Apple's M5 and A19 GPU family introduces the 'Family 10' GPU architecture with significant performance improvements. Key automatic gains include 2x faster FP16/complex ALU execution, doubled geometry throughput, 30% more memory bandwidth, second-generation dynamic caching with a smarter occupancy management unit, universal texture compression (now covering compute shaders), and third-generation hardware ray tracing with 70% faster intersection function buffer indexing and reduced memory alignment from 16KB to 1KB. New opt-in features include extended indirect command buffers (ICBs) allowing full GPU-driven pipelines with per-draw device state encoding, visibility buffer rendering, depth bound testing, 8x MSAA resolved on-chip, and 32K texture support. Xcode 26.4 adds new occupancy profiling counters (register pressure, L1 cache pressure, memory stalls, texture decompression stalls) to diagnose and fix low GPU occupancy. Best practices for universal texture compression are covered, including when to disable it for scattered access patterns using the compression ratio and write inefficiency counters.
Sort: