This isn't really an Amber issue, it looks more like your GPUs already have
jobs running on them, or perhaps you don't have 8 GPUs set up for that
node. Do you use a scheduling system like slurm? Have you run mpi jobs on
this computer before?
On Sun, Sep 5, 2021, 9:56 AM jinfeng liu <jinfengliu1119.gmail.com> wrote:
> Dear Amber users and developers,
>
> I am trying to run remd simulation by using Amber20 pmemd.cuda_SPFP.MPI on
> multiple gpus. I have 6 replicas, and correspondingly I want to use 6 gpus.
> Unfortunately, I got error messages like the following,
>
>
>
>
>
>
>
>
>
>
>
>
> *+ mpirun -np 6 /home/jf/soft/amber20/bin/pmemd.cuda_SPFP.MPI -ng 6
> -groupfile equilibrate.groupfile Running multipmemd version of pmemd
> Amber20 Total processors = 6 Number of groups = 6cudaMalloc
> GpuBuffer::Allocate failed all CUDA-capable devices are busy or
> unavailablecudaMalloc GpuBuffer::Allocate failed all CUDA-capable devices
> are busy or unavailablecudaMalloc GpuBuffer::Allocate failed all
> CUDA-capable devices are busy or unavailablecudaMalloc GpuBuffer::Allocate
> failed all CUDA-capable devices are busy or unavailablecudaMalloc
> GpuBuffer::Allocate failed all CUDA-capable devices are busy or
> unavailable*
>
> Could anyone help me to fix this problem? Thanks in advance !
>
> Best regards,
> Jinfeng
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Sep 05 2021 - 09:00:02 PDT
