burn-cuda

name	burn-cuda
description	Burn CUDA backend - High-performance NVIDIA GPU acceleration. Provides optimal performance for NVIDIA hardware.

burn-cuda

Overview

burn-cuda provides NVIDIA GPU acceleration using CUDA. It's optimized for NVIDIA GPUs and provides the best performance for NVIDIA hardware.

Key Features

• NVIDIA CUDA Support: Direct access to CUDA cores
• High Performance: Optimized for NVIDIA GPUs
• Tensor Cores: Support for mixed-precision computing
• CUDA Toolkit Integration: Uses NVIDIA's CUDA toolkit

Usage

Basic Setup

use burn::tensor::Tensor;
use burn_cuda::CudaDevice;

// Create CUDA device
let device = CudaDevice::default();

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

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

Backend Type

use burn_cuda::CudaBackend;

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

burn.toml Configuration

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

[cuda]

Requirements

• NVIDIA GPU with CUDA support
• CUDA Toolkit installed
• Linux or Windows operating system

Best Practices

1. Use for maximum GPU performance
2. Enable mixed-precision training
3. Monitor GPU memory usage
4. Use batch processing for optimal throughput

Performance Considerations

• Best performance on NVIDIA GPUs
• Requires NVIDIA hardware
• Not available on macOS (use burn-metal instead)
• Excellent for large-scale training

When to Use

• NVIDIA GPU deployment
• High-performance training
• Production inference on NVIDIA hardware
• Research with large models

Related Crates

• burn: Core framework
• burn-metal: Apple GPU alternative for macOS/iOS