NVIDIA Vera CPU with 88 Olympus cores, high-bandwidth LPDDR5X memory, and NVLink-C2C for agentic AI, reinforcement learning, data processing, and accelerated host compute.
Fast Shipping
100k+ SKUs
8/5 Support
Certified
NVIDIA Vera CPU for Agentic AI and AI Factory Host Compute | VERA-CPU
NVIDIA Vera CPU (VERA-CPU)
NVIDIA Vera is a purpose-built data center CPU for agentic AI, reinforcement learning, data processing, and the CPU-side work that keeps accelerated systems utilized. Vera combines 88 custom NVIDIA Olympus cores with high-bandwidth LPDDR5X memory and NVLink-C2C connectivity for both standalone CPU deployments and host-compute roles in Vera Rubin and HGX systems.
Technical Specifications
| SKU | VERA-CPU |
|---|---|
| Product Type | AI CPU / Host Compute |
| CPU | 88 NVIDIA Olympus cores / 176 threads with Spatial Multithreading |
| Memory | Up to 1.5 TB LPDDR5X |
| Memory Bandwidth | Up to 1.2 TB/s |
| CPU-GPU Interconnect | NVIDIA NVLink-C2C up to 1.8 TB/s coherent bandwidth |
| Deployment Modes | Standalone CPU; host CPU for accelerated AI/HPC systems |
| Availability Status | Limited / Allocation-Based |
Deployment and Compatibility Considerations
- Validate OEM server or MGX platform support before deployment.
- For accelerated systems, confirm the target Rubin/HGX configuration and NVLink-C2C topology.
- Plan memory capacity and bandwidth around agent sandboxing, data pipelines, ETL, KV-cache management, analytics, and orchestration.
- Confirm networking, storage, power, cooling, firmware, and software support for the selected OEM implementation.
Common Use Cases
- Agentic AI tool execution and sandbox workloads
- Reinforcement-learning environments and evaluation loops
- AI data processing, analytics, ETL, and orchestration
- Host CPU for Vera Rubin and HGX accelerated systems
- HPC and high-memory-bandwidth CPU workloads
Frequently Paired Infrastructure
- NVIDIA Rubin GPUs and HGX Rubin NVL8 platforms
- NVIDIA ConnectX-9 SuperNICs
- NVIDIA BlueField-4 DPUs and STX storage processors
- Spectrum-X or Quantum InfiniBand scale-out fabrics
- High-throughput AI storage
Official NVIDIA Documentation
Frequently Asked Questions
What makes NVIDIA Vera different from a conventional server CPU?
Vera is designed around the CPU work created by agentic AI and accelerated computing, emphasizing high single-thread performance, high LPDDR5X memory bandwidth, and high-bandwidth coherent connectivity to NVIDIA GPUs.
Can Vera be used without a Rubin GPU?
Yes. NVIDIA positions Vera both as a standalone CPU platform and as a host CPU for accelerated systems such as Vera Rubin NVL72 and HGX Vera Rubin NVL8.
| product type | AI CPU / Host Compute |
|---|---|
| Architecture / Series | NVIDIA Vera |
| CPU Configuration | 88 Olympus Cores / 176 Threads |
| Memory | Up to 1.5 TB LPDDR5X |
| Availability Status | Limited / Allocation-Based |
