Why Vera matters
NVIDIA's data center CPU Vera is the latest step in the company's years-long AI infrastructure push. According to Tom's Hardware's independent testing, the chip delivers strong results in the fast memory access and Python performance that agentic AI workflows demand.
Technical specifications
Vera is built on Olympus, NVIDIA's first custom core design. The 88-core, 176-thread chip uses the Armv9.2-A architecture. It has 2 MB of L2 cache per core and 164 MB of L3 per CPU. On the memory side it uses replaceable SOCAMM2 LPDDR5X at 9600 MT/s, with total bandwidth up to 1.2 TB/s and roughly 14 GB/s per core.
Comparison with rivals
In testing, Vera faced AMD EPYC 9965 and 9755 plus Intel Xeon 6980P on dual-socket reference platforms. Results show Vera often matching or beating its rivals in general-purpose server workloads. x86 chips keep an edge in frontend workloads thanks to core density, while Vera pulls ahead in data-rich backend operations.
Market context
Vera enters a competitive server market where custom Arm chips are penetrating data centers and AMD and Intel's x86 chips still hold a strong customer base. The test shows Vera's position before AMD's Venice and Intel's Diamond Rapids generations arrive.
