Nvidia's Vera server recorded a PyBench score 4% higher on average than an AMD EPYC 9755 system, providing a Python-workload comparison across two servers with different hardware configurations.
The Vera setup paired two processors for a total of 176 cores and 352 threads. AMD's configuration joined two EPYC 9755 chips, producing 256 cores and 512 threads. Both machines had 1.5 terabytes of memory, though the memory types and speeds differed.
The result captures the complete server setups and cannot isolate processor performance. The benchmark covers Python code, leaving unanswered how much time either setup needs for a complete AI-agent task. EconoTimes previously examined a Vera benchmark that did not test live-agent performance.
A separate estimate based on SPEC CPU 2026 integer throughput gave Vera a 1.5% edge over EPYC 9755, though the result was not submitted to SPEC. Nvidia's earlier 3% estimate came from a different test, making it a separate result. Nvidia positions Vera for data-center computing in agentic artificial intelligence and reinforcement learning.
