To clarify this. On my RX5700 XT, with ROCm, Stable Diffusion generates images on the default sampler at 512x512 20 passes in about 2 seconds. It's definitively slower than a similar NVIDIA card, but I've no concrete numbers.
It's still a pain in other ways. AMD does not support ROCm on consumer-grade GPUs, it only "happens to work". I had to explicitly install the 3.10 python interpreter and use an older torch version, because seemingly, the newer ROCm releases do not support the RX5700 XT. Also ROCm only works on Linux.
It's still a pain in other ways. AMD does not support ROCm on consumer-grade GPUs, it only "happens to work". I had to explicitly install the 3.10 python interpreter and use an older torch version, because seemingly, the newer ROCm releases do not support the RX5700 XT. Also ROCm only works on Linux.