PyTorch's Metal GPU update delivers up to 3x faster AI training on Mac

PyTorch has migrated its grid_sampler_2d operation to Apple's Metal Performance Shaders, delivering dramatic speed improvements for Mac-based AI training. Benchmarks show the new Metal shader implementation is up to 3x faster than the previous MPSGraph approach, with the 512x512 bilinear float32 operation dropping from 4311.3 microseconds to just 1483.5 microseconds. This optimization was entirely coded by Claude AI, showcasing AI's growing role in improving AI infrastructure itself.