1U NVIDIA Grace CPU server
Arm based server with the NVIDIA Grace CPU Superchip, 144 Neoverse V2 cores and up to 960 GB of LPDDR5X at about 1 TB/s, for bandwidth bound HPC and data.
- CPU
- 1 x NVIDIA Grace CPU Superchip, 144 Arm Neoverse V2 cores
- Memory
- Up to 960 GB LPDDR5X with ECC, soldered, about 1 TB/s bandwidth class
- Storage
- 2 x M.2 boot, up to 8 x E1.S or U.2 NVMe
- Networking
- 1 x 400 Gb/s ConnectX-7 or BlueField-3, 2 x 25 GbE
- Software
- Arm Linux distributions, NVIDIA HPC SDK, container runtimes
- Power
- Redundant PSUs, about 1 kW per node planning figure
Overview
The NVIDIA Grace CPU server brings 144 Arm Neoverse V2 cores and LPDDR5X memory at about 1 TB/s to a conventional 1U chassis. The memory bandwidth per core is well beyond what DDR5 systems provide, which is why Grace suits memory bound HPC codes, graph analytics, in memory databases and the CPU side of data pipelines that feed GPU clusters.
It is listed as available on request because Grace systems are built in smaller volumes than x86 servers and configurations are fixed at order, including the soldered memory capacity. We recommend it when your software already runs on Arm, or when a benchmark on the memory bandwidth shows a clear win over the EPYC 9005 and Xeon 6 classes.
Networking and storage follow the rest of the CPU range: a 400 Gb/s ConnectX-7 or BlueField-3 adapter, dual 25 GbE and up to eight NVMe drives. We can arrange evaluation access before purchase and deliver the node imaged with an Arm Linux distribution and the NVIDIA HPC SDK.
Use cases
- Memory bandwidth bound HPC
- Graph analytics and in memory databases
- Arm native cloud services
- Data pipelines for GPU clusters
Specifications
| CPU | 1 x NVIDIA Grace CPU Superchip, two Grace dies with NVLink-C2C, 144 Arm Neoverse V2 cores |
|---|---|
| System memory | Up to 960 GB LPDDR5X with ECC, fixed at order, about 1 TB/s class aggregate bandwidth |
| Architecture | Arm v9, SVE2, runs standard Arm Linux and NVIDIA HPC SDK |
| Boot | 2 x M.2 NVMe, mirrored |
|---|---|
| Data | Up to 8 x E1.S or U.2 NVMe |
| RAID | Software RAID |
| Fabric | 1 x 400 Gb/s ConnectX-7 or BlueField-3, PCIe Gen5 |
|---|---|
| Frontend | 2 x 25 GbE onboard |
| Management | Dedicated 1 GbE BMC port |
| Form factor | 1U rackmount, air cooled; 2U two node variants available |
|---|---|
| Power supplies | Redundant 1,200 W class PSUs, 200 to 240 V input |
| Planning power | About 1 kW per node |
Figures describe the system class. Exact configuration, weights and power draw are confirmed on the quotation.
What's included
- Rails, power cords and cable management arm
- Burn in and acceptance test report
- Operating system imaging and firmware baseline on request
- 3 year warranty, options to 5 years with advance replacement
- Arm Linux image and NVIDIA HPC SDK installed on request
Compliance
This product is not a data center GPU system and does not require NVIDIA end user registration. Standard export documentation applies. Read about export compliance.
Related
Related systems
1U dual socket EPYC 9005 server
- CPU
- 2 x AMD EPYC 9005 series
- Memory
- 24 x DDR5 RDIMM, 12 channels per socket
- Expansion
- 2 x PCIe Gen5 x16 slots, OCP 3.0 mezzanine
- Chassis
- 1U, about 1 kW, air cooled
High frequency single socket server
- CPU
- 1 x AMD EPYC 9005 high frequency model
- Memory
- 12 x DDR5 RDIMM, up to 768 GB
- Use
- Per core licensed databases, trading
- Chassis
- 1U, about 0.6 kW, air cooled
NVIDIA GB300 NVL72 rack
Allocation required- GPUs
- 72 x NVIDIA Blackwell Ultra, 288 GB HBM3e each
- CPUs
- 36 x NVIDIA Grace, 2,592 Arm Neoverse V2 cores
- NVLink
- 130 TB/s total NVLink bandwidth across all 72 GPUs
- Chassis
- Single rack, power not yet published, liquid cooled
Compare it with the next class up or down
The nearest classes we quote are the 1U dual socket EPYC 9005 server and the High frequency single socket server. If neither is quite right, tell us the workload and we specify the system between them.
- (858) 613-1818
- sales@computermodules.com
- Monday to Friday, 8am to 5pm Pacific