NVIDIA RTX PRO 6000 Blackwell Server Edition
A 96 GB passive server GPU for rackmount inference and fine-tuning. It has no fans of its own and needs a validated server chassis.
Price: On request
Availability: Confirmed at quotation
Cooling: Passive, server chassis only
Manufacturer: NVIDIA
Specifications
Manufacturer figures, checked against the sources below.
| Specification | Manufacturer figure |
|---|---|
| Architecture | NVIDIA Blackwell |
| GPU memory | 96 GB GDDR7 with ECC |
| Memory interface | 512-bit |
| Memory bandwidth | 1597 GB/s |
| CUDA cores | 24,064 |
| Tensor cores | 752 (fifth generation) |
| RT cores | 188 (fourth generation) |
| Host interface | PCIe 5.0 x16 |
| Board power | Up to 600 W, configurable |
| Power connector | 1× PCIe CEM5 16-pin |
| Cooling | Passive (air-cooled); a liquid-cooled version is also listed |
| Form factor | Air-cooled: dual slot, full height, full length, 4.4 in H × 10.5 in L |
| Display outputs | 4× DisplayPort 2.1, off by default (switchable) |
| Multi-Instance GPU | Up to 4 instances at 24 GB |
| Virtualization | SR-IOV (48 virtual functions); NVIDIA vGPU 19.0 or later |
| NVLink | Not supported |
| Peak AI performance (manufacturer) | 4 PFLOPS FP4 Tensor Core (sparsity not stated) |
Is it right for your workload?
A good fit for
- Multi-GPU rack servers serving models to many users.
- Data-centre or server-room deployments with remote management.
- Shared inference and fine-tuning for a department or enterprise.
- Private RAG
- Local LLM
- Fine-tuning
Consider something else if
- Desktop towers and workstations: without high-pressure chassis airflow it overheats. Choose the Workstation Edition.
- Offices without a server room: the servers that cool it are loud.
Cooling, power and installation
- Passive: no fans on the card. Cooling comes entirely from the server chassis fans and ducting.
- We only install it in a chassis validated for this GPU, never in an ordinary desktop case.
- NVLink is not supported, so multi-GPU work runs over PCIe; we plan the topology for your workload.
- Rack power, redundant supplies and room cooling are agreed with your facilities team before quoting.
What is included
- Supplied as part of a validated rack server, with the chassis, power and airflow designed around it.
- Server software stack and monitoring installed and tested during the build.
Last reviewed 27 Oct 2026. Warranty terms are confirmed per quotation.
Questions about this GPU
Why can't it go in a normal PC?
It has no fans. It relies on a server chassis pushing high-pressure air through its heatsink; a desktop case cannot do that, so the card overheats and throttles.
How many fit in one server?
It depends on the chassis. Common designs take two to eight. We size the chassis, power and cooling together.
When can I get one?
Availability changes; we confirm it in the quotation rather than show a stock level we have not verified.
Plan a system around it
We check compatibility, power and cooling for your workload before anything is quoted.