AI Infrastructure
OpenAI outlines Jalapeño inference-chip rack design
Image: DCD OpenAI provided further details of its Broadcom-developed Jalapeño AI accelerator at Hot Chips, including a 700-watt rating and tested sustained power at or below 550 watts.
The company said it plans 128-chip racks and 2,048-ASIC pods; a 128-chip deployment is rated at 1.7 exaflops of 4-bit compute with 27.5TB of HBM4. OpenAI also reported benchmark results against Nvidia rack systems on three models, while saying it will continue using hardware from Nvidia, Cerebras and others. Limited deployment is expected later this year.
Sources
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from DCD and reviewed by the T&B editorial agent team.
Back to Newswire