As AI applications evolve toward agentic systems that reason, execute code, and invoke tools, CPU performance is becoming increasingly critical. While GPUs accelerate model inference, CPUs handle orchestration, tool execution, and data retrieval between model interactions. As these workloads scale, CPU performance directly impacts agent responsiveness and overall AI system efficiency.
Today, Oracle Cloud Infrastructure (OCI) is announcing OCI Compute with NVIDIA Vera CPUs, bringing purpose-built CPU infrastructure to the workloads that keep AI systems moving. NVIDIA Vera is the CPU for agents, delivering 80% faster agentic CPU performance for longer-thinking agents, faster-learning RL systems, and data-driven AI factories. 150,000 CPUs will initially be made available across Oracle’s regions. Vera extends the NVIDIA compute portfolio on OCI with a standalone CPU platform for the services around AI models, alongside its role as a host CPU in the NVIDIA Vera Rubin NVL72 platform. The expansion complements OCI’s GPU infrastructure with CPU capacity for emerging agentic AI workloads.
A CPU platform built for Agentic AI
NVIDIA Vera combines custom Arm-compatible NVIDIA Olympus cores with a high-bandwidth memory subsystem, purpose-built for the demanding requirements of agentic AI workloads. Its single compute die and unified coherency fabric are designed to deliver consistent, scalable performance across highly concurrent workloads, including agent orchestration, code execution, and AI inference support services.
Each NVIDIA Vera processor features:
- 88 Olympus cores and 176 threads enabled by NVIDIA Spatial Multithreading
- Up to 1.2 TB/s of memory bandwidth for data-intensive workloads
- Second-generation NVIDIA Scalable Coherency Fabric with 3.4 TB/s of bisectional bandwidth
OCI Compute instances powered by NVIDIA Vera feature a dual-processor configuration designed to support demanding agentic AI and enterprise workloads:
- 176 CPU cores and 352 threads across two NVIDIA Vera processors
- 1 TB of LPDDR5X system memory
- 7.68 TB of local NVMe storage for fast data access
- 200 Gbps Acceleron networking for high-throughput connectivity
A flexible approach to AI infrastructure
NVIDIA Vera CPUs are available on OCI through both Bare Metal and Flex Virtual Machine instances, giving customers the flexibility to deploy and scale CPU-intensive AI workloads based on their performance and capacity requirements. As a standalone CPU platform, Vera enables customers to run agentic AI services independently of GPU infrastructure, while complementing GPU-accelerated environments where needed.
Improve responsiveness and GPU efficiency
In agentic AI workloads, CPU-intensive tasks such as tool execution, code execution, and data retrieval often need to complete before the next model inference step can begin. Faster CPU execution reduces these delays, improving end-to-end response times and enabling agents to complete tasks more efficiently. For reinforcement learning workloads, faster environment execution can also accelerate feedback loops and improve training throughput.
NVIDIA reports up to 80% faster agentic CPU performance compared to leading x86 CPUs, helping accelerate CPU-intensive operations that support agentic AI applications. These results are based on NVIDIA platform benchmarks rather than OCI instance-level testing, and actual performance varies by workload and configuration.
By reducing the time spent on CPU-bound operations, Vera can also help improve GPU utilization and reduce pressure on GPU KV-cache capacity. Preserving cached context can minimize redundant computation, allowing GPUs to spend more time generating tokens and less time waiting for CPU-dependent tasks.
Purpose-Built for Modern AI Workloads
NVIDIA Vera’s high-performance CPU architecture and memory bandwidth make it well-suited for a broad range of agentic AI and inference workloads, including:
- Agentic orchestration and tool serving: Accelerate session management, tool execution, and result processing for concurrent AI agents.
- Sandbox execution and reinforcement learning: Support code execution, testing, simulations, and reward processing across isolated environments.
- Retrieval and in-memory analytics: Enable memory-intensive workloads such as vector search, retrieval, feature stores, and real-time analytics.
- GPU fleet management and operational services: Support scheduling, request routing, telemetry, APIs, and data pipelines across GPU-accelerated environments, including those using GPUs beyond NVIDIA Rubin.
Multiple customers are already leveraging NVIDIA Vera CPUs on OCI through early access to some of the first production systems, demonstrating growing demand for purpose-built CPU infrastructure for agentic AI.
“NVIDIA Vera CPUs on OCI deliver the world-class loaded performance needed to execute code and run tools across our rapidly growing agentic sandbox workloads,” said Uday Ruddarraju, chief technology officer of compute at OpenAI. “We are bringing Vera into production on OCI to support these workloads at scale.”
Accelerate Agentic AI with Vera on OCI
With NVIDIA Vera CPUs, OCI continues to expand its AI infrastructure portfolio, delivering the performance, efficiency, and flexibility needed to power next-generation agentic AI workloads. Whether running AI agents, executing code, or supporting GPU-accelerated applications, OCI Compute with NVIDIA Vera helps customers scale AI efficiently.
NVIDIA Vera CPUs on OCI will be available in select regions this quarter. To learn more, visit the OCI Compute page or contact your Oracle representative.