NVIDIA RTX PRO 6000 Blackwell Max-Q
96 GB of GDDR7 ECC memory for AI models, inference, rendering and professional compute workloads.
96 GBBARE-METAL GPU SERVER · B2B · EUROPE
Dedicated RTX PRO 6000 96GB bare-metal server in Europe
Monthly rental of a physical machine with an NVIDIA RTX PRO 6000 Blackwell Max-Q. One business, all resources, predictable cost.
999 EURnet / monthServer with 4 TB NVMe · no setup fee
Checking current availability…
NVIDIA RTX PRO 6000BLACKWELL MAX-Q / 96 GB ECC
96 GBGDDR7 ECC
2×10Gnetwork card
99.9%Standard SLA
L3 / L4DDoS filter
ONE PROFILE. THE FULL MACHINE.
Built for demanding GPU workloads.
No shared hypervisor and no hourly meter. You receive a physical server for the agreed term.
- Processor
- AMD EPYC · 16C / 32T · 4.3–5.7 GHz
- Memory
- 192 GB DDR5
- Storage
- 4 TB NVMe
- Network
- 2×10G · 10 Gbit/s included
- Addressing
- /30, /29 or /28 per server included
- Model
- Single tenant · administrative access
WHAT WILL YOU RUN?
One server. Your GPU project.
Choose a use case, explore our test results and configure your server.
No preference selected. Choose an example or go straight to the calculator.
LLM / chatbot
Your own assistant, under your control.
Connect a model to your application for draft summaries, documentation questions or team support. Check response quality against your own examples.
256 tokens in 1.5 s
Measurement conditions
Qwen3-8B Q4_K_M · 2048 input tokens · one request. Median complete local API response across 5 runs after warm-up, excluding Internet latency. This measures speed, not chatbot quality.
Test: RTX PRO 6000 Blackwell Max-Q and 96 GB of system RAM. Rental offer: the same GPU and 192 GB RAM. Your task's runtime depends on its data and settings; the article provides the full method and results.
AI video
Explore the idea before expanding production.
Animate a product illustration, try a direction of movement or create a short clip for further editing. Play an actual result from our test below.
A 5.1 s clip in 5.1 min
Measurement conditions
Wan2.2 · headphones · 832 × 480 · 20 steps. One measured run after a separate warm-up, not a median. Generation time excludes the environment's first startup.
Test: RTX PRO 6000 Blackwell Max-Q and 96 GB of system RAM. Rental offer: the same GPU and 192 GB RAM. Your task's runtime depends on its data and settings; the article provides the full method and results.
3D rendering
Give rendering its own machine.
Transfer your project with its dependencies, run the GPU calculations and collect the result. Your workstation can stay available for further project work.
Full HD classroom: 8.9 s
Measurement conditions
Blender Cycles / OptiX · 512 samples · median of 3 renders after warm-up. Timing covers the render stage with scene preparation, not program startup, project loading, saving or file transfer.
Test: RTX PRO 6000 Blackwell Max-Q and 96 GB of system RAM. Rental offer: the same GPU and 192 GB RAM. Your task's runtime depends on its data and settings; the article provides the full method and results.
Clean Linux
Start with the OS. Choose the rest yourself.
Debian 13 or Ubuntu 26.04 LTS, administrative access and your own workflow. Set up applications, containers and job queues to suit your project.
Debian 13 / Ubuntu 26.04 LTS
Choose the operating system in the order form. Install the software yourself or agree its preparation with us.
Configure your serverChoose how to use the same dedicated server. We will agree software preparation with you before activation.
RENTAL CALCULATOR
Configure the monthly service.
The slider range adjusts to the number of servers currently available to order.
A 10% discount applies to recurring fees for terms from 7 to 12 months.
KATOWICE · SLA · NETWORK PROTECTION
One clearly defined service SLA.
Our Standard Service SLA is 99.9% and covers the data-centre infrastructure and our network within the scope defined by the Terms and the Order.
Monthly availability
The Standard SLA of 99.9% is included with the service. It covers power and cooling that permit Server operation, and our network to the demarcation point, under the Terms and the Order.
Read about measurement and creditsInfrastructure in Katowice
4DC2 has two independent power paths, independent UPS systems, a standby generator, duplicated cooling and round-the-clock technical duty, 24/365. Guaranteed 4DC2 infrastructure availability is 99.982%; our service SLA remains clearly defined at 99.9%.
4DataCenterBandwidth per server
10 Gbit/s is included. Bonded 20 Gbit/s adds EUR 99 net for each server and month. A 40 Gbit/s connection is prepared on request after individual confirmation of the terms.
Compare optionsDDoS filtering included
Every active service is protected by network-level DDoS filtering by default. The protection layer continuously analyses traffic, detects anomalies and automatically suppresses volumetric and protocol-based L3/L4 attacks, including SYN, UDP, ICMP and amplification floods. Rules automatically adapt to the detected attack vector so legitimate traffic can continue to reach the server.
Read the protection scopeSERVER LIFECYCLE
Control from installation to a safe exit.
- 01
B2B verification
We confirm the company, hardware availability, VAT treatment and expected activation date.
- 02
Installation
Choose Debian 13, Ubuntu 26.04 LTS or agree an optional AI-ready profile.
- 03
Operation
Use the full machine for the prepaid term. The Order defines parameters, support and responsibilities.
- 04
Export or erasure
Before service ends, choose migration, data recovery or controlled erasure under the Terms.
FAQ
Before you send a request.
Is the server or GPU shared?
No. It is a physical server dedicated to one B2B customer. The Order confirms access scope and configuration.
How many servers can I order?
The page retrieves the current available server count and automatically limits the slider. Hardware and the activation date are finally confirmed during Order verification.
How does the term discount work?
For 7–12 months, 10% is deducted from all recurring fees. Server setup is free of installation and activation fees. The full term is paid upfront.
What does IPv4 addressing include?
Choose /30, /29 or /28 per server at no extra cost. Routing details and usable-address count are recorded in the Order.
What does DDoS protection cover?
Every active service is protected by a network filter that continuously analyses traffic and automatically suppresses volumetric and protocol-based L3/L4 attacks, including SYN, UDP, ICMP and amplification floods. Filtering rules adapt to the detected attack vector.
Do you back up my data?
The standard rental does not include Customer Data backups. Off-server backups remain the customer's responsibility unless the Order says otherwise.
TESTS & GUIDES
Explore what you can run.
First-party measurements, a transparent methodology and practical GPU server resources.
GPU server tests and guides
Our GPU server tests and practical guides: AI response time, model memory, Blender, Linux and costs. Start with your use case.
How we test AI models and explain the results
Our shared LLM test methodology: response time, memory, simultaneous requests and correct answers. Understand the rules and differences between measurements.
About Bestconnect
Meet GPU Server Hub by Bestconnect: dedicated GPU servers in Poland for Polish and EU businesses, monthly pricing and control of your environment.
GPU server infrastructure and security
A dedicated GPU server in Katowice: a whole machine for your company, networking, DDoS protection and clear customer and operator responsibilities.
RTX PRO 6000 Blackwell Max-Q vs Server Edition
Which RTX PRO 6000 do we offer? Max-Q 96 GB versus Server Edition: the differences, what they mean for customers and our AI and Blender tests.
Monthly GPU server or hourly GPU cloud?
A dedicated monthly GPU server or hourly billing? Practical examples, total costs and straightforward selection criteria for businesses.
AI and GPU servers: the terms explained
Plain-language terms for choosing an AI server: GPUs, LLMs, tokens, context, memory, APIs, MoE experts and offloading, with practical examples.
Blender on RTX PRO 6000 Blackwell Max-Q — measured render times
How fast does RTX PRO 6000 Blackwell Max-Q render? Our three-scene Blender test, measured times, practical uses and reproducible settings.
A large AI model on one GPU: Qwen 72B speed test
Our Qwen2.5-72B test on RTX PRO 6000 Blackwell: text-processing speed, the effect of longer inputs and the complete measured results.
Can a 235B model run on a single GPU server?
Qwen3-235B with GPU, CPU and RAM: shared speed, memory and 24-task PL/EN tests. We explain when a larger model could be a practical choice.
How quickly does an AI server answer one or several requests?
We measured AI response times with one and several simultaneous requests: results for 8B, 32B and 72B models and a four-hour GPU server test.
What does the CPU do when an LLM runs on the GPU?
What does the CPU do when an AI model runs on a GPU? Our CPU measurements, percentages explained, and practical examples of work around the model.
A model on GPU or split across GPU and CPU?
The same Qwen2.5-72B model on GPU and with CPU/RAM assistance. Response-time and memory comparisons under our shared CORE 1.1 methodology.
How much energy does a GPU need for AI work?
GPU power during a four-hour AI test. Understand the difference between board power, energy per task and whole-server consumption.
Ubuntu, Debian or Rocky Linux for LLM inference?
Ubuntu, Debian or Rocky Linux for your own AI? Choose by application requirements and team experience, with details of our verified environment.
System RAM and cache: how much memory does an LLM server need?
RAM versus VRAM in an AI server: memory during loading and responses, model comparisons, and a practical guide to file cache and application headroom.
Which AI models fit in 96GB of GPU memory?
GPT-OSS-120B and Qwen 72B Q8 on one 96GB GPU, plus a 235B model assisted by RAM. Our memory and speed measurements, with practical examples.
How to run your own AI model API on a GPU server
Run an AI model and connect it to your application. A practical Qwen3-8B and llama.cpp guide with verified files, commands and private SSH access.
15 seconds of AI video from one image: Wan2.2, LightWan2.2 and LTX-2.5
9 image-to-video AI clips: Wan2.2, LightWan2.2 and LTX-2.5 on RTX PRO 6000. Compare 15-second scenes, sound, generation time, GPU memory and energy.
From a still image to video: Wan2.2 on RTX PRO 6000 Blackwell
12 Wan2.2 AI videos on RTX PRO 6000 Blackwell: 480p and 720p, generation times, GPU memory and practical comparisons. Watch the clips and download the settings.
READY FOR YOUR OWN GPU INFRASTRUCTURE?
Configure the server and send a B2B request.
Choose your configuration and see the cost of the full term instantly. The calculator shows the current number of servers available to order.