deepseek-ai/DeepGEMM

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

Cuda7.7K stars1.2K forks+109 today

What it does

DeepGEMM is a high-performance CUDA library for tensor core kernels, offering efficient FP8 GEMM operations and more. It simplifies GPU kernel optimization techniques while matching or surpassing expert-tuned libraries.

Star history

Not enough history yet — 1 day(s) recorded. The daily snapshot builds this up.

Tracking

Last trending: 2026-04-25

Creator kit

Hook

Revolutionize your AI workloads with DeepGEMM, a high-performance CUDA library for tensor core kernels. #AI #CUDA

Content angles

Exploring the performance benefits of DeepGEMM in large language model training.
A deep dive into how DeepGEMM optimizes FP8 GEMM operations on modern GPUs.
Comparing DeepGEMM's efficiency and ease-of-use against traditional CUDA libraries.

Who should care

Researchers, developers, and engineers working with AI models and GPU optimization.