NCI recently hosted three GPU training sessions for our users: the Accelerated Data Science GPU Bootcamp, the Distributed Deep Learning Bootcamp and the CUDA Python GPU Bootcamp. Presented jointly by NCI and NVIDIA, the three two-day training sessions focused on exciting and unique ways for data scientists and researchers to gain the skills needed to start quickly accelerating codes on GPUs – Graphics Processing Units. Around 150 participants from various organisations and government agencies learnt about the foundations of working with multi-GPU codes in a range of different use cases.

NCI provides researchers access to more than 600 GPUs through the Gadi supercomputer and Nirin cloud computer. All NCI users have access to these systems and can make use of the powerful technologies to accelerate and improve their scientific research codes.

The series of Bootcamps featured GPU parallelisation tools such as the RAPIDs software suite, CuPY, Numba and concepts of System Topology. Together with NVIDIA and NCI staff, the participants tackled data science and analytic pipeline challenges entirely on GPUs. The sessions contained interactive hands-on exercises with distributed training (TensorFlow, Horovord) and CUDA Python programming, running analytic pipelines and parallel GPU codes on a cluster of one of the world's leading GPU systems, NVIDIA’s DGX A100. The NCI experts used live demonstrations to show how the skills and tools acquired from the Bootcamps are directly transferable to running applications within the Gadi supercomputer's data-analysis environments.

Recent Bootcamp attendees said the following things about the sessions:

  • “Content was delivered clearly and had the right level of interaction and self-learning.”
  • “Good course pace, clear explanations from the tutor, and the breakout room was helpful.”
  • “Gave me more contextual knowledge for developing my research and code for future GPU acceleration.”
  • “Well prepared training materials, information delivery was also clear and the mentors were very helpful.”

The NCI Bootcamps are ideal for researchers who need to confirm that their self-learning has covered all angles. Looking ahead, we are planning two more Bootcamps in coming months: OpenACC and Advanced CUDA,hopefully to be jointly hosted with other supercomputer centres in the APAC region to increase availability and engage with more researchers. The total of five Bootcamps over five months of learning and practice from April to September will lay out a strong, practical foundation for researchers.

“NCI and NVIDIA have been working together to build a consolidated learning journey for researchers who are keen to solve issues in their research by making use of NCI’s infrastructure in multiple ways,” said Dr. Jingbo Wang, Training and Research Engagement Manager at NCI. The wrap-up event for this series, a highly engaging and practical Hackathon, will take place at the end of the year. Participants will be asked to express their interest, then selected teams will submit a proposal so that we match them with the best supervisors to help them during the Hackathon.

“I would love to see how users adopt what they have learned and apply it to their own research. It would be great for them to achieve something that they could not have done without the knowledge gained through these bootcamps, or that would have been prohibitive in terms of cost or technical complexity,” said Dr Wang.

Please keep an eye on NCI’s Training and Educational Events page for further announcements about future Bootcamps and the Hackathon.