AI Hardware - WestAI

AI-Hardware

Training large AI models requires strong computing power. With WestAI, you can access the specialized High Performance Computing (HPC) computing infrastructure of RWTH Aachen University and Forschungszentrum Jülich. Access is exploratory for research institutions and companies.

Use our AI infrastructure for your AI idea!

How to make use of our AI computing resources

RESEARCH Institutions

If you come from an academic enviroment and are interested in computing time on WestAI’s High Performance Computing(HPC) infrastructure, you can submit an application directly. The application process has been designed to be very lean and can be realized with little effort by following the five steps described below.

Important: Please note the project-related requirements.

To the application

Companies

Do you belong to a small or medium-sized company or start-up and are pursuing an exciting AI research project? Do you have an idea for an AI solution, but lack the necessary hardware? We will be happy to support you personally with your application for our AI computing resources and advise you beyond that. Please contact us directly.

Contact

Note: We can currently offer you these services free of charge. Up to 10,000 GPU hours can be used. We reserve the right to reduce the number of computing time projects if there is a high demand for the systems.

Your partner in High Performance Computing

With decades of experience, the IT Center of RWTH Aachen University and the Jülich Supercomputing Centre at Forschungszentrum Jülich (FZJ) are your experts for High Performance Computing (HPC) at WestAI. We not only guarantee stable operation of our computing clusters, but also ensure continuous updates of the software stack and progressive modernization of the infrastructure. Thanks to our established processes for handling support requests, we can respond quickly to your needs. With us as your partner, you leave the demanding responsibility for HPC infrastructure to proven IT professionals so that research and development can take center stage.

WestAI and its partners, RWTH and FZJ, have created easy access for you to specialized computing infrastructure for training large AI models.

Our AI hardware in detail

To enable the training of large AI models, WestAI offers specialized HPC computing infrastructure. We have a total of 31 servers, each equipped with four NVIDIA H100 GPUs. The exact specifications can be found by clicking on the modules below.

JURECA / Forschungszentrum Jülich

16 computing nodes each with

2x Intel Xeon Platinum 8462Y (Sapphire Rapids) CPUs
A total of 64 cores per node @2.8 GHz (Base Freq.)
512 GB DDR5 RAM per node
4x NVIDIA H100 GPUs (incl. NVLink Interconnect)
94 GB HBM2e per GPU

CLAIX-2023 / RWTH Aachen University

15 computing nodes each with

2x Intel Xeon Platinum 8468 (Sapphire Rapids) CPUs
Total of 96 cores per node @2.1 GHz (Base Freq.)
512 GB DDR5 RAM per node
4x NVIDIA H100 GPUs (incl. NVLink Interconnect)
94 GB HBM2e per GPU

More Details

Both systems have an InfiniBand high-performance network that connects the individual computing nodes with each other.

We complement the hardware with an extensive software stack. This includes all common machine learning frameworks such as Scikit-Learn, PyTorch and TensorFlow. It is also possible to bring your own Apptainer container and use it on the system. This allows our users to start their research quickly and without much preparation time.

As with all large HPC systems, we use Linux on our computing clusters. Users usually interact with it via the console. For inexperienced users, the lack of a graphical user interface can be a major barrier to entry. We therefore also offer JupyterHub instances. This allows Jupyter Notebooks to be run directly on the computing cluster via the browser. This user-friendly environment makes it possible to test the system and develop applications efficiently. If you need support with this process, we will be happy to help you in this phase too.

Requirements for computing time applications

Applications for computing time can currently only be submitted by people from an academic environment. We are happy to help people from the corporate context personally with the application – please contact us directly.

Each application must name a Principle Investigator (PI) and a Person of Contact (PC) with e-mail addresses who will act as the main scientific and technical contact persons for the project. More information on the requirements for computing time applications can be found by clicking on the following module.

More information

Important: The following authorized users may apply for computing time projects at WestAI as Principal Investigator:

Leading scientists (usually with a doctorate) who belong to a state or state-recognized, institutionally accredited university in Germany (including universities of applied sciences).
Senior scientists (usually with a doctorate) who belong to a public or state-recognized research institution in Germany.

The following requirements must also be met or the following information provided for a computing time application:

A title (250 characters) and a project description (500 to 2000 characters – approx. one page). This description should have a scientific and project-oriented character. The AI reference and GPU requirements of your project should be evident from the description.
An indication of the research focus. This is done according to DFG guidelines, but can be conveniently selected from main and sub-categories via drop-down lists.
A technical description of the project (300 to 1000 characters – approx. half a page), e.g. which codes, frameworks, data sets or models will be trained and used.

Note: The online form for computing time applications is in English and can also be filled out in German if necessary.

Since our offer is specifically designed for GPU resources and AI, please request computing time from the following alternative providers in other cases:

If you prefer to use CPU resources:
- National High Performance Computing(NHR)
- Gauss Center for Supercomputing (GCS)
If you need computing time for Bachelor’s or Master’s theses: Please request Tier 3 resources from your local university HPC provider. You can find more information at e.g:
- HPC.NRW

Five steps to your computing time

Do you come from an academic environment? Then you can submit your application for computing time directly via WestAI. The following instructions will show you step by step how to submit your application for computing time. If you do not know your exact requirements or need support, we will be happy to advise you.

Or do you come from a business context? Then we will be happy to help you with your application. Simply get in touch with us.

Step 1: Location selection

Visit our online form for WestAI computing time applications via the “Joint Application, Review, and Dispatch Service” (JARDS) at the following link.

Now select the location (RWTH Aachen or FZJ) where the computing time is to be requested. To make the right choice, the following special features of the individual systems should be taken into consideration:

CLAIX-2023 at RWTH Aachen University

Allows shared use of computing nodes (shared nodes) so that, for example, jobs that only require 1 GPU can also be executed.
The 15 GPU computing nodes that were purchased for WestAI are embedded in a larger partition consisting of a total of 52 systems.

JURECA of the Jülich Supercomputing Center (JSC)

Shared use of nodes is not yet planned here. Applicants should therefore be able to utilize at least one entire node with 4x NVIDIA H100 GPUs.
Computing time projects at the JSC have higher file quotas (data volume and number of files). If you need to process or store larger data sets or a large number of files in your project, the JSC is the right choice.

Step 2: Login / Registration

The login procedure is either via RegApp (if you already have an HPC account at RWTH Aachen University) or via e-mail verification.

Step 3: Application

Fill out the computing time application. Please note the requirements described above and the information to be provided.

Please also note that the scope of computing time projects within WestAI is limited to 10,000 GPU hours. The standard duration of WestAI projects is 12 months and cannot be extended.

If you have a greater need for resources or wish to continue your project afterwards, you can apply for larger proposals via the following links:

Note: We reserve the right to reduce the scope of computing time projects in the event of high demand for the systems.

Step 4: Finalization

Finalization of the application. This can be done on the last page of the application. After finalization, a summary of the application will be sent by e-mail to the specified PI, who must sign the document. The signed application must then be uploaded to the JARDS website.

Step 5: Review and Assessment

Done! Your computing time application will now be reviewed by us. After the assessment, you will be informed about the approval decision and the next steps.

We look forward to helping you realize your AI idea with our computing power!