Have just started trying out bminer and have found it definitely squeezes out some additional hashrate.

One request would be to honor CUDA_VISIBLE_DEVICES, or to allow "-devices" to take the GPU's UUID.

I use some scripts to manage the miner processes and have mappings to what miners should be running on which GPUs. One annoyances with the index based settings of "0,1,2" is that the ID used does not match the ID from nvidia-smi. I found this the hard way and it isn't well documented. I could find a simple way to find out what the order was, but did find that CUDA_VISIBLE_DEVICES supports a comma-separated list of the GPU's UUID, so that was my solution.

With bminer, I had the env var set to the UUIDs, but it was trying to use all GPUs. When I used -devices, it was the same thing of IDs not matching the ordering in nvidia-smi or the order the NVML API returns them in.

Thanks!