XARIV Fusion Studio

See kernel fusion — not just read about it

Replay one decode step side by side: hundreds of micro-kernel launches vs fused Metal blocks from llama.cpp. Timings are scaled from local llama-bench on Apple Silicon.

Run fusion compare Read the study

Browser replay: kernel timeline + launch counters — works on Vercel.
Your Mac: toggle real fusion with GGML_METAL_FUSION_DISABLE=1 and llama-bench.

← XARIV Relay (quantization compare)