What changed

AMD announced support for multi-GPU diffusion inference using xDiT, an open-source inference engine, on its Radeon AI PRO R9700S GPUs. A four-GPU system can run large image and video generation workloads such as FLUX.2, Qwen-Image, Wan 2.2, Stable Diffusion 3.5, and Z-Image. The system uses distributed memory and reduces PCIe communication to improve scaling.

What this means in practice

The Radeon AI PRO R9700S GPU combines 32GB of memory with RDNA 4 compute. A four-GPU system has 128GB of aggregate GPU memory, enabling the execution of large models. Users can split requests across GPUs to reduce latency or run independent requests for higher throughput. The xDiT engine provides tuned RDNA 4 configurations for diffusion models, including native FP8 execution and optimized VAE decoding.

Limitations and availability

The feature is available for Linux systems with one, two, or four Radeon AI PRO R9700S GPUs and a host AMD GPU driver that supports ROCm 10.0. The xDiT model runner is packaged in a Docker container and requires Docker, a Hugging Face account, and sufficient local storage. The supplied announcement does not specify exact performance benchmarks or independent testing results. The feature is currently in preview and may require further configuration for optimal results.

Source facts for AMD Enables Multi-GPU Diffusion Inference on RDNA 4 GPUs: How to Scale Diffusion Inference Across AMD RDNA 4 GPUs# November 02, 2026 by Paul dos Santos. 11 min read. | 2645 total words. It explains when multiple GPUs help, how to fit larger models, and how to reduce the PCIe communication that can limit scaling. AMD’s Diffusion Model Inference repository packages xDiT in a container and provides tuned RDNA 4 configurations for the featured workloads.
Original AINewsTrend explainer based on the cited source. This graphic summarizes source statements; it is not an independent benchmark.

Sources and further reading

  1. amd: How to Scale Diffusion Inference Across AMD RDNA 4 GPUs