Brand: NVIDIA | Category: CPUs
SKU: NVIDIA-900-21001-0010-000 | Part #: 900-21001-0010-000 | MPN: 900-21001-0010-000
Contact for Pricing — Request a Quote
The NVIDIA Grace CPU Superchip (GH200 NVL2) delivers 144 CPU cores and 2 integrated H100 SXM5 GPUs within a unified memory fabric architected for AI inference and high-performance computing workloads that demand extreme bandwidth efficiency in compact form factors. Two Grace Hopper Superchips are connected via NVLink-C2C (NVL2), creating a tightly coupled compute module that eliminates traditional CPU–GPU latency bottlenecks through 900 GB/s bidirectional interconnect bandwidth per superchip and NVLink 4 GPU-to-GPU peer connectivity at 900 GB/s between the two H100 SXM5 units.
IT infrastructure and AI operations teams will recognize the GH200 NVL2 as a purpose-built alternative to discrete multi-socket server designs, collapsing memory hierarchy into a unified 720 GB pool: 240 GB LPDDR5X CPU memory and 480 GB HBM3 GPU memory. The Arm Neoverse V2 (Armv9) CPU architecture, extended with SVE2 vector instructions, pairs with 4 TB/s HBM3 memory bandwidth per superchip to sustain dense inference and ML acceleration at power levels sustainable in shared data center cooling. Multi-Instance GPU (MIG) support enables up to 7 isolated instances per H100 GPU, maximizing utilization across multi-tenant and containerized workloads. NVIDIA MGX platform compatibility ensures integration into standard enterprise deployment patterns. For organizations seeking to consolidate CPU and GPU compute into a single-module architecture with unprecedented memory bandwidth and low-latency interconnect, contact Omnixon Global to request a quote on part number 900-21001-0010-000.
| Brand | NVIDIA |
| Category | CPUs |
| SKU | NVIDIA-900-21001-0010-000 |
| Part Number | 900-21001-0010-000 |
| Condition | New |
| Interface | NVLink-C2C |
| Manufacturer Part Number | 900-21001-0010-000 |
| Product Name | Grace CPU Superchip (GH200 NVL2) |
| Configuration | 2x Grace Hopper Superchips connected via NVLink-C2C (NVL2) |
| CPU Architecture | Arm Neoverse V2 (Armv9) |
| CPU Cores per Superchip | 72 (144 total across NVL2 module) |
| CPU ISA Extensions | SVE2, Armv9 |
| GPU per Superchip | NVIDIA H100 SXM5 |
| Total GPUs in NVL2 Module | 2x H100 SXM5 |
| GPU Memory per Superchip | 96 GB HBM3 |
| Total HBM3 GPU Memory (NVL2) | 480 GB HBM3 |
| CPU Memory per Superchip | 120 GB LPDDR5X |
| Total CPU Memory (NVL2) | 240 GB LPDDR5X |
| Total Combined Memory (NVL2) | 720 GB |
| CPU-GPU Interconnect | NVLink-C2C at 900 GB/s bidirectional per Superchip |
| HBM3 Memory Bandwidth per Superchip | 4 TB/s |
| CPU Memory Bandwidth per Superchip | 512 GB/s |
| Multi-Instance GPU (MIG) Support | Yes (up to 7 instances per H100 GPU) |
| NVLink GPU-to-GPU (NVL2 peer) | NVLink 4 (900 GB/s bidirectional between the two GPUs) |
| Platform Compatibility | NVIDIA MGX |
| Thermal Design | Supports air-cooled and liquid-cooled server configurations |
The NVIDIA Grace CPU Superchip (GH200 NVL2) is enterprise IT hardware deployed in production data centers, server rooms, and edge sites across the GCC and EMEA region.
Key specifications for the NVIDIA Grace CPU Superchip (GH200 NVL2): interface NVLink-C2C; 3 years support; new condition; processor ARM CPUs with H100 GPUs via NVLink-C2C; support 3 years standard (manufacturer). Manufacturer part number 900-21001-0010-000. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.