OpenAI Deploys Jalapeño ASIC with AMD EPYC Turin Hosts
OpenAI's new Jalapeño ASIC is being deployed internally alongside AMD EPYC Turin hosts, each with 1.5TB of memory, according to SemiAnalysis and confirmed by OpenAI's hardware VP.
OpenAI's new Jalapeño ASIC is being deployed internally alongside AMD EPYC Turin hosts, each with 1.5TB of memory. The deployment was revealed by SemiAnalysis and confirmed by OpenAI's hardware VP, Robert Ho.
Ho described the choice to use Turin as 'pragmatic' and said that Nvidia's Vera standalone is 'a little bit behind on that maturity level.'
In the past, Nvidia claimed its Arm-based Vera CPU offers a 1.8x improvement over a competing AMD Turin chip, but a benchmark suite showed Vera just 3% ahead of AMD's EPYC 9755. Nvidia has also listed OpenAI among the customers 'exploring the Vera CPU.'
Separately, Arm has claimed its AGI CPU provides more than twice the performance of modern x86 platforms based on internal estimates, and OpenAI is one of Arm's deployment partners for AGI.
Quick answers
What is OpenAI's Jalapeño ASIC?
OpenAI's Jalapeño ASIC is a new custom chip being deployed internally alongside AMD EPYC Turin hosts.
Why did OpenAI choose AMD EPYC Turin for its Jalapeño ASIC deployment?
OpenAI's hardware VP Robert Ho called the choice 'pragmatic' and said Nvidia's Vera standalone is 'a little bit behind on that maturity level.'
How much memory do the AMD EPYC Turin hosts have?
Each AMD EPYC Turin host has 1.5TB of memory.