RCC supports artificial intelligence research at UQ in direct and indirect ways, from provision of specialised supercomputing infrastructure—including an HPC and frameworks—to technical support and training events.
HPC Wiener
Wiener is RCC’s most noticeable infrastructure for image processing, deep learning algorithms and machine learning, with the supercomputer delivering unprecedented performance for UQ’s artificial intelligence community.
Wiener’s graphics processing unit (GPU) architecture provides massively parallel processing of data through the use of hundreds of thousands of stream processor units known as CUDA cores and TensorFlow cores, making it well suited to machine learning and image processing problems.
Wiener has many machine learning, artificial intelligence, deep learning and machine vision frameworks installed, some of which include TensorFlow, PyTorch and Horovod. These are the leading frameworks for writing machine learning applications.
Tensorflow and other frameworks like it are optimised to utilise GPU architectures, which means machine learning training and classification can be about 100 times faster over traditional HPC (core processing unit—CPU) approaches. Further, Wiener is agnostic in its platform support, allowing for the use of CUDA, OpenCL and OpenAAC.
“Much of the current revolution occurring in machine learning is being driven by the fact that problems that were previously intractable are now possible to solve in realistic times because of these orders of magnitude speed-ups in ‘learning’ problems,” said Dr Nick Hamilton, an RCC eResearch Analyst and Bio-Mathematician at UQ’s Institute for Molecular Bioscience (IMB).
Wiener began operating at UQ in late 2017, with recent upgrades to improve performance. RCC—together with leading researchers from UQ’s faculties of Science, Medicine, Engineering, Architecture and Information Technology—recently received a grant to almost triple the size of Wiener.
Wiener was the first supercomputer in the Asia-Pacific region to use the Nvidia Volta GPU, and was originally installed to support UQ’s world-class IMB-based Lattice Light Sheet Microscope (LLSM).
Support
RCC works in collaboration with groups campuswide to appropriately install, compile, optimise and scale machine learning frameworks to support a number of AI-centric initiatives.
One such significant collaborative effort was for a UQ digital pathology project (read our separate story about this project). The work enabled linearly scalable use of GPUs leveraging the supercomputing interconnect in Wiener, to allow for scalable distributed machine learning. This resulted in a “substantial” reduction in time-to-solution for the project team, according to RCC Chief Technology Officer Jake Carroll.
Training
RCC supports training in machine learning methods through weekly UQ Hacky Hours where machine learning experts are on hand to answer questions; running occasional introductory machine learning for bio-imaging workshops; and through events it sponsors, such as UQ’s annual Winter School in Mathematical and Computational Biology and Brisbane’s HealthHack, an annual hackathon solving real-world healthcare and medical research problems proposed by UQ and other researchers. HealthHack is increasingly using machine learning approaches to solve research problems. (See our separate story about this year’s Brisbane HealthHack.)
This year, RCC ran its first Introduction to Network Visualisation and Cytoscape workshops. The September workshop was so popular that RCC ran it again on 1 October. Cytoscape is a free, powerful multi-platform application for visualisation and analysis of networks.
Ask us how we can help your AI research: rcc-support@uq.edu.au.
