(858) 613-1818Quote list 0

4 GPU PCIe AI server, 2U

Four PCIe GPUs, RTX PRO 6000 Blackwell Server Edition, H200 NVL or L40S, in a 2U server with dual EPYC 9005 or Xeon 6 for dense inference racks.

GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition, H200 NVL or L40S

Illustration of a 2U rackmount server
GPUs
4 x RTX PRO 6000 Blackwell Server Edition, H200 NVL or L40S, double width PCIe Gen5
CPU
2 x AMD EPYC 9005 or 2 x Intel Xeon 6
Memory
Up to 1.5 TB DDR5 RDIMM
Storage
2 x M.2 boot, up to 4 x U.2 or E3.S NVMe
Networking
2 x 400 Gb/s ConnectX-7 (one per GPU pair), 2 x 25 GbE
Power
Redundant PSUs, about 3.5 kW per node planning figure

Overview

The 2U four GPU server is the densest way to deploy PCIe GPUs in a standard rack, with twenty GPUs in ten rack units. It accepts four double width cards, RTX PRO 6000 Blackwell Server Edition for Blackwell FP4 throughput, H200 NVL for HBM capacity with NVLink bridges, or L40S for a lower power universal GPU, behind dual AMD EPYC 9005 or Intel Xeon 6 processors.

It is our most requested inference node. A single RTX PRO 6000 card serves 7B to 32B parameter models with room for KV cache, a bridged pair of H200 NVL serves 70B class models, and MIG partitioning lets one node host many small endpoints. Two 400 Gb/s adapters give the node GPUDirect RDMA into a cluster fabric when it is part of a larger deployment.

The chassis is validated for 600 W GPUs with front to back airflow, so it needs a hall with adequate cold aisle supply but no liquid cooling. Memory is DDR5 to 1.5 TB and storage is mirrored M.2 boot plus up to four NVMe data drives. Plan for about 3.5 kW per node with Blackwell cards.

Use cases

  • LLM and embedding inference
  • Dense inference racks
  • Fine tuning of small and mid sized models
  • Video analytics and transcoding
  • Virtual desktops and workstations

Specifications

Compute
GPUs4 x NVIDIA RTX PRO 6000 Blackwell Server Edition (96 GB), H200 NVL (141 GB) or L40S (48 GB)
GPU interconnectPCIe Gen5 x16 per GPU; NVLink bridges available with H200 NVL
CPUs2 x AMD EPYC 9005 series or 2 x Intel Xeon 6
System memory24 x DDR5 RDIMM, up to 1.5 TB
Storage
Boot2 x M.2 NVMe, mirrored
DataUp to 4 x U.2 or E3.S NVMe, 7.68 TB to 15.36 TB each
RAIDSoftware RAID or optional hardware RAID
Networking
GPU fabric2 x 400 Gb/s ConnectX-7 or 2 x 200 Gb/s, InfiniBand or RoCE
Frontend2 x 25 GbE onboard, optional 2 x 100 GbE
ManagementDedicated 1 GbE BMC port, TPM 2.0
Physical
Form factor2U rackmount, air cooled, GPUs in the rear bay
Power suppliesRedundant 2,000 W class PSUs, 200 to 240 V input
Planning powerAbout 3.5 kW per node with 600 W GPUs, about 2.5 kW with L40S

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
  • NVIDIA end user registration handled with the order

Compliance

Compare it with the next class up or down

The nearest classes we quote are the 4 GPU PCIe AI server, 4U and the 2 GPU 2U inference server. If neither is quite right, tell us the workload and we specify the system between them.

Talk to an engineerRequest a quotation