Vera CPU Brings Lightning-Fast Storage to AI Workloads

NVIDIA has unveiled significant performance gains in AI-native storage systems using its Vera CPU. The Vera BlueField-4 STX platform, powered by the custom-built Vera CPU, outperforms traditional x86 processors across critical storage operations such as encryption, recovery, and compression.

The benchmarks reveal that the Vera CPU achieves up to 1.43x higher encryption throughput, 3.26x faster recovery using Reed-Solomon encoding, and 3.67x higher data integrity validation throughput compared to x86 CPUs. Its compression capabilities are equally impressive, with up to 3.29x higher throughput for reducing storage footprints and bandwidth usage.

The Vera CPU's architecture features 88 Olympus cores, 176 NVIDIA Spatial Multithreading threads, and a proprietary Scalable Coherency Fabric delivering 3.4 TB/s bandwidth. This design ensures the CPU can sustain high single-thread performance while efficiently handling concurrent workloads, essential for AI factories where storage systems need to manage massive data volumes without creating bottlenecks.