# TriAttention — trigonometric KV cache compression

> NVIDIA's new attention technique compresses the KV cache by exploiting a geometric property of Q/K vectors before RoPE is applied.

- Org: NVIDIA
- Date: 2026-04-07
- Category: algorithm, paper
- Source: https://arxiv.org/abs/2604.04921
- Canonical page: https://ai-tldr.dev/releases/nvidia-triattention/

## In short

A new attention scheme that compresses the KV cache by exploiting Q/K vector concentration in pre-RoPE space. Targets long-context inference memory bottlenecks.

## What it is

TriAttention is a new variant of attention designed to make long-context LLM inference cheaper. It targets the KV cache — the chunk of GPU memory that stores per-token keys and values during generation, which dominates memory cost as context length grows past 100k tokens.

## How it works

The authors observe that Q and K vectors concentrate in specific angular regions in the pre-RoPE (rotary positional embedding) space. TriAttention uses this concentration to estimate the importance of each cached key cheaply, without computing the full attention matrix. Less-important keys get evicted or compressed; the rest stay in cache. The trigonometric framing comes from working in the angular RoPE space.

## Why it matters

Long-context models all hit the same wall: the KV cache eats GPU memory faster than the model's parameters do. KV compression that does not degrade quality is the bottleneck on cheap 1M-context serving. If TriAttention's quality holds at scale it pushes that wall back significantly.

## Who it's for

ML systems researchers, anyone building inference infrastructure for long-context models.

## Links

- [arXiv](https://arxiv.org/abs/2604.04921)
- [HF paper page](https://huggingface.co/papers/2604.04921)