(858) 613-1818Quote list 0

2 GPU 1U inference server

Two L40S or RTX PRO 4500 Blackwell GPUs in a 1U server with single or dual socket host, 200 Gb/s networking and low power, for edge inference.

GPU: NVIDIA L40S or RTX PRO 4500 Blackwell

Illustration of a 1U rackmount server
GPUs
2 x NVIDIA L40S (48 GB) or 2 x RTX PRO 4500 Blackwell (32 GB), double width PCIe
CPU
1 or 2 x AMD EPYC 9005, or 2 x Intel Xeon 6
Memory
Up to 768 GB DDR5 RDIMM
Storage
2 x M.2 boot, up to 4 x U.2 or E3.S NVMe
Networking
1 x 200 Gb/s ConnectX-7 or 2 x 100 GbE, 2 x 25 GbE onboard
Power
Redundant PSUs, about 1.2 kW per node planning figure

Overview

The 1U two GPU server is the compact inference node for racks where power and space are tight, and for edge sites that need GPU acceleration without a full AI rack. It fits two double width cards: L40S for the widest software validation and video engines, or RTX PRO 4500 Blackwell for Blackwell FP4 support at only 200 W per card.

At about 1.2 kW per node it runs from a standard 120 V or 230 V circuit, which makes it the usual answer for branch data centers, retail and industrial sites, and departmental racks. Typical work is serving small and mid sized language models, embedding and reranking, computer vision pipelines and video transcoding, with 200 Gb/s or 100 GbE connectivity back to the core.

The host is a single or dual socket AMD EPYC 9005 or dual Intel Xeon 6 with DDR5 to 768 GB, mirrored boot and up to four hot swap NVMe drives. Short depth variants are available for telecom and edge racks. Every unit is burned in with both GPUs under load and delivered with the acceptance report.

Use cases

  • Edge inference
  • Small model serving and embeddings
  • Computer vision pipelines
  • Video transcoding and analytics
  • Branch and departmental AI

Specifications

Compute
GPUs2 x NVIDIA L40S (48 GB GDDR6, 350 W) or 2 x RTX PRO 4500 Blackwell (32 GB GDDR7, 200 W)
GPU interconnectPCIe Gen5 x16 per GPU, no NVLink
CPUs1 or 2 x AMD EPYC 9005 series, or 2 x Intel Xeon 6
System memory12 or 24 x DDR5 RDIMM, up to 768 GB
Storage
Boot2 x M.2 NVMe, mirrored
DataUp to 4 x U.2 or E3.S NVMe, hot swap
RAIDSoftware RAID
Networking
Fabric1 x 200 Gb/s ConnectX-7 or 2 x 100 GbE, PCIe Gen5
Frontend2 x 25 GbE onboard
ManagementDedicated 1 GbE BMC port, TPM 2.0
Physical
Form factor1U rackmount, air cooled, short depth options for edge racks
Power suppliesRedundant 1,200 W class PSUs, 100 to 240 V input
Planning powerAbout 1.2 kW per node with L40S, about 0.9 kW with RTX PRO 4500

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 for L40S

Compliance

Compare it with the next class up or down

The nearest classes we quote are the Single GPU 1U edge inference server 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