PulseAugur
LIVE 01:01:35
research · [3 sources] ·
0
research

NVIDIA open-sources cuDNN kernels after 12 years, including MoE and sparse attention

NVIDIA has open-sourced parts of its cuDNN library, a significant move after 12 years of it being closed-source. This release includes over 20 Mixture-of-Experts (MoE) kernels and NSA sparse attention kernels. The codebase for these kernels is largely written in Python CuTe-DSL, with public documentation now available. AI

Summary written by gemini-2.5-flash-lite from 3 sources. How we write summaries →

IMPACT Open-sourcing of cuDNN kernels could accelerate research and development in AI infrastructure and model optimization.

RANK_REASON Open-sourcing of a significant software library component by a major tech company.

Read on X — SemiAnalysis →

NVIDIA open-sources cuDNN kernels after 12 years, including MoE and sparse attention

COVERAGE [3]

  1. X — SemiAnalysis TIER_1 · SemiAnalysis_ ·

    @manicely6005 The public documentation can be found here too (3/3)

    @manicely6005 The public documentation can be found here too (3/3) https://t.co/jUnoOJ7kj7 https://t.co/M74p6UyTKc

  2. X — SemiAnalysis TIER_1 · SemiAnalysis_ ·

    @manicely6005 The codebase for these kernels, a lot of these OSS cuDNN kernels are written in Python CuTe-DSL (2/3)

    @manicely6005 The codebase for these kernels, a lot of these OSS cuDNN kernels are written in Python CuTe-DSL (2/3) https://t.co/7awOYa0dbd https://t.co/euu5qaKveW

  3. X — SemiAnalysis TIER_1 · SemiAnalysis_ ·

    For the past 12 years, cuDNN has been completely closed sourced (besides the .h files), until this week! OVER 20 MoE kernels & NSA sparse attention kernels

    For the past 12 years, cuDNN has been completely closed sourced (besides the .h files), until this week! OVER 20 MoE kernels & NSA sparse attention kernels from cuDNN has been open sourced! Great work to @manicely6005 & the rest of the team on seeing that parts of NVIDIA …