AMD and Intel agreed on something pretty unusual. A local PyTorch AI app shouldn't have to care whether the CPU underneath it is AMD or Intel. AMD is folding its ZenDNN AI optimizations into oneDNN, the library originally developed by Intel. AMD + Intel have now aligned on oneDNN as the default open-source x86 acceleration path in PyTorch. ZenDNN has been working on refining for ... 🧠 LLM inference ⚡ INT4 / INT8 / BF16 / FP16 🧩 fused MoE operations 🚀 vLLM CPU acceleration 🔧 optimized attention/matmul Now that work is being brought into the common oneDNN ecosystem AMD + Intel are also jointly developing ACE, new x86 AI matrix instructions. The plan is for future AMD/Intel AI instructions → oneDNN → PyTorch So local PyTorch software gets one x86 target instead of separate vendor paths. ⚠️ This does NOT make llama.cpp/GGUF faster, and AMD/Intel haven't published new performance numbers from the merger. 🎯But for CPU Local AI, the two companies that spent decades fighting over x86 are now cooperating on the AI software layer. Link in ALT

AMD is bringing FSR Upscaling 4.1 to Xbox. The tech is coming first to the ROG Xbox Ally handheld, with Xbox Series X|S support also planned. AMD is also teasing FSR “Redstone”/Diamond for Project Helix, Microsoft’s next-generation Xbox hardware. AMD’s latest upscaling tech is now being developed across both Xbox and PlayStation hardware, with Sony involved in Project Amethyst.

AMD $AMD last week unveiled its new ‘Helion’ racks for agentic applications, featuring up to 32,768 EPYC CPU cores in a CPU-only rack configuration or a CPU-plus-GPU configuration including up to 16 MI355X GPUs. $NVDA $INTC $ARM

AMD’s new Helion rack architecture is designed to let cloud operators dedicate CPU resources to running AI agents while handling model inference on separate systems. The CPU-only configuration supports up to 32,768 EPYC cores per rack. An all-in-one version combines EPYC CPUs, Instinct MI350P GPUs and general-purpose compute in a single rack. AMD says it is developing Helion configurations with ecosystem partners and plans to make them available in November 2026. Foxconn, QCT and Wiwynn are among the companies supporting the design.