Credit: Supplied
New Zealand’s most powerful supercomputer for artificial intelligence applications has been installed at the University of Waikato by Nvidia partner Fujitsu.
The Nvidia DGX A100 is the first computer of its kind in New Zealand and the world’s most advanced system for powering universal AI workloads, the University said.
It can rapidly and efficiently process massive amounts of data, enabling machine learning and artificial intelligence that can solve complex problems.
Machine learning uses algorithms to explore huge data sets and create models that can be trained to recognise patterns, facial expressions and spoken words -- or to find anomalies such as credit card fraud.
One of the first projects the computer is being used for is to train models that can learn and classify New Zealand’s plants and animals, based on a publicly available database of more than one million photos.
Credit: Supplied The massive computing engine fits into one quarter of a computing rack in the University’s main server room, powered by processors designed specifically to accelerate the unique needs of AI workloads.
Nvidia's Mellanox InfiniBand networking ensures data is rapidly supplied to the system.
The DGX A100 has eight A100 GPUs containing 40GB of memory each for a total of 320 GB of GPU memory.
“AI is a powerful tool that enables researchers to achieve scientific breakthroughs and discoveries on areas such as climate change and biodiversity, which are critically important to New Zealand and the world,” said Sudarshan Ramachandran, country manager of Nvidia.
Professor Albert Bifet said students and researchers could take months, or even years, to process the data needed to create models like the one they are working on if they had to use more traditional computing.
“This computer will allow our researchers to process that data in a matter of days. It will enable them to gain insights and progress their research at an unprecedented scale," he said.
The purchase was made possible through income from the sale of commercial licenses to the Weka software, a suite of Java-based software tools for machine learning and data mining that the machine learning group at the University has been developing for more than 20 years.
“Being able to use the funds from Weka, which has proved so successful, is a real win for us," said professor Bifet.
"Weka software has been bought by several large international IT companies. It shows the success and depth of expertise we have here and has enabled us to reinvest back into our group.”
