NVIDIA Vera Storage Processor Performance

8/3/2026

This post introduces and benchmarks the NVIDIA Vera BlueField-4 STX Storage Processor, highlighting its performance advantages over x86 CPUs for AI-native storage workloads. It details the Vera CPU architecture, including its 88 Armv9.2 cores, Spatial Multithreading, Scalable Coherency Fabric, and SOCAMM2 LPDDR5X memory, and quantifies its throughput gains in encryption (up to 1.43x), decryption (up to 1.29x), Reed-Solomon recovery (up to 3.26x), CRC32C integrity checking (up to 3.67x), compression (up to 3.29x), decompression (up to 1.72x), and multi-stage pipeline operations (up to 3.21x). The post emphasizes how these improvements enable higher service density and more concurrent data flows for agentic AI workloads by reducing CPU, power, and cooling overhead in the storage data path.

7/7/2026

This post details how the NVIDIA Vera CPU architecture, with its monolithic compute die, unified cache, and Scalable Coherency Fabric, delivers 1.8x faster sustained per-core performance under full socket load, directly improving reinforcement learning (RL) training throughput, policy gradient quality, and environment rollout completion rates. It achieves 40% lower peak loaded latency and over 3x the per-core memory bandwidth at less than half the power of traditional x86 data center CPUs, optimizing agentic inference and interactive deployment responsiveness at scale. By minimizing CPU-side stalls, reducing context reconstruction from KV-cache evictions, and maximizing throughput, NVIDIA Vera CPU enables higher GPU utilization and efficiency in densely loaded agentic AI factories, directly increasing overall system productivity and service level agreement adherence.

6/1/2026

This post introduces the NVIDIA Vera CPU as a new design point for AI factories, specifically addressing the increased importance of CPU performance in agentic AI and reinforcement learning workloads. It details how the Vera CPU, with its 88 NVIDIA Olympus cores and 1.2 TB/s LPDDR5X memory bandwidth, delivers high per-core performance, concurrency, and energy-efficient memory bandwidth. Key architectural features like the neural branch predictor, wide decode unit, deep out-of-order engine, graph prefetcher, and NVIDIA Scalable Coherency Fabric are highlighted for their role in accelerating agentic tasks such as sandboxed code execution, data retrieval, and orchestration. Performance benchmarks show over 1.8x higher agentic sandbox performance compared to x86-based architectures.