ISTA advances academic AI research with new NVIDIA GPUs

The Austrian Institute of Science and Technology (ISTA) is investing in a state-of-the-art computing cluster with more than 100 NVIDIA H100 Tensor Core graphics processors (GPUs) to expand its machine learning computing infrastructure in academic research. The new GPU cluster specializes in training deep learning algorithms for generative AI. The multi-million dollar investment by the institute will help accelerate academic AI research on a large scale and strengthen ISTA as a European computing hotspot, the institute announced in Klosterneuburg today.

In the era of AI and generative AI, AI and machine learning research is making huge strides, primarily led by the private sector. With a €3.6 million investment in NVIDIA H100 Tensor Core graphics processors (GPUs), the Austrian Institute of Science and Technology (ISTA) is increasing its computing power for training deep learning algorithms: so-called “large language models” (LLMs) are used in generative AI and machine learning in academic research.

“With this comprehensive upgrade, ISTA is expanding its computing power and can thus compete with major international research institutions at the forefront of academic AI research,” says Martin Hetzer, President of ISTA.

“The installation of the new NVIDIA GPU cluster at ISTA provides powerful computing capabilities for LLM training and advanced research,” said John Josephakis, global vice president of business development for high-performance computing (HPC) and supercomputing at NVIDIA. “The cluster will help drive breakthroughs, for example, in the life sciences, including structural biology and neuroscience, to advance scientific knowledge and education.”

ISTA will continue to invest in the Group to maintain it to high international standards.

Modern high-performance computing cluster with dedicated graphics processors and optimized CPUs

The NVIDIA H100 GPU nodes are part of a larger update to the ISTA High Performance Computing (HPC) cluster. The upgraded infrastructure consists of two new types of NVIDIA GPU nodes with complementary capabilities in addition to updating existing CPUs. The first new GPU node type consists of 13 nodes, each containing eight NVIDIA HGX H100 systems. Together, the H100 nodes achieve a peak computing power of 416 petaflops. With more than 100 powerful GPUs, this part of the cluster is designed to handle complex AI tasks: for example, training LLMs similar to ChatGPT or applications like the AlphaFold AI program for predicting protein structures. The second new GPU node type consists of 12 nodes, each containing four NVIDIA L40S GPUs, and is intended for graphics applications that require training smaller models. These applications include structural biology calculations and image analysis, for example in cryogenic electron microscopy. Part of this second expansion is the replacement of legacy GPU systems, with over 150 previous-generation NVIDIA GPUs continuing to operate.

In addition to the two new GPU node types, the new CPUs will provide 147 million core hours per year (one hour of CPU operation equals one core hour). The 147 million core hours per year figure is derived from usage data from recent years, additional needs of existing research groups, and projected needs for new professors in the coming years. ISTA currently has over 80 active research groups and is expected to grow to 150 by 2036.

Developing AI research through collaboration between academic research and the private sector

Recent developments in the fields of AI and machine learning have shown that size plays a crucial role. “Currently, training large AI models is primarily the domain of tech giants. With this high-performance computing cluster, ISTA enables its researchers to experiment with cutting-edge AI technology and develop it in potentially new and unexpected directions. This contributes to “this initiative will help to better anchor AI and machine learning research in publicly funded research and promote open, curiosity-driven science, which is one of ISTA’s key strengths,” says Professor Dan Allistare, an AI and machine learning specialist at ISTA, who also highlights the need to “democratize” AI. The Allistare group is exploring ways to create more powerful but small versions of large AI models that people can use on everyday devices, such as a phone or PC. “By modernizing the computing infrastructure and targeting it specifically for AI and machine learning research, ISTA is demonstrating its collaboration with the private sector,” adds Hetzer. “This is an important mission to ensure that academic research helps shape the future of AI.”

This cluster, a project initiated by ISTA professors and strongly influenced by Alistarh, brings far-reaching progress to the Institute and its various research areas. In addition to the possibility of training numerous high-performance LLM holders, the cluster will also be used for research in the life sciences, for example in structural biology and neuroscience. The new computing infrastructure is currently designed to be used exclusively by ISTA scientists and staff; however, there is potential to expand the service to external users in the future.

The extensive modernization funded by the Institute is an ongoing process to maintain a high international standard. In this way, ISTA creates and maintains a competitive computing infrastructure for the next generation of researchers.

Minister welcomes ISTA initiative

The Austrian Federal Minister for Education, Science and Research, Martin Polaszek, praised the institute’s initiative: “I welcome ISTA’s decision to continue investing in the future of AI. ISTA is already part of the Research Excellence Cluster recently presented by the FWF duo. Bringing together the best minds and providing them with modern research infrastructure will help take AI research in Austria to the next level. I wish ISTA, its AI researchers and their collaboration partners: good luck with your innovative projects.”

Rückfragehinweis:
Andreas Rothe
[email protected]
+43 664 8832 6510