(858) 613-1818Quote list 0

2U 24 bay all NVMe storage server

All flash storage node with 24 U.2 or E3.S NVMe bays, up to 737 TB raw, dual EPYC 9005 or Xeon 6 and 2 x 400 Gb/s ports, for datasets and checkpoints.

Illustration of a 2U rackmount server
Drives
24 x U.2 or E3.S NVMe, 7.68 TB to 30.72 TB each, up to 737 TB raw
CPU
2 x AMD EPYC 9005 or 2 x Intel Xeon 6
Memory
24 x DDR5 RDIMM, 512 GB to 1.5 TB
Networking
2 x 400 Gb/s ConnectX-7 with RDMA, 2 x 25 GbE
Software
Parallel file systems, NVMe over Fabrics, object storage, software defined storage
Power
Redundant PSUs, about 1.3 kW per node planning figure

Overview

This is the all flash node that feeds GPU clusters. Twenty four NVMe bays hold up to 737 TB raw behind dual AMD EPYC 9005 or Intel Xeon 6 processors, and two 400 Gb/s ConnectX-7 adapters with RDMA let the node deliver data straight into GPU memory with GPUDirect Storage. It is the usual hot tier for datasets, checkpoints and model repositories.

Our cluster sizing rule is one terabyte of usable storage per GPU, so a 64 node cluster with 512 GPUs gets about 512 TB usable, typically from three or four of these nodes under a parallel file system with erasure coding. We quote the node with your choice of software, whether a parallel file system, NVMe over Fabrics, an object store or a software defined storage product.

Drives are enterprise NVMe in 7.68 TB to 30.72 TB capacities in U.2 or E3.S, with memory from 512 GB to 1.5 TB for caching and metadata. Plan for about 1.3 kW per node. Burn in exercises every drive bay and both storage ports and documents per drive health data.

Use cases

  • Training datasets and checkpoints
  • Parallel file systems for GPU clusters
  • NVMe over Fabrics targets
  • Object storage for AI
  • Database and analytics storage

Specifications

Storage
Drive bays24 x 2.5 inch U.2 or E3.S NVMe, front hot swap, PCIe Gen5
CapacityUp to 737 TB raw with 30.72 TB drives, 184 TB with 7.68 TB drives
Boot2 x M.2 NVMe, mirrored
Sizing rule1 TB usable storage per cluster GPU, so one node serves about 128 to 256 GPUs at the capacity tier
ThroughputTens of GB/s per node, limited by the 2 x 400 Gb/s ports
Compute
CPUs2 x AMD EPYC 9005 series or 2 x Intel Xeon 6
System memory24 x DDR5 RDIMM, 512 GB to 1.5 TB for caching and metadata
Expansion4 x PCIe Gen5 x16 slots
Networking
Storage network2 x 400 Gb/s ConnectX-7, InfiniBand or RoCE with GPUDirect Storage
Frontend2 x 25 GbE onboard
ManagementDedicated 1 GbE BMC port, TPM 2.0
Physical
Form factor2U rackmount, air cooled
Power suppliesRedundant 1,600 W class PSUs, 100 to 240 V input
Planning powerAbout 1.3 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
  • Drive health report with every drive serial

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 12 bay E3.S NVMe storage server and the 4U 36 bay HDD storage server. If neither is quite right, tell us the workload and we specify the system between them.

Talk to an engineerRequest a quotation