Open-Sourced PTX KDA Kernel Delivers 1.59x Speedup Over FlashKDA
JFPuget · x · 2026-07-28
A developer has open-sourced PTX KDA, a fully automated generated KDA kernel designed for Kimi (Moonshot).
According to the announcement, the kernel achieves a 1.59x speedup over FlashKDA. The code is now available on GitHub for exploration.
Related event: Open-Source PTX KDA Kernel Released for Blackwell(2 posts)→
More from Infra
- Kimi K3 GGUF IQ1_S lands on Hugging Face as 1-bit and 2-bit fixes continue — victormustar · 2026-07-28
- Hyperscalers are tying 2026 capex to customer commitments to avoid fiber-era returns — inductionheads · 2026-07-28
- LightRAG v1.5.3 adds a one-time Milvus migration and hardens production edge cases — JeremyCMorgan · 2026-07-28
- llama.cpp adds DSpark speculative decoding and asks for speed results — pmttyji · 2026-07-28
- Nvidia puts an open vision-language-action model on Hugging Face — theteknosaur · 2026-07-28
- “Model eats harness” is really about deployment-driven harness evolution — m_wulfmeier · 2026-07-28