GPU GUIDE · UPDATED 11 NOVEMBER 2026
RTX PRO 6000 Blackwell: price, editions and specifications
The NVIDIA RTX PRO 6000 Blackwell is a professional GPU with 96GB of GDDR7 memory and fifth-generation Tensor Cores with FP4 support. It comes in three editions — Server, Workstation and Max-Q — that share the same memory but differ in cooling, power and where they belong. This page compares the editions, lists the specifications, shows our indicative EU prices and explains when an RTX 5090 or an H200 NVL is the better choice.
At a glance
96GB GDDR7 with ECC
Enough for a 70B-parameter model in 8-bit on a single GPU, with room left for context.
Blackwell with FP4
Fifth-generation Tensor Cores add hardware FP4, which halves the memory per parameter compared with FP8.
Three editions
Server Edition for rack servers, Workstation Edition for desktops, Max-Q for multi-GPU workstations at 300W.
From $11,900 at NodeX
Indicative price of the Server Edition in our configurator on 11 November 2026, confirmed in your written quote.
The three editions compared
Figures from NVIDIA’s published specifications for each edition. NVIDIA states that the Workstation and Max-Q AI TOPS figures are theoretical FP4 with sparsity; it does not say whether the Server Edition’s 4 PFLOPS FP4 figure assumes sparsity, so treat the FP4 row as approximate.
| Server Edition | Workstation Edition | Max-Q Workstation Edition | |
|---|---|---|---|
| GPU memory | 96GB GDDR7 with ECC | 96GB GDDR7 with ECC | 96GB GDDR7 with ECC |
| Memory bandwidth | 1,597 GB/s | 1,792 GB/s | 1,792 GB/s |
| FP4 AI performance | 4 PFLOPS | 4,000 AI TOPS (with sparsity) | 3,511 AI TOPS (with sparsity) |
| FP32 | 120 TFLOPS | 125 TFLOPS | 110 TFLOPS |
| RT Core performance | 355 TFLOPS | 380 TFLOPS | 333 TFLOPS |
| Maximum power | Up to 600W, configurable | 600W | 300W |
| Cooling | Passive (air) or liquid-cooled version | Double flow-through fans | Active (fan) |
| Size | Dual-slot, full height, full length (air) | Dual-slot, 5.4″ × 12.0″ | Dual-slot, 4.4″ × 10.5″ |
| Built for | Rack servers with front-to-back airflow | Desktop workstations | Workstations with several GPUs and limited power |
| Indicative NodeX price | $11,900 | $12,400 | On request |
Prices are NodeX indicative prices in US dollars as shown in our configurator on 11 November 2026. VAT, delivery and availability are confirmed in your written quotation.
Which edition do you need?
- Server Edition — for rack servers. It has no fans of its own and relies on the server’s front-to-back airflow, which is how our 4U GPU servers are built. Choose it for 2 to 8 GPUs in one chassis.
- Workstation Edition — for a desktop workstation with one or two GPUs. Its fans push air through the card, which suits an open workstation case rather than a dense server.
- Max-Q Workstation Edition — for workstations that need several GPUs on a limited power budget. At 300W it gives up some performance for half the power.
Whichever edition you choose, we confirm compatibility with the chassis, power supplies and cooling in your quotation.
RTX PRO 6000 Server Edition specs
| Specification | RTX PRO 6000 Blackwell Server Edition |
|---|---|
| Architecture | NVIDIA Blackwell |
| CUDA cores | 24,064 |
| Tensor Cores | 5th generation |
| RT Cores | 188 (4th generation) |
| GPU memory | 96GB GDDR7 with ECC, 512-bit |
| Memory bandwidth | 1,597 GB/s |
| FP4 Tensor Core | 4 PFLOPS |
| FP8 Tensor Core | 2 PFLOPS |
| FP16 / BF16 Tensor Core | 1 PFLOPS |
| TF32 Tensor Core | 234 TFLOPS |
| FP32 | 120 TFLOPS |
| Multi-Instance GPU | Up to 4 isolated instances |
| Video engines | 4 encoders (9th gen NVENC), 4 decoders (6th gen NVDEC) |
| Interface | PCI Express Gen 5 |
| Maximum power | Up to 600W, configurable |
The RTX PRO 6000 has no NVLink: several cards in one server communicate over PCIe. This works well for inference and for running several models side by side; models split across many GPUs gain more from NVLink, which the H200 NVL offers.
RTX PRO 6000 price in the EU
NodeX sells the RTX PRO 6000 as part of a complete GPU server, built and tested in Bulgaria. Indicative GPU prices at NodeX on 11 November 2026:
| Configuration | Total GPU memory | GPUs at indicative prices |
|---|---|---|
| 1× RTX PRO 6000 Server Edition | 96GB | $11,900 (about $124 per GB) |
| 1× RTX PRO 6000 Workstation Edition | 96GB | $12,400 (about $129 per GB) |
| 2× Server Edition | 192GB | $23,800 |
| 4× Server Edition | 384GB | $47,600 |
| 8× Server Edition | 768GB | $95,200 |
These totals cover the GPUs only. The platform, processors, memory, storage, networking, assembly and warranty are added in your quote. For a complete, priced eight-GPU build, see our 8× RTX PRO 6000 reference configuration.
RTX PRO 6000 vs RTX 5090
Both are Blackwell GPUs with GDDR7 on a 512-bit interface. The difference that matters most for AI is memory: 96GB against 32GB. At our indicative prices the RTX 5090 server version costs $4,690 and the RTX PRO 6000 Server Edition $11,900 — about $147 against $124 per GB of GPU memory.
- Choose the RTX 5090 when your models fit comfortably in 32GB per GPU — for example 8B to 14B models, or image generation — and the budget per GPU matters most.
- Choose the RTX PRO 6000 when you need memory. One card holds what would otherwise be split across three RTX 5090s.
Specifications, cost at equal memory, power and use cases are in our full RTX PRO 6000 vs RTX 5090 comparison.
What 96GB of GPU memory is good for
LLM inference
A 70B model in 8-bit needs about 70GB for the weights, so it fits on one card with room for context. A 32B model fits even in 16-bit.
Larger models across GPUs
Two to four cards give 192–384GB for 70B models in 16-bit or 120B-class models in 8-bit; eight cards give 768GB.
Fine-tuning
LoRA and medium-scale fine-tuning benefit from 96GB per GPU, which reduces the need to split work across cards.
Rendering and video
RT Cores and four encoders and decoders per card suit rendering, simulation and video pipelines.
RTX PRO 6000 servers from NodeX

We build 4U GPU servers in Bulgaria with one to eight RTX PRO 6000 Server Edition GPUs — for example on our NX-G8 Dual SP5 platform with AMD EPYC 9004-series processors, DDR5 and PCIe 5.0. Your written quotation confirms compatibility, power and cooling, delivery and warranty. We can also arrange colocation in Bulgaria and operating system installation.
Frequently asked questions
Is the RTX PRO 6000 the same as the RTX 6000 Ada?
No. The RTX 6000 Ada is the previous generation, with 48GB of GDDR6 and no FP4 support. The RTX PRO 6000 Blackwell has 96GB of GDDR7. Both are in our configurator: the RTX 6000 Ada at $6,726 and the RTX PRO 6000 Server Edition at $11,900 (indicative).
Can I put the Workstation Edition in a rack server?
The Workstation Edition is designed for the airflow of a desktop workstation, while the Server Edition is designed to be cooled by a server’s fans. For multi-GPU rack servers we recommend the Server Edition and confirm the final choice in your quotation.
Does the RTX PRO 6000 support NVLink?
No. Cards communicate over PCI Express Gen 5. If you need NVLink between GPUs, compare it with the H200 NVL.
Do you sell the Max-Q edition?
It is not in our configurator at the moment. Ask us in your request and we will check availability and price.
Sources
Specifications and platform details on this page come from the published sources below. Prices come from our configurator on the date shown at the top of the page.
