4 GPU AI workstation, RTX PRO 6000 Blackwell Max-Q

Tower workstation with four RTX PRO 6000 Blackwell Max-Q GPUs, 384 GB of GPU memory at 300 W each, Threadripper PRO and up to 2 TB DDR5, on the desk.

Illustration of a tower workstation
GPUs
4 x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition, 96 GB each, 384 GB total
CPU
AMD Ryzen Threadripper PRO, 32 to 96 cores
Memory
8 channel DDR5 RDIMM, 256 GB to 2 TB
Storage
2 TB NVMe boot, up to 4 x M.2 and 4 x U.2 NVMe
Networking
10 GbE onboard, optional 100 GbE or 200 Gb/s adapter
Power
2,000 W class PSU, about 1.8 kW under full load, 200 to 240 V recommended

Overview

Four RTX PRO 6000 Blackwell Max-Q cards put 384 GB of GPU memory and 16 PFLOPS of FP4 in a tower that sits under a desk. The Max-Q edition runs at 300 W with an active blower cooler, which is what makes four cards possible in one chassis without liquid cooling, while keeping most of the performance of the 600 W edition.

This is the workstation for teams that fine tune mid sized models, run several models at once, serve a local inference endpoint for a department, or render and simulate at scale without booking cluster time. AMD Ryzen Threadripper PRO supplies 128 PCIe Gen5 lanes so each GPU has a full x16 link, and eight channels of DDR5 up to 2 TB feed them.

A 200 to 240 V supply is recommended for full performance; on 120 V the GPU power limits are reduced. We burn in all four GPUs under sustained load, record thermals and deliver the system imaged with drivers, CUDA and your container runtime.

Use cases

  • Fine tuning mid sized models
  • Departmental inference endpoint
  • Multi model experimentation
  • Rendering and simulation
  • Digital twins and visualization

Specifications

Compute
GPUs4 x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition, 300 W, active blower, dual slot
GPU memory96 GB GDDR7 with ECC per GPU, 384 GB total
GPU interconnectPCIe Gen5 x16 per GPU, peer to peer over the CPU root complex, no NVLink
CPUAMD Ryzen Threadripper PRO, 32 to 96 cores, 128 PCIe Gen5 lanes
System memory8 x DDR5 RDIMM, 256 GB to 2 TB, ECC
Storage
Boot2 TB M.2 NVMe PCIe Gen5
DataUp to 4 x M.2 and 4 x U.2 NVMe
RAIDSoftware RAID for data volumes
Networking
Onboard10 GbE and 1 GbE
Optional1 x 100 GbE or 200 Gb/s ConnectX-7
Remote managementRemote desktop and SSH access, optional BMC on rack conversion
Physical
Form factorTower, 5U equivalent, rack conversion kit optional
Power supply2,000 W class, 200 to 240 V recommended, 120 V operation with reduced GPU power limits
Planning powerAbout 1.8 kW under full load

Figures describe the system class. Exact configuration, weights and power draw are confirmed on the quotation.

What's included

  • Power cord, keyboard and mouse
  • Burn in with all GPUs under sustained load and acceptance report
  • Operating system imaged with NVIDIA drivers and CUDA toolkit on request
  • 3 year warranty, options to 5 years with advance replacement

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 Liquid cooled 4 GPU AI workstation and the Rackmount 4U workstation, up to 4 GPUs. If neither is quite right, tell us the workload and we specify the system between them.