MacBenchForge — The Ultimate LLM Benchmarking Suite for Apple Silicon

Capabilities

Everything you need to benchmark LLMs

Designed with a zero-bloat philosophy. No Node.js, no npm, no Electron. Just pure C++ and Vanilla JS delivering native performance.

Apple Silicon Native

Aggressively leverages Metal Performance Shaders via llama.cpp, pushing Unified Memory and GPU cores to their absolute limit.

HuggingFace Integration

Browse, search, and download .gguf models directly from the HuggingFace Hub. Curated Apple Silicon recommendations included.

Advanced Benchmarking

Measure TTFT, tokens/sec, prompt eval, and generation speed. Customizable presets from Quick to Thorough with varied prompt types.

Hardware Telemetry

Real-time detection of your M-series chip, CPU/GPU core counts, system RAM, and dynamic Metal VRAM wiring limits.

Beautiful Visualizations

Glassmorphic dark UI with live Chart.js graphs, radar comparisons, progress bars, and toast notifications for a premium feel.

Data Portability

Export your complete benchmark history to JSON or CSV. Compare runs, share results, and track performance across model upgrades.

Interface

Designed to be beautiful

A premium glassmorphism UI that makes benchmarking feel delightful. Dark mode, live telemetry, and stunning visualizations.

Under the Hood

Clean, separated architecture

A C++17 backend with an ultra-lightweight REST API, SQLite persistence, and a zero-dependency frontend SPA served from memory.

Backend

C++17 Engine

cpp-httplib Ultra-lightweight REST API
SQLiteCpp Persistent history tracking
osascript Native macOS integration
libcurl HuggingFace Hub downloads

Engine

Inference Core

llama-bench Isolated LLM throughput
Metal Full GPU acceleration
IOKit Hardware detection via sysctl
Memory Poller Real-time RAM telemetry

Frontend

Zero-Bloat SPA

Vanilla JS No frameworks, no build tools
CSS3 Custom properties + glassmorphism
Chart.js Performance visualizations
REST Clean API client layer

Get Started

Up and running in minutes

Download the pre-built .dmg for Apple Silicon or build from source with a few commands.

Recommended

Download .dmg

Pre-built for Apple Silicon. Just download, mount, and drag to Applications.

v2.1.1 ~1.5 MB Apple Silicon

Download for macOS

Requires macOS Ventura (13.0)+ and brew install llama.cpp

Build from Source

Full control. Clone, build with CMake, and run locally.

Terminal

# Clone with submodules
git clone --recursive https://github.com/AdityaGuhaa/MacBenchForge.git
cd MacBenchForge

# Install dependencies
brew install curl openssl cmake llama.cpp

# Build and launch
mkdir build && cd build
cmake ..
cmake --build . -j
./bin/MacBenchForge

Installation Troubleshooting

Running into an error? Watch this quick preview guide on how to resolve common installation issues.

Ready to benchmark your Apple Silicon?

Join the community of developers pushing their M-series Macs to the limit.

Star on GitHub Download v2.1.1

The Ultimate LLM Benchmark for Apple Silicon