How to run faiss add_index on a selected GPU in a multi GPU system ?

[shamanez]

My training process uses 4 32-GB GPUs and during the training process I want to update the index. There I have add_index function to run. But I want to run only on the 5th GPU. Is there a way that I can select the exact GPU?

Running on:

• CPU
• [X ] GPU

Interface:

• C++
• [ X] Python

[wickedfoo]

Specify the device in the GpuIndexConfig passed to the index constructor. Different GPU indices have different config objects (GpuIndexIVFPQConfig, GpuIndexFlatConfig, etc) but they all have device as a member.

e.g.,

faiss/benchs/kmeans_mnist.py

Line 63 in efa1e3f

cfg.device = i

devices are numbered from 0 (following the CUDA convention).

[shamanez]

@wickedfoo

Thanks a lot. Can you please explain to me what is happening in the this script?

# build a flat (CPU) index
index_flat = faiss.IndexFlatL2(d)
# make it into a gpu index
gpu_index_flat = faiss.index_cpu_to_gpu(res, 0, index_flat)

What is the second parameter? is it GPU index? if not how can use a specific GPU in this case?