ABINIT Discussion Forums

Hello, I installed ABINIT successfully with GPU flags and it is running smoothly too. I am having two NVIDIA Tesla C2050 cards in my machine . when I run abinit it shows running on only one card , the other card is free when I run it. My question is how can I run / divide the processing/computing between the two cards.

At the moment you can only use one GPU per CPU. So if you have 2 GPUs, you should use 2 MPI processes. Each one should use one GPU.

Good luck

Jordan

Jordan wrote:At the moment you can only use one GPU per CPU. So if you have 2 GPUs, you should use 2 MPI processes. Each one should use one GPU.

Good luck

Jordan

Hi Jordan,

I have 2 GPU cards and I use 2 MPI processes to run the program, but it used only one GPU too. May I know if there is any configuration options to enable multiple GPUs? Thanks.

Shaohuai

Hi,

I'm not quite sure, but I think that if there are 2 GPUs on one node, abinit checks which one is the fastest and only uses this one. If there are equivalent, I don't know what happens.
So maybe the 2 MPI processes want to use the same GPU so only one is used.

I'll try to ask someone who knows more than me about GPUs

Cheers

Hey all,

we have managed to compile gpu with openmp on a XCray machine.
Is there any benchmark for abinit 8.2 we can use to test our compilation?

thanks a lot
m.