Skip to content
Home Strato UCloud AI Cloud AI-LAB TAAURUS External HPC Mistral
Get Support

TAAURUS system overview

Key characteristics:

  • Secure-by-design platform architecture
  • Support for sensitive and regulated research data
  • Flexible compute resources for data processing, AI, ML, and HPC workloads
  • Role-based access to data

Platform Architecture

TAAURUS is structured in four main layers:

  1. Frontend
    Project-specific virtual machines with preinstalled software packages.

  2. Backend
    GPU servers for AI/ML and high-performance computing, managed through a Slurm queueing system.

  3. Storage Platform
    Secure, high-performance storage with a separate protected storage area for each project.

  4. Data Transfer Layer
    Controlled import and export services for moving data into and out of the platform.

Hardware and Compute Resources

GPU compute cluster

The GPU compute environment is built for high-performance, GPU-accelerated workloads such as machine learning, AI training, and advanced data processing.

Compute node specification

Each of the two physical compute nodes includes:

  • Model: Lenovo SR675v3
  • CPU: Dual AMD EPYC 9354 (2 × 32 cores)
  • Memory: 768 GB RAM per node
  • GPU: 8 × NVIDIA L40S per node
  • GPU memory: 48 GB GDDR6 per GPU
  • CUDA cores: 18,176 per GPU
  • Networking: 2 × 100 Gbps Ethernet

Storage and Data Handling

Secure storage

TAAURUS uses Hitachi HCSF as its high-performance storage platform.

Storage features include:

  • project-specific secure storage areas
  • encryption at rest
  • high-performance flash storage

Data isolation

Each project is assigned its own dataset and storage area. Users cannot access datasets outside their own project scope.

Access and Authentication

User access requires:

  • login through a Remote Desktop Gateway
  • multi-factor authentication
  • an AAU account
  • Connection to AAU network
  • membership in the relevant project-specific AD groups

Security higlights

TAAURUS includes:

  • isolation from the public internet
  • role-based access control
  • project-based access separation
  • encrypted data transfer
  • encryption at rest on storage systems
  • controlled and logged import/export workflows
  • restricted network traffic via firewall rules

Data Import and Export

Import and export are controlled workflows and require approval within the platform governance model.

Import

  • Data import requests are handled through ServiceNow.
  • Import must be approved by the Principal Investigator (PI).
  • Imported data must be correctly classified and labelled.

Export

  • Data export must also be approved by the PI.
  • Smaller exports can be transferred to a dedicated export web server for download.
  • Export workflows are logged and controlled.
  • The longer-term design includes automated transfers through NiFi.

Available Software

Research tools

The platform provides access to a range of commonly used research and analysis tools.

Available software includes:

  • Firefox
  • LibreOffice Calc
  • LibreOffice Writer
  • Atril Document Viewer
  • DICOMscope and Weasis
  • R and RStudio
  • Python and Anaconda
  • Stata
  • Matlab

Software packages can be updated by ITS through service requests.