Metal
A low-level graphics and compute API introduced by Apple in 2014, deeply optimized for Apple platform hardware. It communicates directly with the GPU to provide low CPU overhead and high rendering throughput, and supports custom shader programming.
Core Facts
llama.cpp是由Georgi Gerganov开发的纯C/C++推理引擎，支持在纯CPU环境下运行大模型
Metal is a low-level graphics API introduced by Apple in 2014, optimized specifically for Apple hardware.
llama.cpp supports GPU acceleration backends including Metal, CUDA, Vulkan, and SYCL.
Timeline (last 90 days)
The core change in v0.40.13-rc1 is a temporary compatibility patch for the Metal backend, used to fix crashes caused by failure to compile the unary library on older versions of macOS.
After upgrading to v0.40.13-rc1, Metal inference on older versions of macOS should return to normal.
This patch only targets the Metal backend and does not constitute a general solution for similar library compilation errors on other platforms (such as the CUDA and Vulkan backends).
The unary library refers to a set of Metal kernels that carry element-wise unary operations (such as negation, absolute value, exponentiation, etc.).
At runtime, ggml dynamically compiles Metal compute kernels, in the form of source-code strings, into libraries callable by the GPU, a mechanism known as runtime compilation.
Through the Metal backend of the ggml tensor computation library, llama.cpp offloads inference workloads such as matrix operations to the Mac's GPU for execution.
This regression only affects older versions of macOS, with no triggering conditions on newer systems.
Apple provides Core ML, Metal and the MLX framework launched in recent years in macOS and iOS for local model deployment.
FlyEnv's llama.cpp plugin supports installing and retaining multiple Metal runtime versions coexisting side by side.
When mapping GameCube's GX graphics instructions to modern graphics APIs (Vulkan, Metal), AI can assist in generating shader conversion code and state management logic, but developers still ultimately need to verify the correctness of rendering results.
40 more timeline events
All Facts (20)
llama.cpp是由Georgi Gerganov开发的纯C/C++推理引擎，支持在纯CPU环境下运行大模型
90%VerifiedMetal is a low-level graphics API introduced by Apple in 2014, optimized specifically for Apple hardware.
80%Verifiedllama.cpp supports GPU acceleration backends including Metal, CUDA, Vulkan, and SYCL.
80%VerifiedOllama automatically detects NVIDIA GPUs in the system and selects the optimal inference backend (CUDA, Metal, or pure CPU), and achieves GPU acceleration for Apple Silicon users via the Metal API.
80%Verifiedllama.cpp is a highly optimized C++ inference engine that supports Metal acceleration for Apple Silicon, CUDA acceleration for NVIDIA, and ROCm acceleration for AMD.
80%VerifiedGPUI is the GPU-accelerated rendering framework used by the code editor Zed, implementing rendering on macOS via Metal.
75%VerifiedMetal is Apple's low-level GPU computing interface designed for its hardware, analogous to CUDA in the NVIDIA ecosystem.
75%Verifiedllama.cpp supports GPU acceleration via CUDA for NVIDIA GPUs, ROCm for AMD GPUs, and Metal for Apple Silicon, and also supports pure CPU inference.
70%Verifiedllama.cpp has evolved into a general-purpose local inference framework covering multiple backends including CPU, CUDA, Metal, and Vulkan.
65%VerifiedThe Metal API is a low-level graphics interface introduced by Apple in 2014, replacing OpenGL.
65%VerifiedMetal is a low-level graphics API introduced by Apple in 2014 that can reduce CPU overhead by approximately 40% compared to traditional OpenGL ES.
65%UnverifiedApple Silicon M-series unified memory architecture allows CPU and GPU to share the same physical memory, eliminating data copy overhead beneficial for LLM inference
90%UnverifiedPorting Metal kernels to CUDA is not a simple syntax translation; it involves unified memory model data residency assumptions, threadgroup-to-threadblock mapping, and redesigning register and shared memory utilization strategies for quantization and dequantization.
70%UnverifiedSimultaneous support for Metal, CUDA, and ROCm is uncommon among open-source inference projects
70%UnverifiedWebGPU introduces compute shader capabilities comparable to Vulkan/Metal/D3D12, enabling general-purpose GPU computing (GPGPU).
60%Unverifiedds4 is written in pure C with native support for Metal, CUDA, and ROCm backends
60%UnverifiedThis regression only affects older versions of macOS, with no triggering conditions on newer systems.
50%UnverifiedThrough the Metal backend of the ggml tensor computation library, llama.cpp offloads inference workloads such as matrix operations to the Mac's GPU for execution.
50%UnverifiedAt runtime, ggml dynamically compiles Metal compute kernels, in the form of source-code strings, into libraries callable by the GPU, a mechanism known as runtime compilation.
50%UnverifiedThe unary library refers to a set of Metal kernels that carry element-wise unary operations (such as negation, absolute value, exponentiation, etc.).
50%