Tagged articles
Metal
19 articles · Page 1 of 1
DwarfStar is a specialized local inference engine by Redis creator antirez, optimized exclusively for DeepSeek V4, GLM 5, and Qwen3.8 Flash Next models across Metal, CUDA, and ROCm platforms, achieving 790 tokens/s prefill on M5 Max and 126 aggregated tokens/s on 8×L40S, with built-in coding agent and HTTP server.
CUDACoding AgentDwarfStar
0 likes · 16 min read

A CUDA/HIP‑based 3‑D fluid simulation was ported to an Apple M3 Pro device via a Clang‑HIP → SPIR‑V → Vulkan → MoltenVK conversion pipeline, achieving roughly ten‑fold speedup without any Metal‑specific API changes and preserving numerical accuracy.
3D fluid simulationCUDAGPU
0 likes · 7 min read

The article details how the open‑source Flash‑MoE engine enables the 400B‑parameter Qwen3.5‑397B‑A17B mixture‑of‑experts model to run on an iPhone 17 Pro, achieving about 0.6 tokens per second through a custom Metal pipeline, GCD‑driven SSD streaming, and aggressive caching strategies.
400BFlash-MoELLM inference
0 likes · 6 min read

The ds4.c project, authored by Redis founder Salvatore Sanfilippo, is a Metal‑only C inference engine that uses asymmetric 2‑bit quantization, disk‑based KV caching, and OpenAI/Anthropic‑compatible APIs to achieve usable performance for DeepSeek V4 Flash on high‑end Apple Silicon Macs.
Apple SiliconC++DS4
0 likes · 9 min read


Geek LabsMay 13, 2026 · Artificial Intelligence
This article compares two recent GitHub LLM inference engines—ds4.c, a Metal‑optimized engine for DeepSeek V4 Flash on Apple Silicon Macs, and TokenSpeed, a Python/C++‑based, data‑center‑grade engine for GPU clusters—detailing their design choices, performance numbers, usage instructions, and suitable scenarios.
DeepSeekGPULLM
0 likes · 8 min read

Redis founder antirez unveiled ds4.c, a pure‑C inference engine that leverages Objective‑C and Metal to run DeepSeek V4 locally on Mac devices, delivering about 27 token/s on an M3 Ultra—far slower than GPU servers but offering a dependency‑free, on‑device solution that keeps data private.
AIC++DeepSeek
0 likes · 8 min read

Redis creator Salvatore Sanfilippo (antirez) released DS4, a Metal‑only C inference engine tailored for DeepSeek V4 Flash on high‑end Macs, featuring narrow model focus, 2‑bit quantization, disk‑based KV cache, benchmark speeds around 26 tokens/s, and a dual OpenAI/Anthropic compatible server.
2-bit QuantizationAI inference engineKV Cache
0 likes · 13 min read

The article details how the open‑source Flash‑MoE engine streams a 400‑billion‑parameter Mixture‑of‑Experts language model on an iPhone 17 Pro, achieving interactive‑level token throughput by eliminating Python dependencies, crafting a custom Metal pipeline, and streaming weights directly from SSD.
Apple SiliconFlash-MoEGCD
0 likes · 7 min read

The article details the Flash‑MoE project that streams the 400 billion‑parameter Qwen3.5‑397B‑A17B mixture‑of‑experts model on an iPhone 17 Pro, achieving up to 0.6 tokens per second with a custom Metal‑GPU pipeline, zero‑Python code, and SSD‑backed weight streaming that keeps only 5.5 GB in RAM.
Flash-MoELLMMetal
0 likes · 7 min read

This article provides an in‑depth technical overview of Flutter's Impeller rendering engine, covering its origin, Jank classification, Metal shader compilation evolution, vector rendering fundamentals, DisplayList architecture, Impeller's rendering pipeline, and the ImpellerC shader compiler, with code examples and performance insights.
DisplayListFlutterImpeller
0 likes · 31 min read

This article details Youku's technical approach to rendering bullet‑screen pass‑through on mobile devices, covering cloud‑based and on‑device segmentation pipelines, GPU‑accelerated rendering steps, performance optimizations, and engineering challenges to deliver seamless immersive viewing.
Bullet ScreenGPUMetal
0 likes · 11 min read

Apple’s WWDC21‑introduced Adaptive‑Sync on macOS and ProMotion on iPad Pro and iPhone 13 Pro enable variable‑refresh‑rate displays, and developers can use Metal and CADisplayLink APIs to dynamically pace frames, query hardware limits, and adjust rates based on GPU load for smoother motion.
Adaptive SyncCADisplayLinkMetal
0 likes · 18 min read

This article provides a comprehensive overview of WebRTC video rendering across major platforms, detailing the pipelines and rendering technologies such as OpenGL, Metal, Vulkan, and Direct3D, and explains how these APIs are employed on iOS, Android, macOS, and Windows for efficient real‑time video playback.
Direct3DMetalOpenGL
0 likes · 15 min read

The article analyzes why Flutter crashes on iOS when accessing the GPU in the background, explains the official SyncSwitch fix for ImageDecoder, and details Xianyu’s additional patches for MultipleFrameCodec, EncodeImage, and Rasterizer::DrawToSurface that together, via PR #28383, fully resolve the GPU‑background crash.
FlutterGPUImageDecoder
0 likes · 11 min read

This article explains how Apple Silicon affects Metal applications, outlines migration steps from Intel to Apple Silicon, describes new GPU architectures and API features, and provides practical best‑practice guidelines to achieve optimal performance and correctness on the new platform.
Apple SiliconGPUMetal
0 likes · 11 min read

This article compares forward and deferred rendering techniques, analyzes their performance trade‑offs on mobile GPUs, explores tile‑based and hardware‑TBDR approaches, and presents a Metal‑based single‑pass deferred shading solution for modern mobile graphics pipelines.
GraphicsMetalMobile GPU
0 likes · 15 min read

This article traces the evolution of Baidu Maps' mobile rendering engine, explains why Metal outperforms OpenGL, outlines a cross‑platform engine redesign, and quantifies the CPU, GPU, and memory improvements achieved after the Metal upgrade.
Baidu MapsGraphics APIMetal
0 likes · 11 min read

GCanvas is a lightweight, W3C‑compliant graphics engine that evolved from H5 acceleration to server‑side rendering, offering front‑end developers high‑performance canvas rendering, extensive platform support, and detailed optimization techniques such as JSBinding and Metal integration.
Frontend DevelopmentGCanvasGraphics Rendering
0 likes · 9 min read

This article explains ARKit’s architecture, core tracking components, scene‑recognition features such as plane detection, hit‑testing and light estimation, and provides a step‑by‑step demo using Xcode templates to render a 3D airplane model on iOS devices.
ARKitMetalSceneKit
0 likes · 10 min read
