PC Praha Rome AI server – 4U rack, front fans

AI server PC Praha Rome – EPYC, up to 6 GPUs

2× RTX 5080 16 GB (32 GB VRAM) / 256 GB DDR4 ECC / 2 TB NVMe
333 890 Kč
Skip to product information
PC Praha Rome AI server – 4U rack, front fans

PC Praha · AI and GPU servers

AI server PC Praha Rome – EPYC, up to 6 GPUs

Specifications by configuration

Graphics cards
2–6 GPUs
Graphics memory
32–384 GB
Processor
AMD EPYC 7643
Memory
256–512 GB
Form factor
4U rack
275 942 Kč excl. VAT
333 890 Kč incl. 21% VAT
Graphics cards
RAM
Storage

Assembly and testing: usually 5–7 working days. We often have the server in stock and ship it straight away – we always confirm the date individually.

Graphics cards
RAM
Storage
  • 24-month warranty, claims handled directly by us
  • We help with configuration and software
  • We can run it in our data centre on request

What the build includes

Graphics cards
2× RTX 5080 16 GB 32 GB VRAM in total
Processor
EPYC 7643 · Milan 48C
Motherboard
ASRock ROMED8-2T
Memory
4× 64 GB DDR4 ECC RDIMM 256 GB in total
Storage
2 TB NVMe
Power supply
2000 W power supply
Chassis
4U rack
GPU risers
2× PCIe riser
Cooling
CPU cooler and server fans
Assembly
Assembly, OS installation and 72-hour stress test

The price above is for the complete server including assembly. We are happy to quote a different CPU, network card or chassis.

A 4U rack server on the AMD EPYC platform for companies that want to run language models in-house. Two to six graphics cards, up to 512 GB of ECC memory and remote management. Choose the combination of cards, memory and drive below.

Full description, video and specifications

Which models this build can run

The selected configuration has 32 GB of graphics memory (VRAM). It comfortably fits, for example, these open language models:

  • Seed-OSS-36B28 GB
  • DeepSeek-R2 32B24 GB
  • Qwen3-30B-A3B MoE20 GB
  • Gemma 3 27B18 GB
  • Mistral Small 3 24B14 GB
  • Qwen3-14B12 GB

The larger model Llama 3.3 70B / Qwen3-72B needs roughly 48 GB of VRAM.

Indicative estimate for quantised models (mostly Q4) with headroom for context. Actual memory needs depend on context length and the number of concurrent users.

What we take care of

  • Custom pre-installation

    We pre-install Linux, NVIDIA drivers, CUDA, Docker and the software for your workload (e.g. vLLM or Ollama) – you can start working as soon as you switch it on.

  • 72 hours under load

    Before handover every server runs at full load for 72 hours while we check stability, temperatures and ECC memory errors.

  • Service within 24/48 hours

    We keep key spare parts in stock in Prague. Service visits and part replacement within 24/48 hours.

  • Hosting in our data centre

    No server room? We place the server in our data centre, take care of power, cooling and connectivity, and you access it remotely.

How we work on AI servers

Description

AI server PC Praha Rome is our versatile platform for business AI: a chatbot over internal documents, an assistant for programmers, processing of contracts and e-mails or an API for your own applications. Your data stays in the company and you pay neither for tokens nor for a cloud subscription. The 48-core AMD EPYC 7643 has enough PCIe lanes for all the cards, and the ASRock Rack ROMED8-2T board offers 2× 10 GbE and IPMI remote management.

Configuration

Processor AMD EPYC 7643 (48 cores, SP3 socket)
Motherboard ASRock Rack ROMED8-2T – 2× 10 GbE, IPMI
Graphics cards 2–4× RTX 5080 / RTX 4090 / RTX 5090 / RTX Pro 6000, 2–6× L4 / L40 / Intel Arc Pro B70; connected via PCIe risers
Memory 256 or 512 GB DDR4 ECC RDIMM
Storage NVMe 2 or 4 TB
Power supply 2 kW ATX (2 cards) or 2× 2 kW ATX (4–6 cards)
Chassis and cooling 4U rack, tower CPU cooler, system fans

How much graphics memory you need

Graphics memory Example configuration What runs comfortably
32–64 GB 2× RTX 5080, 2× RTX 4090, 2× RTX 5090, 2× Arc Pro B70 Llama 3.3 70B / Qwen3-72B from 48 GB, smaller models for more users
96–128 GB 4× RTX 4090, 4× RTX 5090, 2× L40, 4× L4 GLM-4.5-Air 106B, Mixtral 8x22B, Qwen3.5-122B-A10B (MoE)
144–192 GB 6× L4, 4× L40, 6× Arc Pro B70, 2× RTX Pro 6000 Qwen3-235B-A22B (Q3–Q4), GLM-4.6 355B from 180 GB
288–384 GB 6× L40, 4× RTX Pro 6000 DeepSeek-V3.2 / R1 671B (Q2), Qwen3-Coder-480B, Kimi-K2 1T (1.58-bit)

Model sizes are indicative (Q4 quantisation unless stated otherwise), based on our configurator. Actual memory requirements depend on context length and the number of concurrent users. For language models the memory of the cards adds up; image and video generation needs the whole model on a single card.

Passive NVIDIA L4 and L40 cards need rack airflow – this 4U chassis is prepared for that. The L4 draws only 72 W, which makes it a good fit for speech transcription, embeddings and document search with low power consumption.

What you get

  • A server built from new components – we assemble and test it in Prague.
  • Pre-installed software for whatever you want to run on the server (e.g. Ubuntu, drivers and CUDA, Docker, vLLM, llama.cpp, Ollama, ComfyUI). It is ready to work as soon as you switch it on.
  • 72 hours of stress testing under full load before handover.
  • 24-month warranty. We keep spare parts in Prague and provide service intervention within 24/48 hours.
  • The server does not have to stand at your premises – you can keep it in our data centre (hosting) and we take care of power, cooling and connectivity.

Operation and installation

The server belongs in a rack or a technical room with cooling. At full load it draws roughly 1.1 to 2.8 kW depending on the cards – we will advise you on the circuit breaker and placement. If you need a different processor or another combination of cards, we will build it to order.

Frequently asked questions

How many cards fit in the Rome server?
Up to four large RTX 5080, RTX 4090 and RTX 5090 cards, and up to six dual-slot L40, Intel Arc Pro B70 and single-slot L4 cards. We build six RTX Pro 6000 cards on the Genoa platform.
Which model will it run?
It depends on the total graphics memory: from 48 GB Llama 3.3 70B runs comfortably, with 128 GB models of around 120 billion parameters (MoE), and with 4× RTX Pro 6000 (384 GB) even the largest open models in strong quantisation.
Why DDR4 and not DDR5?
The EPYC 7003 platform with DDR4 ECC offers the best price/performance for inference on graphics cards – the computation runs on the GPU. For DDR5 and PCIe 5.0 we offer the Genoa and Turin servers.
Can I place the server with you?
Yes. You can keep the server in our data centre; we take care of power, cooling and connectivity, and you access it remotely.

Customer reviews

There are no reviews yet.

Only customers who bought the product from us can leave a review – we send them a review link by e-mail after the purchase.

More AI servers

View all