https://github.com/carlosacchi/apple-silicon-bench
Lightweight benchmark tool for Apple Silicon. Test CPU single/multi-core, memory bandwidth, and disk I/O performance on M1, M2, M3, M4, M5 chips.
https://github.com/carlosacchi/apple-silicon-bench
apple apple-silicon benchmark benchmarking benchmarks blackmagic cpu-benchmark disk-benchmark geekbench macbook macbook-air macbook-pro memory-benchmark osx
Last synced: 5 months ago
JSON representation
Lightweight benchmark tool for Apple Silicon. Test CPU single/multi-core, memory bandwidth, and disk I/O performance on M1, M2, M3, M4, M5 chips.
- Host: GitHub
- URL: https://github.com/carlosacchi/apple-silicon-bench
- Owner: carlosacchi
- License: mit
- Created: 2025-12-30T22:48:33.000Z (5 months ago)
- Default Branch: main
- Last Pushed: 2026-01-03T21:50:00.000Z (5 months ago)
- Last Synced: 2026-01-04T11:43:19.550Z (5 months ago)
- Topics: apple, apple-silicon, benchmark, benchmarking, benchmarks, blackmagic, cpu-benchmark, disk-benchmark, geekbench, macbook, macbook-air, macbook-pro, memory-benchmark, osx
- Language: Swift
- Homepage:
- Size: 1.16 MB
- Stars: 10
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- License: LICENSE
- Security: SECURITY.md
Awesome Lists containing this project
README
# Apple Silicon Bench
**A native macOS benchmark tool for Apple Silicon Macs (M1, M2, M3, M4, M5)**
[](https://github.com/carlosacchi/apple-silicon-bench/releases)
[](https://opensource.org/licenses/MIT)
[](https://www.apple.com/macos/)
[](https://swift.org/)
A lightweight, native Swift benchmark tool designed specifically for Apple Silicon processors. Compare your M1, M2, M3, M4, M5 (and future chips) performance with comprehensive CPU, GPU, memory, disk, and AI benchmarks.
## Features
- **CPU Single-Core & Multi-Core**: Integer, floating-point, SIMD, cryptography, compression
- **GPU Benchmark (Metal)**: Compute shaders, particle simulation, image processing
- **Memory Benchmark**: Bandwidth and latency measurements
- **Disk Benchmark**: Sequential and random I/O with cache bypass
- **AI/ML Benchmark (NEW in v2.0)**: CoreML inference (CPU/GPU/Neural Engine), BNNS operations
- **Thermal Monitoring**: Real-time throttling detection
- **HTML Reports**: Beautiful interactive reports saved to Desktop
- **Lightweight**: ~2MB standalone binary, no dependencies
## Quick Start
### Download Binary
```bash
curl -LO https://github.com/carlosacchi/apple-silicon-bench/releases/latest/download/osx-bench-macos-arm64.tar.gz
tar -xzf osx-bench-macos-arm64.tar.gz
xattr -cr osx-bench && chmod +x osx-bench
./osx-bench run
```
### Build from Source
```bash
git clone https://github.com/carlosacchi/apple-silicon-bench.git
cd apple-silicon-bench
swift build -c release
./.build/release/osx-bench run
```
## Usage
```bash
# Full benchmark (recommended)
osx-bench run
# Quick mode (~3s per test, less accurate)
osx-bench run --quick
# Custom duration
osx-bench run --duration 30
# Stress test (60s per test)
osx-bench run --stress
# Selective benchmarks
osx-bench run --only cpu-single,gpu
osx-bench run --only memory,disk
osx-bench run --only ai
# AI benchmark options
osx-bench run --only ai --model-path /path/to/model.mlmodelc
osx-bench run --offline # Skip AI if model not cached
# Advanced profiling (v2.1.0+)
osx-bench run --advanced # Stride sweep, QD matrix, thread scaling
# System info
osx-bench info
osx-bench info --extended
# Export results
osx-bench run --export results.json
```
## Scoring
### Total Score
- **Baseline**: M1 base chip = 1000 points per category (calibrated from median of 5 full runs)
- **Method**: Geometric mean of ratios (commonly used in benchmark suites)
- **Weights**: CPU-Single 25%, CPU-Multi 25%, Memory 15%, Disk 15%, GPU 20%
| Chip | Expected Score (Rule-of-Thumb) |
|------|----------------|
| M1 | ~1000 |
| M2 | ~1100 |
| M3 | ~1290 |
| M4 | ~1600 |
### AI Score (Separate)
The AI/ML score is reported **separately** from the Total Score (similar to Geekbench AI):
- **AI-CPU**: CoreML inference with CPU-only compute
- **AI-GPU**: CoreML inference with GPU acceleration
- **AI-Neural Engine**: CoreML inference with Neural Engine (when available)
- **AI-BNNS**: Accelerate framework matrix operations
*Note: Actual results may vary based on chassis, cooling, and configuration*
For detailed methodology, see the [Wiki](https://github.com/carlosacchi/apple-silicon-bench/wiki).
## Thermal States
- 🟢 **Nominal**: No throttling
- 🟡 **Fair**: Minor throttling possible
- 🟠**Serious**: Significant throttling
- 🔴 **Critical**: Maximum throttling
## Requirements
- macOS 13.0 (Ventura) or later
- Apple Silicon (M1, M2, M3, M4, M5 family)
## Documentation
See the **[Wiki](https://github.com/carlosacchi/apple-silicon-bench/wiki)** for:
- [Scoring Methodology](https://github.com/carlosacchi/apple-silicon-bench/wiki/Scoring-Methodology) - How scores are calculated
- [Benchmark Details](https://github.com/carlosacchi/apple-silicon-bench/wiki/Benchmark-Details) - Technical details of each test
- [Advanced Profiling](https://github.com/carlosacchi/apple-silicon-bench/wiki/Advanced-Profiling) - PassMark-inspired deep analysis
- [FAQ](https://github.com/carlosacchi/apple-silicon-bench/wiki/FAQ) - Common questions and troubleshooting
- [Roadmap](https://github.com/carlosacchi/apple-silicon-bench/wiki/Roadmap) - Planned features
## Why Apple Silicon Bench?
| Feature | Apple Silicon Bench | Geekbench 6 | Cinebench |
|---------|---------------------|-------------|-----------|
| Open Source | Yes | No | No |
| Offline | Yes | Account required | Yes |
| Transparent Scoring | Yes | Closed | Closed |
| Thermal Monitoring | Yes | No | No |
| Binary Size | ~2MB | ~200MB | ~1GB |
| Price | Free | $15 (Pro) | Free |
## Contributing
Contributions welcome! See the [Wiki](https://github.com/carlosacchi/apple-silicon-bench/wiki) for guidelines.
## License
MIT License - see [LICENSE](LICENSE) file.
## Author
**Carlo Sacchi** - [@carlosacchi](https://github.com/carlosacchi)
---
Made with Swift for Apple Silicon