Single GPU 1U edge inference server

Short depth 1U server with one L40S or RTX PRO 4500 Blackwell GPU, single socket host and dual 25 GbE, for edge inference, vision and video sites.

Illustration of a 1U rackmount server
GPU
1 x NVIDIA L40S (48 GB) or 1 x RTX PRO 4500 Blackwell (32 GB), double width PCIe
CPU
1 x AMD EPYC 9005 or 1 x Intel Xeon 6, 16 to 64 cores
Memory
Up to 384 GB DDR5 RDIMM
Storage
2 x M.2 boot, up to 2 x U.2 NVMe
Networking
2 x 25 GbE onboard, optional 100 GbE or 200 Gb/s adapter
Power
Redundant PSUs, about 0.8 kW per node planning figure

Overview

The single GPU edge server brings one data center class GPU to sites that are not data centers: factory floors, retail back rooms, hospitals, telecom cabinets and remote offices. The short depth 1U chassis, single socket host and low power draw let it run in a shallow wall cabinet on an ordinary circuit, with front access options for serviceability.

L40S gives it a widely validated inference and video platform with three encode and three decode engines, while RTX PRO 4500 Blackwell adds FP4 support and keeps the whole node under one kilowatt. Typical deployments run computer vision models on camera streams, local language models for assistants and document processing, and transcoding for media distribution.

The host is a single AMD EPYC 9005 or Intel Xeon 6 with DDR5 to 384 GB, mirrored boot and two hot swap NVMe drives. Dual 25 GbE is standard and a 100 GbE or 200 Gb/s adapter is optional. Units are burned in and can be shipped pre imaged with your container runtime so they are ready at power on.

Use cases

  • Edge inference and vision
  • Local language model assistants
  • Media transcoding at the edge
  • Industrial and retail AI
  • Telecom and 5G sites

Specifications

Compute
GPU1 x NVIDIA L40S (350 W) or 1 x RTX PRO 4500 Blackwell (200 W), PCIe Gen5 x16
CPU1 x AMD EPYC 9005 series or 1 x Intel Xeon 6, 16 to 64 cores
System memory12 x DDR5 RDIMM, up to 384 GB
Expansion1 additional low profile PCIe slot
Storage
Boot2 x M.2 NVMe, mirrored
DataUp to 2 x U.2 NVMe, hot swap
RAIDSoftware RAID
Networking
Frontend2 x 25 GbE onboard
Optional1 x 100 GbE or 200 Gb/s ConnectX-7
ManagementDedicated 1 GbE BMC port, TPM 2.0
Physical
Form factor1U rackmount, short depth class under 600 mm, front I/O options
Power suppliesRedundant 800 W class PSUs, 100 to 240 V input
Planning powerAbout 0.8 kW per node
EnvironmentExtended inlet temperature options for edge cabinets

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 2 GPU 1U inference server and the Compact developer system, RTX PRO 4500. If neither is quite right, tell us the workload and we specify the system between them.