CUDA GEMM

This folder contains two projects exploring different approaches to optimizing matrix multiplication on NVIDIA GPUs:

1. CUDA GEMM with CUDA Cores

• Implements matrix multiplication using traditional CUDA cores
• Otimization techniques including warp tiling, vectorized memory access, etc
• Achieves up to 85% of cuBLAS performance on an NVIDIA A6000 GPU

2. Matrix Multiplication with Tensor Cores

• Implements matrix multiplication using NVIDIA Tensor Cores
• Utilizes inline PTX and shared memory optimization
• Achieves up to 95% of cuBLAS performance on an NVIDIA A6000 GPU

Both projects include performance benchmarks, optimization strategies, and references to relevant resources.