NVIDIA Vera Enhances AI-Native Storage Processing Efficiency, Covering Key Stages Such as Encryption, Compression, and Data Recovery

TechFlow reports, August 05, NVIDIA released Vera BlueField-4 STX storage processor benchmark results, stating that in AI-native storage scenarios, it can accelerate critical data path processing tasks such as encryption and decryption, Reed-Solomon recovery, CRC32C integrity check, and compression and decompression. The article shows that Vera achieved higher throughput performance compared to comparable x86 processors in multiple tests, with improvements in recovery, integrity check, and compression-related tasks being more significant.

The article points out that as agent AI workload concurrency and context scale increase, storage systems, in addition to basic read/write operations, must continuously undertake CPU-side processing tasks such as data protection, verification, compression, and recovery. NVIDIA believes that Vera can improve storage processing efficiency within limited power and thermal budgets through higher single-core performance, memory bandwidth, and multi-threading capability, and help AI-native storage platforms support higher concurrent data streams and service density.