
Which solution is right for your project? Pre-installed software for AI development
Die Auswahl sollte nicht allein anhand der Anzahl der GPUs erfolgen. Entscheidend ist, wie Ihr Workload aufgebaut ist und welche Anforderungen sich daraus an Speicher, Rechenleistung, Kühlung, Stromversorgung und Erweiterbarkeit ergeben.
| Anforderung | Geeignete Systemklasse und typische Einsatzbereiche |
|---|---|
| Lokale Entwicklung und Inferenz bei geringem Platzbedarf | Kompakte Systeme mit Unified Memory eignen sich für Prototyping, Modelltests, lokale Assistenten sowie ausgewählte Inferenz- und Fine-Tuning-Aufgaben. |
| Fine-Tuning und Entwicklung größerer Modelle | Deskside-Systeme mit großem Speicherbereich sind sinnvoll, wenn größere Modelle, lokale Datenbestände oder anspruchsvollere Entwicklungsaufgaben im Mittelpunkt stehen. |
| Klassisches Multi-GPU-Training | Für umfangreiche Trainings- und Entwicklungs-Workloads kommen Multi-GPU-Workstations oder GPU-Server infrage. Bei der Auswahl sind unter anderem GPU-Abstände, PCIe-Anbindung, Stromversorgung, Kühlung, GPU-Speicher und die Skalierbarkeit des Trainingsverfahrens zu berücksichtigen. |
| Individuelle Forschungs- und Engineering-Workloads | Für Forschung, Engineering, Simulation und datenintensive AI-Workloads ist häufig eine individuelle Kombination aus CPU, ECC-Arbeitsspeicher, NVMe-Speicher, einer oder mehreren NVIDIA-GPUs sowie geeigneter Netzwerktechnik sinnvoll. |
AI Workstation products


NVIDIA DGX Spark A Grace Blackwell AI supercomputer on your desk

NVIDIA DGX Spark A Grace Blackwell AI supercomputer on your desk

NVIDIA DGX Spark A Grace Blackwell AI supercomputer on your desk

Gigabyte AI TOP ATOM 4TB PCI-E5 ATAGB10-9000

Gigabyte AI TOP ATOM 1TB PCI-E4 ATAGB10-9002

ASUS Ascent GX10 Compact, Powerful, and Scalable

2U GPU-Server/Workstation inkl. 2x NVIDIA RTX PRO 6000 Blackwell Max-Q WS
SYS-2115HV-TNRT
1x Dedicated IPMI Management

inkl. 1x NVIDIA RTX 5090
devCube-5090
1x 1GbE RJ-45 (IPMI dedicated)

inkl. 1x NVIDIA RTX PRO 6000 WS
devCube-PRO6000WS
1x 1GbE RJ-45 (IPMI dedicated)

inkl. 2x NVIDIA RTX PRO 6000 Max-Q
devCube-PRO-6000-Max-Q
1x 1GbE RJ-45 (IPMI dedicated)

inkl. 2x NVIDIA RTX PRO 5000
devCube-PRO-5000
1x 1GbE RJ-45 (IPMI dedicated)

inkl. 2x NVIDIA RTX PRO 4500
devCube-PRO-4500
1x 1GbE RJ-45 (IPMI dedicated)

inkl. 2x NVIDIA RTX PRO 4000
devCube-PRO-4000
1x 1GbE RJ-45 (IPMI dedicated)

inkl. 2x NVIDIA RTX 6000 Ada
devCube-6000-Ada-R2
1x 1GbE RJ-45 (IPMI dedicated)

inkl. 2x NVIDIA RTX 5000 Ada
devCube-5000-Ada-R2
1x 1GbE RJ-45 (IPMI dedicated)

inkl. 4x NVIDIA RTX PRO 6000 Max-Q
devCube-PRO-6000-MaxQ
1x 1GbE RJ-45 (IPMI dedicated)
What's inside Pre-installed software for AI development

Our AI training workstations are equipped with the latest and most powerful software solutions to optimise your deep learning and AI training:
- Caffe: A fast and efficient deep learning library that is particularly suitable for Convolutional Neural Networks (CNNs).
- Torch: A framework that supports dynamic networks and a simple programming language and is ideal for machine learning and deep learning.
- Theano: A Python library that simplifies the definition, optimisation and evaluation of mathematical expressions with multi-dimensional arrays.
- TensorFlow: An open-source framework specifically designed for training and inference of deep neural networks.
- CUDA (including cuDNN): A parallel computing platform and API from NVIDIA that utilises the power of GPUs to perform intensive computing tasks. cuDNN is a GPU-accelerated library specifically optimised for deep learning.
WHY AI? Applications of AI Training Workstations
AI training workstations provide the computing power that companies, research institutions, and engineering teams need for the development, training, fine-tuning, and local inference of AI models. Depending on the configuration, they are suitable for the following tasks, among others:
Development and Prototyping
Entwickeln und testen Sie KI-Modelle, Machine-Learning-Anwendungen und lokale Assistenten in Ihrer eigenen IT-Umgebung. So können Entwicklungsprozesse unabhängig von verfügbaren Cloud-Ressourcen geplant und durchgeführt werden.
Training and Fine-Tuning
High-performance NVIDIA GPUs support model training and fine-tuning. The choice of a suitable workstation depends, among other factors, on model size, GPU memory, the dataset, the training method, and the desired runtime.
Local Inference
Run trained models locally, for example, for generative AI, computer vision, natural language processing, or intelligent data analysis. Actual performance depends on the model, memory requirements, software, and latency requirements.
Research, Engineering, and Simulation
AI training workstations are suitable for data-intensive research and development tasks, such as image processing, simulation, pattern recognition, and scientific data analysis.
Data-Sensitive Workloads
Local processing can help reduce reliance on external cloud services and ensure that sensitive data is processed within the organization’s own IT environment. Whether a specific use case complies with data protection regulations must be assessed within the relevant security and data protection policy.
Unsere leistungsfähigen Speicherlösungen bieten hohe Kapazität und Geschwindigkeit, um große Datenmengen effizient zu speichern und abzurufen. Perfekt für die Anforderungen von Big Data und maschinellem Lernen.
Unsere AI Training Server sind für maximale Rechenleistung und Flexibilität konzipiert. Sie ermöglichen effizientes Training komplexer AI-Modelle und sind ideal für Forschung und Entwicklung.
Unsere Inference Hardware bietet die nötige Rechenkapazität, um AI-Modelle in Echtzeit auszuführen. Ideal für Anwendungen wie autonome Fahrzeuge, Bildverarbeitung und Sprachsteuerung.
- What is the difference between an AI Training Workstation and a conventional PC?
AI training workstations are designed for high and sustained computational workloads. Depending on the configuration, they feature high-performance NVIDIA GPUs, ample GPU memory, appropriate cooling, fast NVMe storage, ECC memory, and advanced networking or management capabilities.
- Is every workstation suitable for multi-GPU training?
No. Key factors include the number of available PCIe slots, GPU spacing, power supply, cooling, driver and framework compatibility, and the scalability of the training process. Compact unified-memory systems and traditional multi-GPU workstations should therefore be evaluated separately.
- Why are powerful GPUs important for AI training workstations?
GPUs (Graphics Processing Units) are crucial for the training of AI models, as they enable the parallel processing of large amounts of data. This speeds up the training process considerably compared to conventional CPUs.
- How do I choose the right AI Training Workstation for my project?
The choice depends on the specific requirements of your project, including the type of models to be trained, the size of your data sets and your budget. A consultation can help you find the ideal configuration based on your needs.
- Can I use an AI Training Workstation for purposes other than AI training?
Yes, AI Training Workstations are versatile and can be used for a variety of computationally intensive tasks, including video editing, 3D modelling and simulations.
- Can I also perform inference and data analysis using an AI workstation?
Yes. In addition to training and fine-tuning, many systems are also suitable for local inference, computer vision, data analysis, simulation, and development. Which tasks can be meaningfully performed locally depends, among other things, on memory, model size, latency requirements, and software.
- What kind of support do you offer for AI Training Workstations?
Support can range from technical assistance and troubleshooting to regular updates and maintenance services.
- What software comes preinstalled?
That depends on the configuration and the agreed-upon scope of delivery. Options include Ubuntu, NVIDIA drivers, the CUDA Toolkit, cuDNN, PyTorch, TensorFlow/Keras, JAX, and reproducible container environments. Versions and the scope of installation are documented in the proposal.