(858) 613-1818Quote list 0

1U single socket EPYC 9005 control plane server

Single socket AMD EPYC 9005 server with a 32 core class processor, 256 GB memory and mirrored NVMe, for Kubernetes control planes and cluster management.

Illustration of a 1U rackmount server
CPU
1 x AMD EPYC 9005 series, 32 core class, high clock
Memory
12 x DDR5 RDIMM, 256 GB standard, up to 1.5 TB
Storage
2 x M.2 boot, 4 x U.2 or E3.S NVMe, mirrored
Networking
2 x 25 GbE onboard, 1 x 100 GbE or 200 Gb/s adapter
Role
Control plane, scheduler, fabric manager, monitoring, provisioning
Power
Redundant PSUs, about 0.5 kW per node planning figure

Overview

Every cluster needs servers that are not doing the work but keeping it running: the Kubernetes or Slurm control plane, the InfiniBand subnet manager or fabric controller, monitoring, logging, provisioning and identity. This single socket node is specified for exactly those roles, with a high clock 32 core class AMD EPYC 9005, 256 GB of memory and mirrored NVMe for state.

Our cluster designs use six of them: three for the scheduler or Kubernetes quorum and three for fabric management, observability and provisioning, with services spread so that any one node can fail. They are quoted as a set with every cluster of eight nodes or more, and are available individually for expansions and replacements.

The configuration is deliberately conservative, with redundant power, mirrored boot and data drives, dual 25 GbE plus a 100 GbE or 200 Gb/s port, and a planning figure of about 0.5 kW. The same class works well as a jump host, a license server or a small virtualization node.

Use cases

  • Kubernetes and Slurm control plane
  • InfiniBand subnet manager and fabric control
  • Monitoring, logging and provisioning
  • Identity and license services

Specifications

Compute
CPU1 x AMD EPYC 9005 series, 32 core class, 16 to 64 cores available
System memory12 x DDR5 RDIMM, 256 GB standard, up to 1.5 TB
Sizing rule6 control plane nodes per cluster: 3 for the Kubernetes or scheduler quorum, 3 for fabric management, monitoring and provisioning
Storage
Boot2 x M.2 NVMe, mirrored
Data4 x U.2 or E3.S NVMe, mirrored pairs for etcd, databases and logs
RAIDSoftware RAID
Networking
Frontend2 x 25 GbE onboard
Fabric or storage1 x 100 GbE or 200 Gb/s ConnectX-7
ManagementDedicated 1 GbE BMC port, TPM 2.0
Physical
Form factor1U rackmount, air cooled, short depth options
Power suppliesRedundant 800 W class PSUs, 100 to 240 V input
Planning powerAbout 0.5 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

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.

Compare it with the next class up or down

The nearest classes we quote are the 1U dual socket EPYC 9005 server and the 64 node HGX B300 AI cluster, 512 GPUs. If neither is quite right, tell us the workload and we specify the system between them.

Talk to an engineerRequest a quotation