AI-Generated Metal Kernels Deliver 87% Speed-Up on Apple Silicon
Your PyTorch models just got a turbo boost on MacBooks and iPads. Researchers fed frontier models a simple prompt: "write an optimized Metal kernel for this op," and in seconds generated GPU code that outperformed hand-tuned Apple libraries—without a single human-written shader. The takeaway: you can now prototype on Apple devices and keep the same model when you deploy. Grab the open-source kernels at https://gimletlabs.ai/blog/ai-generated-metal-kernels and watch your inference times drop while your laptop battery lasts longer.
hacker_news · 7 min read