(858) 613-1818Quote list 0

8 GPU PCIe AI server with RTX PRO 6000 Blackwell Server Edition

Eight RTX PRO 6000 Blackwell Server Edition GPUs, 96 GB each, in a 4U PCIe server with dual EPYC 9005 or Xeon 6 and 400 Gb/s networking.

GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition

Illustration of a 4U GPU server
GPUs
8 x RTX PRO 6000 Blackwell Server Edition, 96 GB GDDR7 ECC each, 768 GB total
CPU
2 x AMD EPYC 9005 or 2 x Intel Xeon 6
Memory
Up to 2 TB DDR5 RDIMM
Storage
2 x M.2 boot, up to 8 x U.2 or E3.S NVMe
Networking
4 x 400 Gb/s ConnectX-7 (one per GPU pair), 2 x 25 GbE
Power
Redundant PSUs, about 6.5 kW per node planning figure

Overview

The eight GPU PCIe server is the workhorse for inference fleets, fine tuning and mixed AI and graphics workloads where an HGX baseboard is more than the job needs. RTX PRO 6000 Blackwell Server Edition brings 96 GB of GDDR7 with ECC and 4 PFLOPS of FP4 per card, so a single node offers 768 GB of GPU memory and 32 PFLOPS of FP4 at less than half the power of an HGX node.

GPUs sit behind PCIe Gen5 switches with a 400 Gb/s ConnectX-7 adapter per GPU pair, which gives GPUDirect RDMA into the cluster fabric and keeps tensor parallel traffic on the PCIe switch. Models up to roughly 70B parameters run comfortably on a bridged pair or a quad, and the node serves many smaller models at once with MIG partitioning.

We quote it with dual AMD EPYC 9005 or Intel Xeon 6 processors, DDR5 memory to 2 TB and up to eight NVMe drives. At about 6.5 kW per node it runs in a conventional air cooled rack, and the four or five rack unit chassis fits eight to ten nodes per rack. Burn in covers all eight GPUs under sustained load with thermal logging in the acceptance report.

Use cases

  • LLM inference serving
  • Fine tuning up to 70B parameters
  • Multi tenant GPU sharing with MIG
  • Rendering and media pipelines
  • Virtual workstations

Specifications

Compute
GPUs8 x NVIDIA RTX PRO 6000 Blackwell Server Edition, dual slot passive, PCIe Gen5 x16
GPU memory96 GB GDDR7 with ECC per GPU, 1.6 TB/s each, 768 GB per node
GPU interconnectPCIe Gen5 through PCIe switches, GPU direct RDMA to the adapters, no NVLink
FP4 performance4 PFLOPS per GPU, 32 PFLOPS per node
CPUs2 x AMD EPYC 9005 series or 2 x Intel Xeon 6
System memory24 or 32 x DDR5 RDIMM, up to 2 TB
Storage
Boot2 x M.2 NVMe, mirrored
DataUp to 8 x U.2 or E3.S NVMe, 7.68 TB to 30.72 TB each
RAIDSoftware RAID or optional hardware RAID
Networking
GPU fabric4 x 400 Gb/s ConnectX-7, InfiniBand or RoCE, one adapter per GPU pair; 8 x 400 Gb/s optional
Frontend2 x 25 GbE onboard, optional 2 x 100 GbE
ManagementDedicated 1 GbE BMC port, TPM 2.0
Physical
Form factor4U rackmount, air cooled; 5U variants with extra drive bays
Power suppliesRedundant 2,700 W class PSUs, 200 to 240 V input
Planning powerAbout 6.5 kW per node with GPUs at 600 W
GPU powerConfigurable 400 W to 600 W per GPU

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 with multi hour GPU stress and NVLink validation, acceptance report
  • NVIDIA end user registration handled with the order
  • Operating system imaging, driver and fabric manager baseline on request
  • 3 year warranty, options to 5 years with advance replacement

Compliance

Compare it with the next class up or down

The nearest classes we quote are the 10 GPU PCIe AI server, 5U and the 4 GPU PCIe AI server, 2U. If neither is quite right, tell us the workload and we specify the system between them.

Talk to an engineerRequest a quotation