Should training Distributed with bs=x/gpus be equivalent?

[stas (Stas Bekman)]

Or use the parent process id: os.getppid()