using several GPUs on a single workstation

[errinaceus]

Is there a way to use two GPUs on a single workstation, not on a cluster?
export CUDA_VISIBLE_DEVICES doesn't work, neither does AlphaFold's flag --gpu_devices

[kashyapchhatbar]

Here's what we've adapted in our workflow. Create a Docker image with AlphaPulldown configured and fire up one container with --gpu device=0 and another with --gpu device=1. Split your custom_mode.txt in two parts and let the containers run parallel.