# FlashAttention

14 Hacker News commenters recommended FlashAttention and 3 criticized it, November 2025 to November 2026. It was recommended in 7 of 13 months. It meets the Hacker Approved badge rule.

**Verdict:** 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.

## Why people recommend it

- Major gains without algorithmic compromise (2 comments): "Flash attention is the only one that has really moved the needle - because it's not an algorithmic compromise." ([source](https://news.ycombinator.com/item?id=50252033)) "immensely useful new tech and research that benefits many" ([source](https://news.ycombinator.com/item?id=49954679))
- Memory and energy savings for attention (2 comments): "allowed quadratic attention to have reasonable memory requirements again" ([source](https://news.ycombinator.com/item?id=49474369)) "Tri Dao's lab must have saved countless watts with FlashAttention." ([source](https://news.ycombinator.com/item?id=48498200))
- Source code teaches GPU programming (3 comments): "This page gives a nice overview of it" ([source](https://news.ycombinator.com/item?id=48185301)) "This post walks the production CUTLASS 3.x (CuTe) kernel line-by-line on Ampere." ([source](https://news.ycombinator.com/item?id=48187017)) "reading the Flash Attention and vLLM kernel source taught me more than any book" ([source](https://news.ycombinator.com/item?id=48178095))

## What people warn about

- Hard to install or get working (2 comments): "you can install flash attention, et al, but if you're on windows" ([source](https://news.ycombinator.com/item?id=46728076)) "haven't gotten FlashAttention2 to work" ([source](https://news.ycombinator.com/item?id=46723568))

## Cite as

Hacker Approved, "FlashAttention", https://hackerapproved.com/p/flashattention/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).