encountered BUG on previously successfully done calculation (in GPU))

[bhargabkakati]

Hello, I have encountered a bug in my GPU-compatible VASP vasp6.4. I had run this calculation (same inputs) yesterday using the same GPU, but when today I encountered the bug in some different calculations, I checked if this previously successful calculation was executed or not, but I encountered the same bug (note : the calculation runs fine on CPU). I have attached the inputs and bug report. Thank you.

[jonathan_lahnsteiner2]

Dear bhargabkakati,

The reason for the error you are encountering is that you only have a single GPU but trying to run with 2 threads. This is not supported by NCCL. In the latest VASP version the code is able to catch this problem.
The code will inform you of the problem with the following message

Code: Select all

 
 WARNING: INIT_OFFLOAD: several MPI-ranks need to share a GPU, which is not
  supported by NCCL. The use of NCCL will be switched off. To avoid this,
 reduce the number of MPI-ranks: #-of-ranks <= #-of-GPUs (on every node!).

So I would recommend to update to the latest VASP version 6.5.1. Or you have to keep in mind to not set the number of threads higher than the number of available GPUS. For further information I would also like to point you to this forum thread.

All the best Jonathan

[bhargabkakati]

Thank you.