https://github.com/andreasholt/cuda-matmul-benchmarking

Implementing and benchmarking various matmul implementations in CUDA
https://github.com/andreasholt/cuda-matmul-benchmarking

cuda matrix-multiplication

Last synced: about 2 months ago
JSON representation

Implementing and benchmarking various matmul implementations in CUDA

• Host: GitHub
• URL: https://github.com/andreasholt/cuda-matmul-benchmarking
• Owner: AndreasHolt
• Created: 2024-12-11T21:38:01.000Z (2 months ago)
• Default Branch: main
• Last Pushed: 2024-12-23T12:37:56.000Z (about 2 months ago)
• Last Synced: 2024-12-23T13:35:01.381Z (about 2 months ago)
• Topics: cuda, matrix-multiplication
• Language: Cuda
• Homepage:
• Size: 19.5 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# CUDA Matrix Multiplication Benchmarking

This project implements and benchmarks different approaches to matrix multiplication using CUDA:

- Sequential CPU implementation

- Naive GPU implementation

- Tiled GPU implementation with shared memory

- More implementations with further optimizations coming in the future

The implementations and results are discussed in detail in these blog posts on my personal site:

- [Part 1: Naive GPU Implementation, Explanation, and CPU vs naive GPU Benchmarking](https://andreasholt.com/posts/gpu-vs-cpu-matmul/)

- [Part 2: Tiled Matrix Multiplication Explained and Implemented, Benchmarking against naive GPU, and Performance Analysis with Nsight Compute](https://andreasholt.com/posts/shared-tiled-matmul/)

## Building the Project

```bash

mkdir build && cd build

cmake ..

make

```

## Usage

The executable supports different modes:

### Run Full Benchmark Suite

```bash

./matmul

```

This runs benchmarks for all implementations across matrix sizes: 32×32, 256×256, 1024×1024, and 2048×2048.

### Profile Specific Implementation

```bash

./matmul profile  

```

- ``: Implementation type (`naive_gpu` or `tiled_gpu`)

- ``: Matrix dimension (creates dim×dim matrices)

Example:

```bash

./matmul profile tiled_gpu 1024

```

### NVIDIA Nsight Compute Profiling

For detailed GPU metrics:

```bash

ncu --set full -o naive_2048_full.ncu-rep ./matmul profile naive_gpu 2048

ncu --set full -o tiled_2048_full.ncu-rep ./matmul profile tiled_gpu 2048

```

## Implementation Details

The project implements matrix multiplication using different approaches:

- Each thread computes one element of the output matrix (`naive_gpu`)

- Uses shared memory tiling to improve memory access patterns (`tiled_gpu`)

- Basic CPU implementation for baseline comparison (`sequential_cpu`)

Each implementation can be benchmarked and profiled independently to compare performance across different metrics. For now these metrics include GFLOPS and time (ms).