Nvidia released benchmarks for its Vera CPU, the company’s first data-center processor, showing performance up to 3.67 times higher than a standard x86 CPU for AI-native storage workloads. The chip, embedded in the BlueField-4 STX, features 88 custom Olympus cores based on Armv9.2 architecture with 176 threads and SOCAMM2 LPDDR5X memory delivering 1.2 TB/s aggregate bandwidth. Results also show 3.29 times faster data compression and 3.26 times faster Reed-Solomon recovery. Commercial launch is scheduled for fall 2026, with Vera designed to serve as host CPU for Nvidia’s Rubin GPU platforms.
Source: Read the original article
