NVIDIA explains the next generation AI-specific Vera CPU, claiming that its speed is more than double that of other CPUs.

NVIDIA announced that the benchmark test results of DeepInfra show that the speed of NVIDIA Vera CPU is more than twice that of other CPUs, and can support more concurrent AI agents. Cloud platform DeepInfra is an early participant in NVIDIA's open AI ecosystem, independently designed and ran benchmark tests based on its production AI agent infrastructure. DeepInfra processes nearly 50 trillion tokens per week, with about 30% being driven by the agent system. Its cloud platform is built specifically for high-throughput AI inference. Benchmark tests show that, at the same level of service quality, it supports up to 1.6 times more concurrent AI agents, with a speed boost of up to 2.2 times, while also improving infrastructure utilization and cost efficiency. These results indicate that NVIDIA Vera CPU can achieve cost efficiency, low latency, and throughput required for production AI agents. As AI agents take on more complex reasoning, planning, tool usage, and data flow, CPU performance becomes increasingly important in coordinating work around each model call. As part of NVIDIA's extreme co-design strategy for AI factories, Vera CPU is designed specifically for agent-workload performance. The benchmark test of DeepInfra highlights how NVIDIA Vera CPU helps cloud service providers improve infrastructure utilization, increase cost efficiency, and support more concurrent AI agents at the same level of service quality.