https://blog.openvino.ai/blog-posts/techniques-for-faster-ai-inference-throughput-with-openvino-on-intel-gpus We have a follow up blog on the XMX details. Cheers. :)