Regards
Stephen
Ohh I see that is bad. Wait time is a lot of I have to run it on one machine. Why do you not have distributed mode?
I can't find a single computer in library that has multiple GPU in one Comp.
Is there any plans for doing distributed mode in 2025.2 ?
Hi,
EDEM Currently runs on one node only, distributed memory isn't supported.
If you are looking for speed improvements GPU or multi-GPU is the best approach
Regards
Stephen