Super-fast computers for AI: Torsten Hoefler awarded prestigious ACM Prize
Torsten Hoefler wins the prestigious ACM Prize in Computing for his pioneering work in high-performance computing. The fact that supercomputers have become so powerful that AI models can be trained very quickly with very large volumes of data is partly down to his research.
In brief
- The ACM Prize in computing recognises early- to mid-career computer scientists and carries a prize of USD 250,000.
- Hoefler’s research enables more efficient supercomputer networking – a key component of large-scale AI models.
- As a Professor at ETH Zurich, Hoefler helps to ensure that Switzerland remains an attractive location for innovative research in AI.
One of the world’s most important computing awards goes to an ETH Professor: Torsten Hoefler, Professor for high-performance computer systems and the Chief Architect of AI and Machine Learning at the Swiss National Supercomputing Centre (CSCS), wins the ACM Prize in Computing 2024.
Hoefler is receiving the prize for his “fundamental contributions to high-performance computing and the ongoing AI revolution”, states the Association for Computing Machinery (ACM).
The ACM Prize in computing recognises early- to mid-career computer scientists for “research contributions of fundamental importance and broad implications.” The award is endowed with a prize of USD 250,000. Hoefler is one of only a few scientists outside of the USA and the first in Continental Europe to receive this accolade.
Christian Wolfrum, Vice-President of Research at ETH Zurich, comments: “This prize is highly deserved! Hoefler’s pioneering work on high-performance computing is essential for work in the AI area around the world. The award is impressive testimony to the significance of ETH and Switzerland as a location for advanced, leading-edge AI research.”
Computing power massively accelerated
Hoefler conducts research into the performance of parallel computer systems. His work has made a key contribution towards massively speeding up high-performance computing (HPC), improving efficiency and thereby also making the costs acceptable.
While high-performance computing used to be important mainly for scientists, e.g., for weather and climate simulations, it has now grown in significance for the wider public with the current AI boom. Hoefler states: “It is a very wonderful feeling to know that things that were mostly created during my doctorate have now become so important. This prize is a huge honour!”
Specifically, Hoefler changed the way in which several computers work together on a task. When they do so, they have to be able to exchange massive amounts of data quickly and reliably. The network that links these computers must be designed to work quickly, stably, and without bottlenecks - even if it is very large.
Hoefler’s contribution consisted in, among other things, developing new network architectures, routing algorithms, and procedures for avoiding data bottlenecks which enabled this scaling in the first place.
Basis for large AI models and catalyst for science
The ACM writes that the work of Hoefler and his colleagues on scaling network design in supercomputers revolutionised their capabilities. “As a result, AI algorithms, for example, can be processed in parallel on hundreds of thousands of nodes (computers or servers).”
Hoefler is thereby also indirectly accelerating science as it relies ever more heavily on very large data volumes and machine learning in order to gain new insights.
“High-performance computing is a catalyst and a decisive factor for science; the more efficiently we can compute, the greater the potential.”Torsten Hoefler
Today, many of his innovations are incorporated in the largest, most powerful computers, including the Alps supercomputer at the Swiss National Supercomputing Centre (CSCS) in Lugano, where Hoefler is the Chief Architect for AI and Machine Learning.
Biographical background
Torsten Hoefler is a Professor of Computer Science at ETH Zurich (the Swiss Federal Institute of Technology), where he has served as Director of the Scalable Parallel Computing Laboratory since 2012. Before coming to ETH Zurich, he ran the high-performance modelling and simulation for the first petascale supercomputer “Blue Waters” at the University of Illinois in Urbana Champaign. Hoefler has received numerous awards for his work, most recently the Max Planck Humboldt Research Medal and the Jack Dongarra Early Career Award. He is a Fellow of IEEE, ACM and a member of the European Academy of Sciences (Academia Europaea). He was also awarded the Latsis Prize from ETH Zurich in 2015, the ACM Gordon Bell Prize in 2019 and received two grants from the European Research Council (ERC). As part of the Swiss AI initiative, Hoefler is also helping to develop Swiss language and base models which are designed to serve both science and business in Switzerland.
Comments
No comments yet