OpenAI’s Jalapeño ASICs are deployed alongside AMD EPYC ‘Turin’ CPUs as hosts, not Nvidia's Vera
OpenAI chose to pair rack-scale deployments of its Jalapeño ASIC with AMD EPYC Turin CPUs, not the wave of high-performance agentic chips like Arm’s AGI or Nvidia’s Vera.
OpenAI's Jalapeño ASIC has been deployed alongside AMD EPYC Turin CPUs, each equipped with 1.5TB of memory, according to SemiAnalysis. This announcement came following the unveiling of the chip, and in an interview, Richard Ho, VP and Head of Hardware at OpenAI, discussed the rationale behind utilizing the Turin architecture. Ho stated that AMD's Turin was "pragmatic" due to Nvidia's Vera CPU being somewhat behind in terms of maturity.
He also mentioned that the decision was made to de-risk the design process and ensure it could be completed quickly. Ho explained that OpenAI wanted to achieve aggressive performance and cost goals without taking unnecessary risks. While Arm-based CPUs are prevalent among hyperscalers like Google Cloud and AWS, Nvidia's Vera and Arm's AGI are considered agentic CPUs, designed to accelerate complex reasoning in agentic loops.
Ho further explained that OpenAI chose Turin for its maturity and to align with the experience of its partners. Although Arm-based CPUs are widely used in data centers, including Google Cloud's Axiom and AWS's Graviton, Vera and AGI are seen as a new approach focusing on peak performance instead of core density and efficiency. OpenAI's decision to pair Jalapeño with Turin was driven by a desire to minimize risk, given the maturity of the Turin platform.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
