FlashAttention
Hacker ApprovedAttention kernel library in Developer tools
Fast exact attention kernel for GPUs that people credit with making long-context models practical. Caveats: awkward on Windows and runs out of memory past 32K.
Fast exact attention kernel for GPUs that people credit with making long-context models practical. Caveats: awkward on Windows and runs out of memory past 32K.
17 people gave an opinion
- recommend it
- 14
- criticize it
- 3
- approval share
- 82%
Recommended in 7 of 13 months
- November 2025: recommended
- December 2025: recommended
- January 2026: recommended
- February 2026: not recommended
- March 2026: not recommended
- April 2026: not recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: not recommended
- August 2026: recommended
- September 2026: not recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 14 people recommend itNeeds 10 or more, met.
- 4.7 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 7 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Major gains without algorithmic compromise
“Flash attention is the only one that has really moved the needle - because it's not an algorithmic compromise.”
Comment from 26 Oct 2026 in “Attention Is Not What You Need (2025)” “immensely useful new tech and research that benefits many”
Comment from 4 Oct 2026 in “LeCun has "zero concerns" about AI wiping out humanity, recent "rogue" incidents” Memory and energy savings for attention
“allowed quadratic attention to have reasonable memory requirements again”
Comment from 28 Aug 2026 in “Nvidia agrees to acquire Hugging Face for $13B” “Tri Dao's lab must have saved countless watts with FlashAttention.”
Comment from 12 Jun 2026 in “Gram Newton-Schulz: A Fast, Hardware-Aware Newton-Schulz Algorithm for Muon” Source code teaches GPU programming
“This page gives a nice overview of it”
Comment from 18 May 2026 in “CUDA Books” “This post walks the production CUTLASS 3.x (CuTe) kernel line-by-line on Ampere.”
Comment from 18 May 2026 in “Show HN: FlashAttention-2 in Cute, from Scratch” “reading the Flash Attention and vLLM kernel source taught me more than any book”
Comment from 18 May 2026 in “CUDA Books”
What people warn about
Each mark is one comment
Hard to install or get working
“you can install flash attention, et al, but if you're on windows”
Comment from 23 Jan 2026 in “Qwen3-TTS Family Is Now Open Sourced: Voice Design, Clone, and Generation” “haven't gotten FlashAttention2 to work”
Comment from 22 Jan 2026 in “Qwen3-TTS Family Is Now Open Sourced: Voice Design, Clone, and Generation”
What people said
16 of the 18 comments that mention it, newest first. Each links to the original comment.
“Flash attention is the only one that has really moved the needle - because it's not an algorithmic compromise.”
“immensely useful new tech and research that benefits many”
“allowed quadratic attention to have reasonable memory requirements again”
“hand-optimized libraries like FlashAttention”
“cuBLAS and Flash Attention kernels still use the same cp.async transport on consumer Blackwell dies”
“Tri Dao's lab must have saved countless watts with FlashAttention.”
“finally some good oepnsource repo with quantization on cuda level”
“This post walks the production CUTLASS 3.x (CuTe) kernel line-by-line on Ampere.”
“The story of Flash Attention is the best manifestation of power and difficulty of GPU programming.”
“This page gives a nice overview of it”
“reading the Flash Attention and vLLM kernel source taught me more than any book”
“FlashAttention OOMs beyond 32K”
“you can install flash attention, et al, but if you're on windows”
“haven't gotten FlashAttention2 to work”
“While the original paper is brilliant”
“their implementations are often much better (in FLOPS) compared to what were proposed in the original papers”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.