• Skip to primary navigation
  • Skip to main content
  • Skip to footer
Open Edge HPC Initiative

Open Edge HPC Initiative

Open Edge HPC Initiative

  • Home
  • About
    • Our Vision
    • Get Involved
    • Supporting Organizations
  • Undertakings
    • Arm Demonstration Systems
    • Workshops & Hackathons
  • News
  • Contact

OEHI Opens Arm Based HPC and AI Cluster at Goethe University of Frankfurt to Partners

2 February 2022

Huawei is a leading global provider of information and communications technology (ICT) infrastructure and smart devices with innovation focus on customer needs. Huawei’s all-domain, full-lifecycle HPC solution is based on Arm technology. The new pilot cluster at Goethe University Frankfurt aims to grow the Arm ecosystem for HPC and AI application use cases. The Goethe University Frankfurt is part of the National High-Performance Computing Alliance. The hardware was assembled and installed by MEGWARE Computer Vertrieb und Service GmbH in December 2021. The cluster software installation, configuration and validation is jointly managed by Goethe University Frankfurt (GUF), OEHI community efforts and Huawei.

The following table shows the currently installed nodes and servers of the cluster:

OEHI Huawei Arm Cluster at Goethe University of Frankfurt
28xStandard Compute Node – TaiShan 200 (Model 2280)
  • 2 x Kunpeng 920 processor (ARMv8 AArch64; 64 cores; 2.6GHz; 180W)
  • 128GB main memory (16x 8GB, one DIMM per channel)
  • 1x 100Gbit/s EDR Infiniband HCA
1xDevelopment Compute Node – TaiShan 200 (Model 2280)
  • 2 x Kunpeng 920 processor (ARMv8 AArch64; 64 cores; 2.6GHz; 180W)
  • 128GB main memory (16x 8GB, one DIMM per channel)
  • 1x 100Gbit/s EDR Infiniband HCA
  • 2 x 960GB SSD SATA 6Gb/s
2xIO Node – TaiShan 200 (Model 5280)
  • Metadata storage: 2x 960GB SSD SATA 6Gb/s (RAID 1)
  • Object storage: 32x 1.2 TB HDD SAS 12Gb/s; 10.000rpm (RAID 10)
  • Mgmt. storage: 4x 1.2 TB HDD SAS 12Gb/s; 10.000rpm (RAID 10)
1xAI Training Node – Atlas 800 (Model 9000)
  • 4x Kunpeng 920 processor (ARMv8 AArch64)
  • Neural Processing Unit (NPU): 8x Huawei Ascend 910 with 32 AI cores and 32GB HBM2 memory
  • 1024GB main memory: 32x 32GB DDR4 2933MHz RDIMM
  • Local storage: 2x 960 GB SSD SATA 6Gb/s
  • Data storage: 4x 3.2TB SSD NVMe
1xAI Inference Node – Atlas 800 (Model 3000)
  • 2 x Kunpeng 920 processor (ARMv8 AArch64)
  • 512GB main memory: 16x 32GB DDR4 2933 MHz RDIMM
  • Local storage: 2x 960GB SSD SATA 6Gb/s
  • Data storage: 4 x 960GB SSD SATA 6Gb/s
  • GPU: 5x Atlas 300 AI Inference Card; 32GB; PCIe3.0 x16

The nodes are connected via a non-blocking EDR Infiniband 100GBit/s fabric for high bandwidth, low latency communication as well as an Ethernet network for deployment and management.

Open source software is used for the cluster software environment. The cluster is built on Rocky Linux 8 with SLURM as a job scheduler. Lustre in combination with ZFS is used to provide a high-performance parallel filesystem to the users. EasyBuild is used extensively to install and configure software packages: GCC, OpenMPI, Python, SciPy and others. The AI software stack is built on top of Huawei’s CANN (Compute Architecture for Neural Networks) toolkit, which will support AI frameworks like TensorFlow, PyTorch, ONNX and Mindspore.

Arm is the leading technology provider of processor IP, offering the widest range of processors to address the performance, power, and cost requirements of every device. Arm processors are integral to the European Processor Initiative and hence the European HPC ecosystem. This realization of an Arm-based HPC and AI cluster at the Goethe University of Frankfurt aims to support

  1. Porting of applications to Arm based architectures
  2. HPC application performance profiling and exploration
  3. Developing an Arm based software ecosystem
  4. Supporting machine learning applications on Atlas AI computing platform
  5. Evaluating the storage software ecosystem for Arm based architectures
  6. Training and Education

Access to the cluster is open to all OEHI members; for others, access can be provided for evaluation purposes.

For any queries, please contact karthee dot Sivalingam at huawei.com

Top

Footer

Privacy Policy

Imprint

Copyright © 2022 Open Edge HPC Initiative

WebSite by Black Studio