NVIDIA Processor

NVIDIA Grace CPU

72 ARM Neoverse V2 cores per CPU Cores • NVIDIA

AI Workstation Builders 24+ Years IT Infrastructure Custom Configurations Available
Cores
72 ARM Neoverse V2 cores per CPU
TDP
Approximately 250W per CPU
Memory
LPDDR5X (up to 480GB per CPU at 546 GB/s)

Detailed Specifications

  • Cores: 72 ARM Neoverse V2 cores per CPU
  • Architecture: ARMv9.0-A (Neoverse V2)
  • Memory: LPDDR5X (up to 480GB per CPU at 546 GB/s)
  • Interconnect: NVLink-C2C to Blackwell GPUs (900 GB/s coherent)
  • Tdp: Approximately 250W per CPU
  • Form Factors: Grace Hopper Superchip, Grace Blackwell Superchip, DGX Spark, DGX Station GB300

Best For

  • AI inference with unified CPU-GPU memory via NVLink-C2C
  • Data preprocessing at GPU memory bandwidth speeds
  • Energy-efficient ARM-based AI infrastructure
  • NVIDIA DGX and HGX platforms
  • Sovereign AI and government AI infrastructure
  • HPC workloads requiring high memory bandwidth

Available in PTG Systems

The NVIDIA Grace CPU is available in Petronella-built AI workstations and servers, configured and optimized for your workload.

  • DGX Spark (Grace + Blackwell)
  • DGX Station GB300 (Grace + Blackwell Ultra)
  • DGX B200/B300 (Dual Grace + 8x Blackwell)

Why Buy from Petronella

PTG deploys NVIDIA Grace CPU systems for organizations needing the tightest CPU-GPU integration. NVLink-C2C eliminates the PCIe bottleneck, delivering 7x more bandwidth than PCIe Gen 5 for data-intensive AI.

  • Custom-built systems with tested configurations
  • GPU pairing recommendations for AI workloads
  • Enterprise support and warranty
  • On-site deployment available (Raleigh-Durham area)
  • Remote management and monitoring

Order a NVIDIA Grace CPU System

Talk to our hardware team for custom configurations, pricing, and availability.