AI/ML●●●Banger
RunAnwhere – Faster AI Inference on Apple Silicon
Custom Metal shaders beat llama.cpp and MLX—1.67x faster on M4 Max.
WizardrySlickZero to One
sanchitmonga22
2401532mo ago
Back to browse
GitHub Repository
9 starsRust
IEEE-754-Conformant FP64 on Metal (Apple Silicon)
by guyfischman·May 8, 2026·1 point·1 comment
AI Analysis
●●●BangerWizardryNiche Gem
Bit-exact f64 emulation on Metal GPUs where Apple's native double support is missing.
Strengths
- •Pure integer operations on bit patterns enable deterministic math on GPUs lacking f64.
- •Includes unpacked path for tight inner loops skipping special-case dispatch overhead.
- •Rust reference implementation ensures bit-exact verification between GPU and CPU.
Weaknesses
- •Extremely niche audience limited to those needing precise float math on Metal specifically.
- •Performance overhead compared to native f32 operations may limit real-time use cases.
Category
Target Audience
Graphics programmers and simulation engineers on Apple Silicon
Similar To
SoftFloat · Cuda Softfloat
Similar Projects
Developer Tools●●●Banger
GhostVM – macOS Containers on Apple Silicon
Native macOS VMs with APFS snapshots beat Docker for agent isolation.
WizardrySolve My ProblemNiche Gem
CarolineWang
702mo ago
AI/ML●●●Banger
Bonsai 1.7B ternary model at 442T/s on M4 Max
Autonomous agent wrote custom Metal kernels boosting decode speed 42% over upstream llama.cpp.
WizardryDark Horse
hhuytho
1331mo ago
Developer Tools●●Solid
Local TTS for OpenClaw on Apple Silicon (MLX-Powered, Zero Setup)
MLX-powered local TTS plugin for OpenClaw—elegant but audience is Apple Silicon only.
Niche GemCozy
ZacharyZZ
203mo ago
Productivity●●Solid
Disk Inventory X updated for Apple Silicon
Classic treemap UI back on native Apple Silicon, but disk space visualizers already exist.
Dark HorseCozy
lababidi
423mo ago
Data●●Solid
Benchmarking Apple Silicon unified mem for GPU-accelerated SQL analysis
The repo does one practical thing well: quantify the real-world impact of Apple Silicon's unified memory on analytics by running six TPC-H queries plus a GPU-favorable QX and shipping the raw charts and code. It's specific and empirical — you get MLX vs NumPy vs DuckDB numbers and PNGs, not just hand-wavy claims — but it's narrowly scoped to M4 hardware and small-ish scales, so its conclusions are useful for experimentation rather than sweeping generalization.
WizardryNiche Gem
sadopc
313mo ago