GitHub Repository

A zero-dependency, zero-copy Python-to-Metal GPGPU advection engine & local transformer. Runs FP16 text embeddings and parallel reductions entirely on macOS/iOS integrated GPUs.

0 starsPython

Sovereign-Metal – Zero-Dependency Python/Metal GPGPU Advection

Name: Sovereign-Metal – Zero-Dependency Python/Metal GPGPU Advection
Availability: InStock
Author: Jamie_Nixx_CUI

by Jamie_Nixx_CUI·Jul 18, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●●BangerWizardryBig BrainNiche Gem

Zero-copy unified memory bypasses PCIe bottlenecks for local Metal inference.

Strengths

•Direct NumPy to MTLResourceStorageModeShared mapping eliminates host-device transfer latency.
•Branchless toroidal boundary wrapping using bitwise ops prevents thread divergence in shaders.
•Runs FP16 BERT-style embeddings entirely on integrated GPU without external ML frameworks.

Weaknesses

•Strictly limited to Apple Silicon, excluding the vast majority of CUDA-based ML workflows.
•Niche fluid dynamics focus may limit immediate utility for standard transformer applications.