Use only one GPU with Python

[Emvlt]

What I want to do:
Use only one GPU when running TIGRE AND calling its by it's ID (my two GPUs have the same name)

What I understand:
When instantiating a reconstruction algorithm, gpuids=None by defauls and I don't see that it can be changed by passing gpuids as a kwarg.

Example:
iterative_recon_alg.py

• no gpuids in keyworded argumens
• gpuids = None (line 160) which by default grabs all gpus

OS : Linux
Python version : 3.6
CUDA : 10.1

[tsadakane]

@Emvlt Hi.
If you want to use the first GPU,

gpuids = gpu.getGpuIds()
gpuids.devices = [0]  # [1] if you use the second one.
proj = tigre.Ax(head, geo, angles, gpuids=gpuids)

You can find more info in example.py and d22_ListGPUs.py

[Emvlt]

Thanks a lot!
I tried to pass the GPU id as an integer to Ax but code memory error, didn't think about passing the object :)

[epapoutsellis]

I also followed the above python scripts for an fdk reconstruction but when I run

gpuids = gpu.getGpuIds()
img = tigre.algorithms.fdk(stack_img, geo, angles, gpuids = gpuids)

it does not seem (nvidia-smi, task manager) that both my two gpus are utilized.

OS : Windows
Python version : 3.9
CUDA : 11.0

Any ideas?

[Emvlt]

Have you tried adding
gpuids.devices = [0] or gpuids.devices = [1] for device selection?

[AnderBiguri]

I think this was because for very large images, most time is used in filtering in FDK, which is not GPU-based yet (albeit #423 exists, and I think its working), @Emvlt