The Nvidia logo next to NCI's GPU cluster
Gadi's GPU cluster is incredible technology, but its potential is only fully realised through the success of NCI's user community

Throughout November 2021, NCI is delighted to offer the user community for a series of GPU bootcamps. With essential support and guidance from NVIDIA, these multi-day workshops are aimed at users looking to better make use of Gadi’s GPU cluster. At the time of writing, the first of these bootcamps has been warmly received by the NCI user community, with more training to come in the last few weeks of the year.

The appetite for GPU-based computing has accelerated in recent years. As the capabilities of GPU-based research computing have become more apparent, so has the demand from the user community. Responding to the needs of the Australian research community, NCI ensuredQuote: I loved the hands on approach, actually writing code is by far the best way to learn concepts relating to it. Having Fortran options as well was very much appreciated. that the Gadi supercomputer would be a trailblazer for GPU-based supercomputing in our region.

Currently, Gadi contains 640 NVIDIA V100 GPUs across 160 nodes. These nodes are a significant component of NCI’s heterogenous high-performance computing architecture, and represent a substantial increase in overall GPU computing capacity when compared to Raijin, NCI’s previous supercomputing cluster. 

However, hardware procurement is only half the story. Following on from a successful series of GPU workshops last year, NCI and NVIDIA were pleased to be able to offer even more support to the NCI user community in 2021, culminating in in the recently completed N-Ways GPU bootcamp, and the upcoming AI For Science GPU bootcamp.

These bootcamps are hands-on opportunities for NCI users to learn how to use Gadi’s GPU cluster. HPC experts from NVIDIA and NCI were at the ready during the N-Ways bootcamp to assist users with a variety of exercises introducing savvy researchers to GPU programming and analysis.

When surveyed, attendees of this bootcamp responded positively, with many comments lauding the course layout and personnel. Others reflected positively on the hands-on method of teaching, which saw participants engaging with the Gadi GPU cluster in a purposeful way.

Shehzad Hathi from UNSW Canberra at ADFA offered a warm endorsement of the GPU bootcamps, saying that "I often use numerical computations to verify/disprove conjectures and using GPUs could result in a significant speed-up for this kind of work. Therefore, it was quite useful for me to learn and get some hands-on experience with GPUs during the Bootcamp."

Members of the NCI team were also pleased to be a part of the workshop, saying that "Engaging NCI users in the benefits of GPUs is always exciting, and working through issues in real time helps researchers, and let's us learn what the current limits are."

The popularity of this bootcamp, as well as the upcoming AI For Science bootcamp, highlights the user appetite for more training sessions with direct application to Gadi.

NCI's user community can look forward to even more learning opportunities in 2022. Our training team will continue to work with external stakeholders to support Australian research.