TensorRT version Katago can not utilitize multi-gpus when selfplaying

[Grant-Tao]

the TensorRT version is really powerful, it can recognize multi-gpus when playing. But, Whatever using selfplay1.cfg or selfplay2.cfg, the TensorRT version can only recognize device 0 when selfplaying. The CUDA version works very well on multi-gpus for selfplay. Maybe there is a bug.

[lightvector]

Did you try adding the trtDeviceToUseThread0, etc. in the selfplay cfgs?