Can't use GPU for inference.

[Phoenix-Chen-Git]

I used the example command to run RFAA on my cloud server(have an A10 GPU).
However, during diffusion, I used nvtop to visualize the usage of GPU. It shows that the GPU is not working at all.

But why? It's suppose to do inference on GPU.
I have installed right CUDA. And training other neural network is fine with GPU.
Help me, plz.

[yvfan4]

Could you tell me how many amino acids you have and the memory taken by it? I am choosing a gpu for rfdiffusion and MD, so the memory matters. Thanks.

[ikalvet]

Looking at the timestamps in the RFdiffusion output I would say that you are in fact using a GPU for this job. Each step would take much longer than 10 seconds if you were running it on a CPU.
No idea why nvtop doesn't show usage though...

[Phoenix-Chen-Git]

Looking at the timestamps in the RFdiffusion output I would say that you are in fact using a GPU for this job. Each step would take much longer than 10 seconds if you were running it on a CPU. No idea why nvtop doesn't show usage though...

Thanks! I may use other tools to monitor the usage.

[Phoenix-Chen-Git]

Could you tell me how many amino acids you have and the memory taken by it? I am choosing a gpu for rfdiffusion and MD, so the memory matters. Thanks.

I was simply running the example which have 150 ammino acids. I used A10 GPU(24GB VRAM). However, nvtop didn't show the usage of VRAM. So I'm not sure if you can use GPU with fewer VRAM.

[yvfan4]

Could you tell me how many amino acids you have and the memory taken by it? I am choosing a gpu for rfdiffusion and MD, so the memory matters. Thanks.

I was simply running the example which have 150 ammino acids. I used A10 GPU(24GB VRAM). However, nvtop didn't show the usage of VRAM. So I'm not sure if you can use GPU with fewer VRAM.

I think you can check whether the RAM(not VRAM) and CPU occupation increase a lot to confirm whether your equipment is using GPU. If your CPU occupation increases a lot that means the GPU is not working.