HPC facilities

What is High-Performance Computing?

High-performance Computing (HPC) is the use of specially-designed hardware systems (and associated software) that allow computational operations to be undertaken at higher speed than is possible with "typical" computers.
Typically, HPC systems consist of many individual compute nodes that are connected together in such a way as to make the overall system much faster at computational jobs than any of the individual nodes alone would be. Thus, the combined nodes (dozens, to hundreds or even thousands in number) can work togther to undertake tasks that would take a single node a prohibitively long time.
This relies on not only a large number of compute nodes- the way they pass data between eachother, the way they access common data, and the way the software that is run makes use of the many compute nodes, is of vital importance. So, while an individual node may be no faster than a good contemporary laptop or desktop machine, it is the overall capability of the entire system (multiple compute nodes, their interconnection, data and memory access, and optimised software) that gives the increase in performance.
Such systems are quite complex in terms of hardware, management software and maintenance, and they also require purpose-built rooms with air-conditioning, mains power conditioning and similar non-trivial support infrastructure.
However, without the sort of number-crunching capacity enabled by facilities such as eResearch SA's HPC systems, many of the problems encountered in today's research areas would not be able to be investigated.

How much faster will my code run on your supercomputers?

Details coming!

How can I use your HPC facilities?

Details coming!

Will you install or upgrade software package X for me?

Commercially-licensed (or paid for) software can also be installed on our systems by eRSA staff. However it is up to the user to pay for a valid license for the product. Please note that we will not install software on our systems that in any way violates a usage or licensing policy.
Upgrading of software is done based on the resources and effort required. Upgrade licenses for software are the responsibility of the user.
Custom-written user code can be installed and compiled on our systems; please note that we can only provide limited support for custom-written user code.

Can I run my own code on your systems?

Details coming!

Will you help me develop/debug my code?

Details coming!

Can I run Windows code on your facilities?

We make use of the Linux Operating System on our HPC Facilities and as such cannot support native Windows code.

What are the limits on using your HPC facilities?

We do not place any hard limits on HPC usage, however our HPC facilities are a shared resource and as such we ask users keep their usage to a reasonable level.

What is considered "unreasonable or excessive" usage on your HPC facilities?

- Submission of more than 50 jobs to the queue in a single run - Requesting large amounts of wall time on a large number of queued jobs (Submitting 50 jobs to the queue and requesting 300 hours wall time per job)

Which supercomputer should I use

This depends heavily on the type of jobs you are running and what the application you are using supports. It is best to contact the help desk if you need help selecting the best supercomputer for your job. As a rule of thumb:

  • Aquila is best suited for sequential jobs that are difficult to parallelize and require large amounts of shared memory (160 Processors, 160GB RAM).
  • Hydra is best suited for running jobs that are sequential in nature and only require a single nodes worth of resources (less than 2GB RAM , Up to 2 processors).
  • Corvus is best suited to run large MPI based jobs that can run in parallel across multiple nodes. (MPI jobs up to 544 processors and 544GB RAM).

How long will it take my job to start?

Job start times vary based on a number of criteria. These include

  • the number of jobs in the queue
  • the walltime of the queued jobs
  • the number of jobs a user has run in the last week
  • available system resources.

During busy times you may have to wait 2 or 3 days for your jobs to start.
If you think your job has been queued for an excessively long time, contact the Service Desk and let us know.

What about GPUs?

Details coming!

What packages are available?

There are two ways to find this information.

  1. Refer to the software map here. Whilst this page is updated regularly, there can be some delay between a package being installed on one of the eRSA HPC systems, and it being listed here, so this reference is not to be regarded as necessarily being completely up-to-date.
  2. When logged-in to one of the eRSA systems, use the module avail command for an up-to-date list of installed modules on that system. Note that this must be done on each system individually. Refer here to help decide which system is most applicable to the type of job you're considering running.