Luminal launches an AMD backend for production PyTorch workloads
The compiler can run FLUX.2 diffusion workloads on AMD MI300X GPUs, targeting lower cost per image from cheaper compute.
Read the source ↗
Luminal says its compiler can port production PyTorch code to AMD hardware. The company is running large-scale FLUX.2 diffusion workloads on MI300X GPUs and has published an engineering write-up on reducing cost per image.