New capabilities for Machine Learning and GPU pricing updates

We’re excited to announce an addition of new GPU capabilities to our platform and some noteworthy changes to resource pricing as a result.

New Graphics Processing Units (GPUs)

We’ve installed eight NVIDIA A100 GPU cards into the Mahuika HPC system, providing a significant boost in computing performance and an environment particularly suited to machine learning workloads. Over the last few months we’ve worked directly with a group of beta tester researchers to ensure this new capability is fit-for-purpose and tuned to communities' specific software and tool requirements.

These new A100s, alongside software optimised for data science), are available to researchers using machine learning approaches. If this is you, Contact our Support Team to discuss how these new resources could support your work.

Reduced pricing for P100s

We’ve recently reviewed our pricing and reduced the price of our existing P100 GPUs to 7.0 compute units per device-hour. The P100 GPUs are available to any project with a Mahuika allocation so if you have an existing allocation on Mahuika, you can access the P100s right away.

If you need a larger or new allocation on Mahuika, you can apply for access now, but requests will likely be considered as part of our next allocation call window: 31 August - 01 October.

For more technical information about using GPUs on NeSI, click here. If you have questions about allocations or how to access the P100s, Contact our Support Team.

Sharing our learning along the way

If you’re curious about what it takes to get the best of the A100 cards, you can learn about our experiences in the first post of a new ‘Tech Insights' blog series: Tech Insights: A behind-the-scenes look at rolling out new GPU resources for NZ researchers.

Our inaugural post discusses our first tasks with the A100s: thermal and internal software tests. In future posts, we’ll explore user tests we conducted in the spaces of deep learning and molecular dynamics codes, as well as take a closer look at which codes are suitable to run on GPUs and whether your research project is a fit.

Future GPU investments

Looking ahead, we’re currently finalising another investment into additional GPU cards later this year. We’re also exploring the A100s' Multi-Instance GPU (MIG) features, which partition the cards into several independent instances, giving multiple users access to GPU acceleration at the same time.

These activities will enable us to expand GPU access to an even broader community of users, as well as support more advanced and demanding performance needs across domains. If you’re interested in using these A100s for something other than machine learning, let us know by Contact our Support Team that way we can keep you up to date on our plans.

If you have questions or comments on anything mentioned above, please Contact our Support Team

Thank you,

NeSI Support