Skip to Content
DE
Nelpx Systems · Configurator

GPU Servers & AI Workstations.
Configured in minutes.

From 1U GPU servers to the NVIDIA HGX platform, from RTX 5090 workstations to quad RTX PRO 6000 systems — and up to 192 CPU cores for HPC & simulation: put together your configuration and we will typically deliver a consolidated quote within 48 hours.

24base systems — servers, HGX & workstations
up to 8 GPUsper system — RTX PRO, L40S, H200, HGX
up to 4.6 TBDDR5 ECC RAM in the 4U server
B200 & B300NVIDIA HGX platforms available to configure
Configure AI systems

Configuring an AI server:
It's this simple.

Whether you are configuring a GPU server for the data centre or an AI workstation for your desk — the path to a quote is the same. No registration, no hidden steps.

At Nelpx you configure AI servers, GPU workstations and HPC systems online in three steps — and typically receive within 48 hours a consolidated quote from our partner network.

  1. Choose system type & base system

    GPU server (1U–5U rack, incl. NVIDIA HGX) or GPU workstation (AI development / AI training) — 24 validated base systems.

  2. Customise the hardware

    Processor (AMD EPYC, Intel Xeon, Ryzen, Threadripper PRO), ECC RAM, GPUs, NVMe storage and operating system — every option adapts to the platform.

  3. Download PDF & get your quote

    Download your configuration as a PDF and send it to sales@nelpx.de — quote typically within 48 h.

As of July 2026 · No list prices — every configuration is quoted per project across multiple manufacturers.

Konfigurator

Your system.
Your configuration.

First choose between GPU server and GPU workstation, then your base system — all further options adapt automatically to the selected platform.

01 · What would you like to configure?
GPU Server Rack systems from 1U to 5U — AMD EPYC or Intel Xeon, up to 8 GPUs, including NVIDIA HGX platforms (H200, B200, B300).
1–5 Uup to 8× GPUHGX SXM
GPU Workstation AI workstations for development & training — from the RTX 5090 to a quad setup with 384 GB of VRAM, right under your desk.
RTX 5090 / RTX PROup to 4× GPUGB10 Mini
02 · Base system choose the platform — options adapt
03 · Processor
04 · Memory
05 · GPUs / Accelerators
06 · Storage
07 · Operating system
08 · Number of systems
09 · Your details for PDF & quote
Decision guide

GPU server or
GPU workstation?

The most important question before configuring. In short: services for multiple users belong on a rack server — the personal development system on your desk.

Criterion GPU ServerRack · 1U–5U · up to 8 GPUs GPU WorkstationTower / mini PC · 1–4 GPUs
DeploymentData centre or server room (19" rack)Office and workplace — quiet air cooling
Typical workloadsAI inference services, multi-user environments, training, HPC simulation in 24/7 operationAI development, fine-tuning, local training, data analysis, simulation at the workplace
GPUsUp to 8× PCIe (RTX PRO, L40S, H200 NVL) or 8× SXM (HGX H200/B200/B300)1–4× (RTX 5090, RTX PRO 5000/6000 Blackwell) — up to 384 GB VRAM
ProcessorsAMD EPYC (up to 192 cores), Intel Xeon 6 (up to 128 cores), single/dual socketRyzen 9, Core Ultra, Threadripper PRO (up to 96 cores)
MemoryRDIMM ECC DDR5 — up to 4,608 GBDDR5 up to 192 GB, Threadripper PRO with ECC up to 512 GB
Operation24/7 operation, redundant Titanium PSUs, management LAN for remote administrationWorkplace operation, cooling designed for days of full load
ScalingCluster-ready — InfiniBand / Spectrum-X Ethernet up to 800 Gb/s per GPUSingle system; AI Mini (GB10) can be paired

Not sure? Describe your workload in the configurator under "Use case" — we will recommend the right platform in the quote.

Guide

From AI server to
HPC workstation.

Four typical requirement profiles — and the systems we configure for them. Every recommendation links straight into the configurator.

Profile 01 · Inference

LLM inference & AI services

Language models, RAG systems and AI APIs for many concurrent users: what counts here is GPU memory and throughput per rack unit — in 24/7 operation with redundant power.

Recommendation: 2U GPU server with L40S, RTX PRO 6000 Server or H200 NVL · Configure a server
Profile 02 · Development

AI development & fine-tuning

Adapting models, LoRA training, prototyping with PyTorch and TensorFlow — locally, reproducibly and without cloud costs. From the RTX 5090 (32 GB) to the RTX PRO 6000 (96 GB ECC) for 70B models.

Recommendation: AI development workstation or AI Mini (NVIDIA GB10) · Configure a workstation
Profile 03 · Training

Model training

Training your own models from scratch or large fine-tunes: multi-GPU systems with NVLink or PCIe interconnect — from 2× RTX PRO 6000 (192 GB VRAM) under the desk to 8× SXM in an HGX server.

Recommendation: AI training workstation (up to 4× GPU) or HGX H200/B200/B300 · Configure a training system
Profile 04 · HPC

HPC & simulation

CFD, FEM, molecular dynamics and rendering need cores, ECC memory and I/O rather than maximum GPU density: AMD EPYC with up to 192 cores and 4,608 GB RAM in the server — Threadripper PRO with up to 96 cores at the workplace.

Recommendation: HPC server (EPYC/Xeon 6) or HPC workstation (Threadripper PRO) · Configure an HPC system
Why Nelpx

Configuring is easy.
Building it right is our job.

We check every configuration for compatibility, cooling and power — and optimise it for your workload before it goes into production.

01 — SOURCING Across manufacturers

We build on Gigabyte, Supermicro and ASUS platforms — and choose the chassis that fits your GPU loadout, cooling and data centre.

02 — QUALITY Tested before delivery

Every system undergoes intensive load and quality testing before delivery — GPUs, memory and storage included. What arrives at your site is proven stable.

03 — CONSULTING From server to cluster

Whether a single workstation or an HGX cluster with InfiniBand fabric: we advise down to the network and software level — GDPR-compliant, from Germany.

Frequently asked questions

Configuring AI systems:
Answered briefly.

The questions we hear most often before configuring AI servers, GPU workstations and HPC systems.

How can I configure an AI server?
In three steps right on this page: 1. choose the system type (GPU server or GPU workstation), 2. choose a base system (e.g. a 2U AMD EPYC with up to 4 GPUs or NVIDIA HGX B300), 3. customise processor, memory, GPUs, storage and operating system. Download the configuration as a PDF or send it directly to sales@nelpx.de — quote typically within 48 hours.
GPU server or GPU workstation — which is the right system?
GPU servers belong in the rack: 24/7 operation, redundant PSUs, remote management, up to 8 GPUs and cluster scaling — ideal for inference services, multi-user environments and training. GPU workstations sit at the workplace: quiet, 1–4 GPUs (up to 384 GB VRAM), ready to use — ideal for development, fine-tuning and local training. Rule of thumb: services for multiple users → server; personal development system → workstation. Details in the comparison table.
Which GPU do I need for LLM inference and training?
For getting started (models up to ~8B parameters), an RTX 5090 (32 GB) is sufficient. Up to ~32B we recommend the RTX PRO 5000 Blackwell (48 GB ECC), for 70B models the RTX PRO 6000 Blackwell (96 GB ECC). In servers, L40S (48 GB) and H200 NVL (141 GB HBM3e) offer more throughput for inference services. For training large models: multi-GPU systems — workstations with up to 4× RTX PRO 6000 Max-Q (384 GB VRAM) or HGX servers with 8× SXM GPUs.
What is the difference between an AI server and an HPC server?
The platform is often the same — the difference is in the loadout. AI servers are GPU-centric: maximum GPU performance and memory per rack unit. HPC servers for simulation (CFD, FEM, molecular dynamics) are CPU- and memory-centric: many cores (EPYC up to 192, Xeon 6 up to 128), lots of ECC RAM (up to 4,608 GB), fast storage. The configurator covers both profiles — from GPU-dense AI servers to core-heavy HPC systems.
What does an AI server or AI workstation cost?
The price depends heavily on GPU loadout, memory and quantity — a workstation with an RTX 5090 plays in a different league than an HGX system with 8 SXM GPUs. That is why we do not quote flat prices but calculate every configuration per project through our partner network — including alternatives when a different loadout delivers more performance per euro. Quote typically within 48 h.
Which operating system is right for AI servers and workstations?
The standard in the AI world is Ubuntu LTS — best support for CUDA, PyTorch, TensorFlow and container workflows. Alternatively: Windows Server 2025 (server) or Windows 11 Pro (workstation) — or delivery without an operating system. On request we set up drivers and the AI stack for you.
Can the systems be expanded into a cluster later?
Yes. All GPU servers offer PCIe Gen5 or OCP 3.0 slots for InfiniBand or Ethernet cards (10 GbE to 800 Gb/s); HGX B300 systems come with ConnectX-8 SuperNICs onboard. We are happy to plan the cluster fabric (NVIDIA Quantum InfiniBand or Spectrum-X Ethernet) along with your system — including switches and cabling.
How quickly will I receive a quote and delivery?
Your quote typically within 48 hours after you submit the configuration. Delivery time depends on component availability (especially GPUs) and is stated bindingly in the quote.
Next step

Your configuration.
Our quote within 48 h.

Send us your configuration from the configurator — or start with a no-obligation initial call. We advise across manufacturers and deliver across Germany and Europe.