High-Density CPU Compute
Reinforcement learning and agentic AI rely on continuous feedback loops, where models trained on GPUs act within CPU-hosted environments that execute code, call tools, and evaluate results at scale. The NVIDIA Vera CPU Rack integrates 256 Vera CPUs and supports more than 22.5K concurrent environments alongside NVIDIA Vera Rubin NVL72 systems, shortening evaluation cycles and maximizing AI factory output.
Platform Overview
Built on NVIDIA MGX™, the Vera CPU Rack scales agentic AI performance across the data center through a dense, liquid-cooled architecture. With 256 Vera CPUs, high-bandwidth SOCAMM2 LPDDR5X memory, NVIDIA BlueField®-4 DPUs, and NVIDIA Spectrum-X™ Ethernet networking, it delivers CPU capacity in step with NVIDIA Vera Rubin NVL72 deployments.
Features
The NVIDIA Vera CPU Rack combines Vera CPUs with next-generation networking in a rack-scale platform purpose-built for the concurrent execution and evaluation demands of enterprise AI factories. With 100% direct liquid cooling across CPUs and key components, the system delivers sustained performance under continuous, production-scale AI workloads.
IGS-V256
- 256 NVIDIA Vera CPUs
- Up to 384 TB SOCAMM2 LPDDR5X memory, up to 1.5 TB per CPU
- Up to 1.8 TB/s coherent CPU-to-CPU bandwidth via NVIDIA NVLink-C2C
- Dense CPU capacity for agentic AI and reinforcement learning (RL) sandboxes
- Scales CPU capacity alongside NVIDIA Vera Rubin NVL72 for AI factories
- Cableless compute tray design for rapid service and deployment
- 100% direct liquid cooling solution
