AMD xDiT scales diffusion inference on Radeon AI PRO R9700S GPUs

This technical guide explains how to use the open-source xDiT inference engine to scale image and video generation workloads across one, two, or four AMD Radeon AI PRO R9700S GPUs. It details the steps required to configure models like FLUX.2 Klein 4B using FP8 precision, memory-efficient replication, FSDP sharding, and tuned distributed VAE decoding. The author notes that while multi-GPU setups can accommodate larger models and lower latency, independent single-GPU requests often yield higher throughput when models fit within a single card's memory limits. Open article →