New national GPU service goes live

Reserve access to specialised high-end computing power for research

Australian researchers can now access high-end cloud computing power on the ARDC Nectar Research Cloud via the new National Graphics Processing Units (GPU) Service.

GPUs are in high demand by Australian researchers for processing large data sets or images, 3D imaging, machine learning, and computational modelling.

Australian researchers can now reserve GPUs and large memory virtual machines for their research in advance via a user-friendly interface on the ARDC Nectar Research Cloud dashboard. This will help share high-end compute resources amongst researchers and provide quicker and more efficient reservation and access to the resources.

As the Queensland node of the ARDC Nectar Research Cloud, QCIF received $620,000 in funding from the ARDC to procure specialised compute infrastructure for the National GPU Service. QCIF has completed deploying this infrastructure to the QRIScloud and Intersect availability zones, so the service may be accessed by researchers with a national allocation on Nectar. 

Stephen Bird, QCIF’s Manager of Shared Infrastructure Services, including QRIScloud, said the ARDC Nectar Research Cloud’s new, readily accessible national reservation system is “a big leap forward for researchers needing access to specialised compute capability.”

Carmel Walsh, Director of eResearch Infrastructure and Services at the ARDC, said: “We are excited to launch the new National GPU Service on Australia’s national research cloud, the ARDC Nectar Research Cloud. The new service will accelerate research that relies on high-end computer power by providing researchers with a simple way to reserve time to use GPUs and large memory servers.”

Nodes of the ARDC Nectar Research Cloud around Australia, including QCIF, are also able to use the reservation service to provide access to their local (node-funded) GPU servers for their local users.

Professor Jarrod Hurley, Supercomputing Manager at the Centre for Astrophysics and Supercomputing at Swinburne University, said: “The new national GPU service will help speed up the analysis of image data for astronomy researchers at Swinburne, particularly for those involved in multi-institution collaborative projects. 

“Astronomy needs a lot of computing power for machine learning, image processing, and computational modelling, which is one reason why Swinburne was an earlier adopter of GPUs for scientific purposes in our HPC systems. Now being able to expand the available resources to reserve GPUs and large memory servers via an easy-to-use national reservation system is a huge boost to our research.”

The National GPU Service includes new GPU servers, which the ARDC co-invested in, alongside a new reservation system developed by the ARDC. The service is available through a reservation system for projects that meet the criteria for national merit allocation.

QCIF’s infrastructure procurement included a mix of compute nodes for data-intensive and large memory workloads, machine learning and inference workloads, and visualisation workloads, with the specific hardware acquired including:

• 4 x A100 GPU nodes, each with dual NVIDIA A100-80 GPUs (enabling g2.xsmall and g2.medium flavours)

• 4 x A40 GPU nodes, each with dual NVIDIA A40-48 GPUs (enabling g1.medium flavours)

• 4 x 2 TB large memory nodes (enabling h4.large and h4.xlarge flavours)
  

Stephen said the reservation system allows the ARDC and Nectar nodes to fine-tune the service over time based on what is needed to support research. “This includes tuning the duration of reservations and what flavours are available for reservations. The current flavours are just an initial starting point for the service. All of the nodes acquired by QCIF utilise NVMe storage for the root and ephemeral drives of the virtual machine instances to improve I/O performance,” said Stephen.

The ARDC worked with the following partners on the National GPU Service, who provided in-kind co-investment over three years to make the new service available:

• QCIF     

• Intersect

• Monash University

• Swinburne University of Technology

• University of Tasmania.

  
  

View the ARDC’s demonstration of the new National GPU Service via YouTube. The demonstration is a recorded webinar from 25 October 2022. 

Learn more about the National GPU Service.

The ARDC is funded through the National Collaborative Research Infrastructure Strategy (NCRIS) to support national digital research infrastructure for Australian researchers.

This article is based on one published by the ARDC on 13 September 2022.