OpenAI's Jalapeño chip: beats Blackwell perf/watt, but bets on AI-written kernels
nrehiew_ · x · 2026-10-11
nrehiew published notes on OpenAI's in-house "Jalapeño" chip. Headline claim: it appears to beat Blackwell on perf/power, with two caveats — it uses HBM4, making Rubin the fairer comparison, and testing wasn't done at very long context. The core design trade-off exposes more of the physical machine to software to maximize effective memory bandwidth, making kernels harder to write; OpenAI is betting AI-assisted kernel generation/search can absorb that complexity.
Related event: Researcher's notes unpack OpenAI's rumored Jalapeño chip design tradeoffs(8 posts)→
More from Infra
- 5 LLM deployment patterns every AI engineer should know, from API calls to hybrid — goyalshaliniuk · 2026-10-11
- Neoclouds that just buy power and GPUs will lose to software-first players, says Beam founder — edgarpavlovsky · 2026-10-11
- M5 Ultra 256GB vs dual DGX Spark: local LLM buyer weighs throughput vs memory — MasterNomie · 2026-10-11
- AI Infrastructure Borrowing Slumps 80% in Three Months, From $113B to $23B — TansuYegen · 2026-10-11
- OpenAI, Microsoft, Google and 4 others signed a pledge to cover AI data centers' grid costs — ChrisGPT · 2026-10-11
- One-town monopolies: Yiwu makes 80% of Christmas decor, every EUV machine comes from Veldhoven — sahilypatel · 2026-10-11