Skip to content
CUDA Kernel Academy
Search
K
Main Navigation
Guides
Getting Started
Workflow
Reading Paths
Modules
01-SGEMM
02-TensorCraft
03-HPC
04-Inference
Whitepaper
Architecture
SGEMM Optimization
Inference Engine
TensorCraft
Advanced Tech
Reference
Coding Style
Troubleshooting
Integration
Benchmarks
Roadmap
English
简体中文
English
简体中文
Theme
Menu
Return to top
On this page
GEMM Optimization
Detailed walkthrough of the 7-step GEMM optimization path.
Optimization Path
Summary
Step
Technique
TFLOPS (FP32)
Speedup
1
Naive
0.5
1.0×
2
Shared Memory
2.0
4.0×
3
Double Buffer
3.5
7.0×
4
Register Tiling
6.0
12.0×
5
WMMA
50+
100×
6
MMA PTX
60+
120×
7
Pipeline
70+
140×
References