Cray CS-Storm

Image f. Cray CS-Storm

The Cray CS-Storm is a specialized system designed to accelerate artificial intelligence (AI) workloads. The machine complements HLRS’s high-performance computing infrastructure by addressing user demand for processing-intensive applications for machine learning, deep learning, and high-performance data analytics. 

The Cray CS-Storm supports a wide variety of well-known and established AI frameworks and tools such as Apache Spark, Python-based data science libraries like scitkit-learn, and frameworks for deep learning including TensorFlow and PyTorch. It also includes 64 NVIDIA Tesla V100 GPUs.

System components

Cray CS-Storm: compute (deep learning) partition

This partition consists of 8 GPU Nodes, each of which is configured as follows:

  • 8x V100 SXM2 32GB HBM2, NVLink 2
  • 2x CLX 6240, 18c, 2.6 GHz (150W)
  • 24x 32 GiB DDR4-2933; 768 GiB total
  • 4x P4510, NVMe SSD, 2.5”, 2 TB
  • 2x S4510, SATA SSD, 2.5”, 240 GB
  • 4x Mellanox CX-4, x8, VPI Single-Port, QSFP28

Cray CS500: Spark (ETL/ML) partition

8x CPU nodes in 2 CS500 3211 (2HE). Each node is configured as follows:

  • 2X CLX 6230, 20c, 2.1 GHz (125W)
  • 12x 32 GiB DDR4-2933, 384 GiB total
  • 1x P4510, NVMe SSD, 2.5”, 2 TB
  • 2x S4510, SATA SDD, 2.5”, 3.8 TB
  • 1x S4510, SATA SDD, 2.5”, 240 GB
  • 1x Mellanox CX-6 HDR100, 100 GB/s On-board
  • 2x SFP+, 10 GB/s

Software, compiler

  • Urika-CS AI Suite

Network

  • HDR100 Inifiband

Learn more about HLRS's activities in artificial intelligence and high-performance data analytics.

AI and data analytics