NVIDIA's Vera CPU Runs More AI Agents Per Watt Than x86 Chips
NVIDIA has published test data for its Vera CPU on AI agent workloads, saying the chip delivers leading agent speed and per-watt throughput at the same time, helping data centers run more agents on limited power.
Unlike traditional generative AI, agents also need to search for information, query databases, and test code, which relies heavily on CPUs. NVIDIA believes data center bottlenecks are shifting from raw computing power to balancing speed, concurrency, and energy use.
In tests based on a 100-megawatt data center, Vera CPU supports 1.6 times more agents per watt than the fastest x86 CPU, and runs agents 2 times faster than the most energy-efficient x86 CPU.
In partner tests, Perplexity ran coding tasks 1.5 times faster, DeepInfra got up to 2.2 times faster orchestration, Daytona launched 1,000 agent sandboxes in just 10.8 seconds, and Modal hit twice the per-core throughput of x86.
Serverless platform for AI and data teams to run CPU/GPU workloads at scale with sub-second cold start and zero config files
Get instant, comprehensive answers to any question with an AI-powered answer engine that summarizes the web so you don't have to dig through countless tabs.
ByteDance AI chatbot development platform with LLM integration, plugin ecosystem and cross-platform deployment to Doubao and Feishu
Comments
Loading comments…