AI
OpenAI's Jalapeño inference chip, co-designed with Broadcom, packs 216 GiB of HBM4 and hits 27 EFLOP/s per 2,048-accelerator system. Its VP of Hardware explains how OpenAI's own models helped design it.
Loading…
Search
No phones and 1 story for “Jalapeño”
AI
OpenAI's Jalapeño inference chip, co-designed with Broadcom, packs 216 GiB of HBM4 and hits 27 EFLOP/s per 2,048-accelerator system. Its VP of Hardware explains how OpenAI's own models helped design it.