Nvidia unveiled its Vera CPU at Hot Chips 2026 at Stanford University featuring 88 custom Olympus cores and LPDDR5X memory delivering up to 1.2 TB/s of bandwidth. According to Nvidia, Vera compiles the Linux kernel 14 to 22 percent faster than AMD’s EPYC 9655P and delivers a 1.8x improvement in task completion time on agentic AI workloads compared to competing x86 CPUs. The partnership with SpaceXAI plans to deploy Vera CPUs at scale for Grok workloads, including an orbital version of Vera Rubin NVL72 for the Starmind satellite by Q4 2027. This presentation puts direct pressure on AMD, whose Venice CPU line serves as the explicit benchmark Nvidia is measuring against.
Source: Read the original article
