The University of Queensland is set to deliver an impactful set of next generation technologies for researchers with the next major expansion of UQ’s main supercomputer.
UQ is working with Hewlett-Packard Australia to deliver this next supercomputer expansion, locally known as Phase 4.0 of the Bunya HPC.
Site preparation and implementation is in progress and Bunya 4.0 will be available for research use once installation has been completed.
This expansion will include some of the most powerful GPUs worldwide, mixed with the latest CPUs, in a solution designed around efficiency and sustainability.
Research Computing Centre (RCC) Director Jake Carroll said: “Bunya Phase 4.0 is the most significant supercomputing investment UQ has made, both in terms of the leap in technology from year to year and in terms of the total amount of computing capability in the phase.
“Bunya Phase 4.0 is a testament to what we can achieve with an evergreen model of sustained supercomputing investment.”
This upgrade will continue to provide UQ researchers and partners with access to technologies which support the wide variety of research undertaken at the University.
Care has been taken to balance the needs of simulation, data processing, exploratory work, machine learning, and the newly popular field of artificial intelligence.
Inside the expansion is 24 of AMD’s flagship AMD Instinct™ MI355x GPUs, 7,680 AMD EPYC 9005 series CPUs, 43,776 GB of CPU RAM, 6,912 GB of dedicated GPU vRAM, and multiple 400Gb/s NDR InfiniBand networks.
AMD’s new flagship Instinct™ MI355x GPUs will be deployed in Bunya to provide UQ researchers with more memory per GPU than any other GPU currently available on the market.
With 288GB of vRAM per GPU and 2,304 GB of vRAM available for multiple GPU workloads, this will enable researchers to run big large language models (LLMs) more efficiently and consolidate many workloads into a more sustainable package. A single Mi355x GPU can do the work of four of Bunya’s existing H100 GPUs.
(Photo by Jake Carroll.)
The AMD Instinct™ MI355x GPUs have 288 GB of HBM3E memory per GPU (with eight chips tightly linked together), 8 TB/second bandwidth, expanded MXFP6 and MXFP4 datatype support for AI and LLM workloads, and FP32 and FP64 support for advanced computational workloads.
Once online, the GPU infrastructure will be available to all UQ researchers with a Bunya account. Visit RCC’s Bunya webpage to learn how to get an account on the supercomputer.
Bunya Phase 4.0 is also configured with the latest direct liquid cooling (DLC) technologies, which provide a more sustainable and efficient way to cool the increasingly energy-intensive AI engines of modern research infrastructure.
DLC technologies reduce the energy required to cool modern chips, improving environmental outcomes.
“We’re saving power for every operation processed and advancing our operating environment with our data centre partner,” said Jake.
Bunya is housed at the Polaris Data Centre in Springfield, Queensland, about 30km from the UQ St Lucia campus.
