PyTorch 2.10+TorchAO: Powering AIPC scenarios on Intel® Core™ Ultra Series 3 processors – PyTorch

PyTorch 2.10 introduces XPU backend support optimized for Intel Core Ultra Series 3 processors with Arc B-series integrated GPUs. Key features include TorchAO integration for out-of-the-box Int4/Int8 quantization enabling LLM inference on AI PCs, and SYCL custom operator support extended to Windows. Benchmark data shows inference latency for popular models (Qwen3, Phi-4, Llama 3.2, DeepSeek-R1) on the Intel Core Ultra X9 388H. Practical use cases demonstrated include visual anomaly detection with Anomalib and robotics training with LeRobot, both running natively on integrated Intel GPUs with minimal code changes.