Errors regarding the memory allocations

[manoj_wijesingha]

Hi Colleagues,
When I am running the VASP (6.5.1) simulations for the optimization step, it generates errors regarding the memory allocations. This system has 172 atoms. I have attached the files, and the SLURM file is included the memory allocation code like "#SBATCH --mem=768G" or default value "#SBATCH --mem=0". Could please consider this request regarding the memory errors running the VASP codes.
Thank you.

Best Regards,
Manoj.

[jonathan_lahnsteiner2]

Dear Manoj,

This looks like you are running out of memory. Could you please check your OUTCAR file for the lines mentioned in this wiki article and compute the total memory per core. You can compare this value to the maximum available memory on your compute cluster. Please supply the asked information, without I am not able to help further.

All the best Jonathan

[manoj_wijesingha]

Hi Jonathan,
Thank you so much for your prompt response. I checked the "total amount of memory used by VASP" in OUTCAR file, it mentioned that "96505 kBytes". Normally I use "#SBATCH --mem=768GB" or "#SBATCH --mem=0 (default)" per node. The memory allocation error happens either way. For this calculation I used #SBATCH --mem=0 for 640 cores. So, could you please suggest which values is the best to keep for memory allocation (right now, I have put it as #SBATCH --mem-per-cpu=2G). I have attached the files.
Thank you.

Best regards,
Manoj.

[jonathan_lahnsteiner2]

Dear manoj wijesingha,

This does not seem to be an error related to VASP. What you could try is to set

Code: Select all

UCX_IB_RCACHE_MAX_REGIONS=500

as recommend on this page. Or even try larger values for this environment variable.
Another environment variable you could try to set is

Code: Select all

export UCX_IB_REG_METHODS=direct

as recommended here. Setting UCX_IB_REG_METHODS might slow down the performance of the code.

It might also be a good idea to check if the memory used during your calculation increases with the number of calculation steps. This might hint to a memory leak.
It might also be advisable to contact your system administrators because of this issue because it might be related to the used mpi version.

All the best Jonathan

[manoj_wijesingha]

Hi Jonathan,
Thanks for your guidance and appreciate it.

Best Regards,
Manoj.