burn-metal

name	burn-metal
description	Burn Metal backend - Apple GPU acceleration for macOS and iOS. Optimized for Apple Silicon and Intel-based Macs.

burn-metal

Overview

burn-metal provides Apple GPU acceleration using Metal. It's optimized for:

• Apple Silicon: M1, M2, M3 chips
• Intel Macs: With AMD or Intel integrated GPUs
• iOS/iPadOS: Mobile devices with Apple GPUs

Key Features

• Apple Metal Support: Direct access to Metal API
• Apple Silicon Optimization: Optimized for ARM-based Macs
• Unified Memory: Shared memory between CPU and GPU on Apple Silicon
• Cross-platform: Works on macOS, iOS, iPadOS

Usage

Basic Setup

use burn::tensor::Tensor;
use burn_metal::MetalDevice;

// Create Metal device
let device = MetalDevice::default();

// Create tensor on GPU
let tensor = Tensor::from_data(&[1.0, 2.0, 3.0], &device);

// Metal-accelerated operations
let result = tensor.matmul(other_tensor);

Backend Type

use burn_metal::MetalBackend;

type Backend = MetalBackend;
let model: Model<Backend> = Model::default();

burn.toml Configuration

[model_exports]
default_backend = "metal"
default_device = "Gpu"

[metal]

Requirements

• macOS 10.15+ or iOS 13+
• Apple GPU (Apple Silicon or Intel with integrated GPU)
• Xcode command line tools

Best Practices

1. Use for Apple platform deployment
2. Leverage unified memory on Apple Silicon
3. Optimize for mobile on iOS
4. Use with Core ML integration

Performance Considerations

• Excellent performance on Apple Silicon
• Unified memory reduces data transfer overhead
• Best for macOS/iOS applications
• Good for both training and inference

When to Use

• macOS applications
• iOS/iPadOS applications
• Apple Silicon development
• Cross-platform Apple ecosystem apps

Related Crates

• burn: Core framework
• burn-wgpu: Alternative cross-platform GPU backend