Zyphra releases ZYFA FP8 FlashAttention kernels, up to 1.76× faster on AMD MI355X

ZYFA provides open‑source FP8 and BF16 prefill attention kernels for AMD Instinct MI355X, running 1.16‑1.76× faster than AMD’s AITER and delivering 62‑78% lower error at head dimension 128.

Benchmarks are the company's own; not independently verified.

Why it matters

Incorporate ZYFA’s FP8 speedup (up to 1.76×) into product roadmaps for AMD‑based inference services to meet performance targets.

For product managers

For your product

What does this mean for your product?

Try it on your product

Read at MarkTechPost

More from this issueAll AI news from 7 Nov →

Thinking Machines Lab releases Inkling 2 Small 594B MoE with 17B activeAvailable now

OpenAI valued at $1.25 trillion after $11 billion funding round

Amazon launches text‑based Alexa shopping assistant for US customersAnnounced

Ricursive Intelligence releases Ripple-65B Preview, tops open US models on TerminalBenchAvailable now

Get this every morning, written for your product.

Deltaday reads the whole AI news cycle and tells you what changed, whether it's real yet, and what to do about it.

Start free See how it works