NVIDIA’s new Vera CPU has delivered up to twice the task throughput of comparable Intel systems in its first independent AI agent benchmark. Cloud computing startup Modal published the results after receiving its first Vera server from NVIDIA and testing it against Intel Granite Rapids, AMD Zen 5, and AWS Graviton processors.
NVIDIA first unveiled Vera at GTC 2026 in March as a general-purpose data center CPU built specifically for agentic AI workloads. The chip uses NVIDIA’s custom Olympus architecture with 88 Arm-based cores and spatial multithreading. This design physically partitions execution units and caches so each thread progresses without waiting on shared resources. In addition, the chip offers up to 1.2 TB/s of LPDDR5X memory bandwidth, roughly 13.6 GB/s per core under full load.
Modal engineer Erik Dunteman ran the test using a benchmark called Agentless Agent Bench. It covered 50 tasks including code compilation, file searches, Git operations, and device emulation. The tests ran in prewarmed Docker containers without network calls. Vera consistently maintained higher throughput as concurrent tasks increased. However, Dunteman noted that “Vera often performed better than the other CPUs, but not always.”
Modal attributed the throughput gains partly to Vera’s high memory bandwidth. Competing systems hit throughput limits sooner because individual AI agent workloads can spike to 10 times their normal CPU usage. Furthermore, NVIDIA analyzed telemetry from over 163,000 agent sessions and found that more than 97% followed unique execution patterns. As a result, the company designed Vera to maintain single-task performance while handling unpredictable concurrent operations.
Meanwhile, Modal recently raised $750 million at a $15.75 billion valuation. The startup expects $1 billion in sales over the next year and its revenue has grown more than 15 times year over year. CEO Erik Bernhardsson told Reuters that Modal will continue using Intel and AMD processors alongside Vera.
Oracle Cloud Infrastructure, CoreWeave, Lambda, and Nebius have also announced plans to deploy Vera-based systems. In addition, NVIDIA named Anthropic and OpenAI as AI developers exploring the chip. Consequently, Vera marks NVIDIA’s most direct challenge yet to Intel and AMD in the data center CPU market.
