Nvidia’s first in-house CPU benchmarked, beats x86 and ARM chips alike but only in Nvidia-sanctioned tests
Nvidia’s first internally-designed CPU, called Olympus, has been benchmarked as part of the company’s Vera Rubin AI platform. The chip outperformed competing x86 and ARM processors in testing, marking a significant milestone for the company’s processor development efforts.
What sets this apart from Nvidia’s previous CPU offerings is that Olympus was designed entirely in-house by Nvidia engineers, rather than using third-party CPU core designs. While Nvidia’s earlier Grace CPU relied on cores designed by Arm itself, the new Olympus core represents a true in-house design that operates using the Arm instruction set architecture.
The first generation Vera chip features 88 cores and is positioned as part of Nvidia’s next-generation AI platform infrastructure. However, there’s an important caveat: while the testing was conducted independently, Nvidia controlled the selection and nature of the workloads used in the benchmarks. This means the tests were specifically designed to showcase Vera’s strengths rather than providing a comprehensive, industry-standard comparison.
For now, Vera appears primarily targeted at data center and AI infrastructure markets, where Nvidia already dominates with its GPU offerings. The achievement represents a major shift for Nvidia, expanding its capabilities beyond GPU design into custom CPU architecture.