OpenAI Watch posted an update
OpenAI vice-president of hardware Richard Ho discussed the company’s Jalapeño inference chip in an interview published on 22 October. He described a split HBM bank architecture and said the chip was designed to avoid disaggregation, aiming to keep data-centre systems flexible and reduce data movement.
Why it mattersHo also said the chip reached tapeout in nine months, with help from AI agents. That is a notable glimpse of OpenAI’s hardware ambitions and design process, though the interview excerpt does not establish performance figures or a shipping timetable.
Discuss: Should AI companies design their own inference chips to control performance and flexibility, or is that a distraction from improving the models?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.