NVIDIA RTX PRO Server – a rack GPU server for AI and graphics
A universal platform for enterprise AI and industrial workloads: up to 8× RTX PRO Server Edition cards, vGPU/MIG sharing for multiple users. Delivery and commissioning in Serbia.
-
Up to 8× RTX PRO 6000 / 4500 Server Edition
-
vGPU / MIG – multiple users on one server
-
Rack audit, integration and commissioning at your request
What the RTX PRO Server is
The GPU power of the RTX PRO line – in a server edition
The RTX PRO Server is a server with NVIDIA RTX PRO Blackwell Server Edition cards – a platform on the Blackwell architecture with enterprise drivers and vGPU/MIG resource sharing.
It is not a desktop card in a box: the Server Edition is passive, built for dense servers, and comes through validated reference designs (NVIDIA MGX) and systems from manufacturers such as Supermicro.
One server can carry up to 8 cards – that is tens of thousands of CUDA cores and up to 768 GB of GPU memory in a single rack device, available to multiple users at once.
Users see it as a shared GPU resource: AI inference for applications, vGPU workstations for designers, rendering jobs and analytics – all from the same cabinet, with the data staying at your location.
We deliver it as a complete Supermicro RTX PRO BSE series system, with rack integration, testing and commissioning – and we stay with the equipment through warranty service.
Differences
Server, Desktop, Station or Spark?
Four tiers of NVIDIA infrastructure – chosen by the number of users, not by the wish for the strongest box.
| RTX PRO Server | RTX PRO Desktop | DGX Station | DGX Spark | |
|---|---|---|---|---|
| Where it lives | Rack, 1U–6U | Desk | Tower, ~1600 W | Desk, ~240 W |
| GPU | RTX PRO 6000 / 4500 Server Edition | RTX PRO Workstation | GB300, ~748 GB coherent | GB10, 128 GB unified |
| GPU count | 1–8 in one server | 1–2 | 1 | 1 (or 2 Sparks via cable) |
| Sharing | vGPU / MIG – multiple users | A single user | MIG on one machine | Local |
| OS | Linux server + NVIDIA AI Enterprise / vGPU | Windows / Linux | Ubuntu / a Windows variant | DGX OS |
GPU Server Edition
Two cards on offer – both passive, built for servers
The RTX PRO Server Edition is not a desktop card with a different sticker: different cooling and profile, and it goes through validated server platforms.
RTX PRO 6000 Blackwell Server Edition
| GPU memory | 96 GB GDDR7 with ECC |
|---|---|
| Bandwidth | ~1597 GB/s |
| Format | double-width FHFL, passive cooling in the server |
| Power draw | 400–600 W |
| MIG | up to 4 instances per card |
| FP32 | ~120 TFLOPS (system manufacturer spec) |
Dense deployment for AI inference, vGPU workstations in the rack and rendering – up to 8 cards in one 4U/6U server.
RTX PRO 4500 Blackwell Server Edition
| GPU memory | 32 GB GDDR7 |
|---|---|
| Bandwidth | ~800 GB/s |
| Format | single-width FHFL, passive, ~165 W |
| Power draw | ~165 W |
| MIG | up to 2 instances per card |
| FP32 | ~51 TFLOPS (system manufacturer spec) |
More cards in a narrower 2U format and edge systems – cheaper GPU density per user.
Reference design
Configurations on the NVIDIA MGX platform
The official reference configurations of the NVIDIA RTX PRO Server – three formats, from a 2U air-cooled to a 6U liquid-cooled system. The list does not guarantee every SKU is in stock – the concrete system is confirmed in the quote.
NVIDIA MGX 2U
NVIDIA MGX 4U
NVIDIA MGX 6U
Supermicro RTX PRO BSE
Supermicro systems with the NVIDIA RTX PRO Blackwell Server Edition
Ready-made, NVIDIA-compatible systems – from a 1U short-depth edge to 4U with 8 GPUs. Three representatives – other models on request.
Supermicro SYS-222H-TN
- up to 4× RTX PRO 4500 Blackwell Server Edition
- 2× Intel Xeon 6700/6500 (P-core), up to 350 W
- up to 8 TB DDR5, 8× 2.5-inch NVMe/SAS/SATA
- 1+1 redundant power, 1200–2600 W Titanium
- for vGPU workstations and inference in the cabinet
Supermicro SYS-422GA-NRT
- up to 8× RTX PRO 6000 Blackwell Server Edition
- 2× Intel Xeon 6900P, up to 128 cores per processor
- up to 6 TB DDR5 (24 DIMM), EDSFF E3.S NVMe
- 4× 3200 W Titanium, redundant (3+1)
- for AI factories, inference farms and rendering
Supermicro SYS-112B-FWT
- 1× RTX PRO 4500 Blackwell Server Edition
- Intel Xeon 6700 (E-core), single socket
- up to 1 TB DDR5, NVMe for local models
- a short 1U chassis for shallow cabinets
- for retail, manufacturing and 5G/edge sites
Workloads
What the RTX PRO Server is most often used for
One server, several scenarios – a range from AI inference to render farms, with a clear boundary between what is CPU work and what is GPU work.
Agentic / generative AI
Local inference for assistants and agents, multimodal inputs and documents – data stays with you.
Typical: 2× 6000 SE for a team, 8× 4500 SE when there are more users.
Industrial / physical AI
3D review, real-time simulation and robotics workflows connected to production data.
Typical: 1–2× 6000 SE per project, with headroom for peaks.
Simulation and analytics
CFD/FEA, per-scenario simulations and analytics over large datasets without waiting in cloud queues.
Typical: a GPU per task – vGPU sharing for occasional users.
AI rendering and graphics
Render farms and virtual workstations: vGPU sessions from the rack instead of machines under desks.
Typical: dense 4500 SE deployment – more users per rack unit.
FAQ
Frequently asked questions
Request a specification and a power check
Describe the workload and the number of users – we propose the GPU class and server format, and check what your rack and installation can realistically take before the order.
NVIDIA
More from the NVIDIA range
NVIDIA DGX Spark
If one user at a desk is enough: GB10, 128 GB unified memory, a regular outlet.
NVIDIA DGX Station
One large deskside tower for the team: GB300 Ultra, up to 748 GB of coherent memory.
RTX PRO Desktops
CAD and 3D at the engineer’s desk: professional Dell, HP and Lenovo workstations with an RTX PRO GPU.