# matx.com

> AI-optimized mirror of matx.com containing 10 pages totalling 9,403 words of clean markdown content, structured data, and semantic HTML. Original source: https://matx.com/. Last updated: 2026-04-18T23:15:24.110Z.

Each page is available as HTML (with JSON-LD structured data) and Markdown (text-only, ideal for LLMs and RAG).

## Homepage

- [MatX: High-throughput chips for LLMs](/site-root.html): Our goal is to make the best chips physically possible for the large model needs of frontier labs. (179 words)

## Articles & Blog Posts

- [Prioritize values over keys: faster attention with many sparsely accessed value heads | MatX](/research/smva/index.html): During Transformer decoding, KV cache size and memory bandwidth requirements can limit overall throughput. Multi Query Attention (Shazeer, 2019) is a powerful technique to mitigate this, but some m... (2,084 words)
- [Speculative Decoding with Blockwise Sparse Attention | MatX](/research/sd_nsa/index.html): Speculative decoding (SD) and blockwise sparse attention both accelerate LLM decoding, but when combined naively, the KV cache may lose sparsity during the verification step of SD. We show that for... (1,842 words)
- [Simple and fast Rust deriving using macro_rules | MatX](/research/rules_derive/index.html): Rust’s #[derive] saves a ton of boilerplate, but custom derive implementations typically require complicated procedural macros—a high-friction process that limits adoption. We have released a libra... (1,225 words)
- [Join Us | MatX](/jobs/index.html): We build the world's best chips and racks for training and serving large language models. Our work spans the entire range from silicon and rack design, through compilers and kernels, to ML models.W... (916 words)
- [Optimize for inference too, not just training FLOPs | MatX](/research/lifetime_llm_cost/index.html): Large Language Models (LLMs) have been shaped by the scaling laws, as established by Kaplan et al., 2020 and Hoffmann et al., 2022. They guide us to design models that optimize for training cost bu... (1,503 words)
- [Research | MatX](/research/index.html) (95 words)
- [Future leakage in block-quantized attention | MatX](/research/leaky_quantization/index.html): Quantizing attention improves efficiency on two fronts: the model has higher compute throughput, and loads fewer bytes per key/value. However, training with block quantized attention can break caus... (985 words)
- [Introducing seqax: A Simple and Efficient LLM Research Codebase | MatX](/research/seqax/index.html): We’re excited to announce seqax, a research-focused LLM codebase that is simple, efficient, and performs well on up to 100 GPUs or TPUs. Everything you need to edit, from the math, to parallelism, ... (560 words)

## About Pages

- [404: NOT_FOUND](/about/index.html): 404: NOT_FOUND Code: NOT_FOUND ID: iad1::8nd4c-1776550162263-40812ab45af5 (14 words)

## Resources

- [Full Page Index](/index.html): Browse all cached pages with rich metadata
- [About This Cache](/content/about.html): Methodology, technical details, and usage guidelines
- [XML Sitemap](/content/sitemap.xml): Machine-readable sitemap for crawler discovery
- [Robots.txt](/content/robots.txt): Crawler directives