Scalable AI

Usability

The existing HPC and storage infrastructure will be extended with modern hardware suited especially for data analytics and machine learning applications. This is accompanied by a versatile and extendable software stack for analytics and the simplification of access to analysis backends with notebooks (e.g. Jupyter, RStudio, Apache Zeppelin), complemented by virtual research environments and Sandboxes Streaming applications on HPC.

Performance and Scalability

• Convergence of Big Data and High Performance Computing (HPC)
• Scientific investigations of architectures for classical HPC applications as well as data intensive and iterative work loads
• Performance measurements and optimization of applications and frameworks
• Consideration of energy efficiency especially for Machine Learning/Artificial Intelligence applications

Available Frameworks and Services

offered by the ScaDS.AI Service Center

• Provides modules on the HPC system
• Consults in selection of technology
• Supports development of the solution
• Develops solutions for the simplification of often-used workflows and programs

Available Hardware

• High-Performance computer with Data Analytics island (HPC-DA)
• Focus: fast I/O (high bandwidth, low latency)
• HPC-DA extends previous installation by:
• 32 Machine Learning nodes
  (each: 2 x Power9-CPUs + 6 x NVIDIA V100 with NVLink)
• 2 PB Non-volatile Memory (NVMe), 10 PB Object Storage