Splash

3 articles tagged with this topic

Mac Local LLMs Hit 50% Faster — AI Wrote the Performance Code

Developer used Claude to rewrite Splash inference engine for M5 Max: 50% faster concurrent, ~25% single request. AI now writes performance code.

Sep 272 min read

SplashApple Silicon

Splash 1.1: Apple Silicon Now Runs 27B AI Models Smoothly

Splash 1.1.0 runs 27B models locally on top-spec Macs at ~50 tokens/sec. Local AI inference becomes practical—enterprises can skip cloud fees.

Sep 262 min read

AppleM5 Max

M5 Max Mac Local AI 20% Faster—Hidden Cost: Per-Gen Kernel Retuning Required

Developer retunes inference kernels for 40-core M5 Max, boosting local LLM generation 20%. Each Apple Silicon generation needs manual AI re-optimizati

Sep 252 min read