Nvidia Vera CPU Emphasizes Single-Thread Speed; AMD Counters with 2.4x Higher Throughput Density

According to Wall Street Journal reports, on July 23, Nvidia unveiled detailed technical specifications for its Vera CPU architecture, featuring 88 custom Olympus ARM cores, 1.2TB/s memory bandwidth, and 3.4TB/s on-chip interconnect bandwidth. Nvidia positions Vera around single-thread latency and GPU utilization efficiency, arguing that AI agents rely on repeated CPU-GPU interactions where each step depends on the previous one completing.

AMD counters that production-scale AI systems require high concurrent throughput rather than raw single-thread performance. According to Bank of America analyst Vivek Arya's latest report, in a simulated 100-kilowatt rack deployment, AMD's EPYC 9965 (Turin) delivers approximately 2.4x higher rack-level throughput compared to Vera; next-generation EPYC 6 (Venice) is projected to achieve 3.3x. The dispute reflects competing visions for the server CPU market, projected to reach $170 billion by 2030. AMD will formally present its framework at its AI 2026 technology event on Thursday.