Key information
OpenAI is deploying its new Jalapeño ASIC internally alongside AMD EPYC Turin host systems, each equipped with 1.5TB of memory. Richard Ho, VP and Head of Hardware at OpenAI, explained in an interview that selecting Turin was a pragmatic choice aimed at de-risking the design process and moving quickly.
Ho stated that Nvidia’s new Vera CPU is considered a little bit behind on maturity as a standalone option for this specific project. Additionally, partners already had experience working with Turin, and practical factors such as Turin chips being socketed rather than board-mounted influenced the hardware team's decision.
While OpenAI is listed among customers exploring Nvidia's Vera CPU and is an Arm AGI deployment partner, the organization ultimately prioritized platform maturity and risk reduction for the rack-scale Jalapeño deployment. Future deployments might utilize different host systems, but Turin was determined to be the correct choice for the current program.
