CUDA Neural Network Inference Engine

A minimal GPU implementation of a 2-layer feedforward neural network using CUDA. Demonstrates matrix multiplication with shared memory tiling and GPU vs CPU performance.

Features

• Forward pass: input → hidden → output
• ReLU and Softmax activation kernels
• Tiled matrix multiplication (16×16 blocks)
• GPU vs CPU timing comparison

Requirements

• NVIDIA GPU with CUDA support
• CUDA Toolkit installed
• C++17 compiler

Build & Run

nvcc -arch=sm_75 -O3 src/main.cu -o build/main
./build/main

Sample Output

GPU forward pass time: 0.31 ms
probability of class 0: 0.00
CPU forward pass time: 4.19 ms

Tested with CUDA 12.4 and C++17. Adjust -arch to match your GPU’s compute capability.