Alvis is an old Nordic name meaning "all-wise". An appropriate name, one might think, for a computer resource dedicated to research in artificial intelligence and machine learning. The first phase of Alvis has been used at Chalmers and by Swedish researchers for a year and a half, but now the computer system is fully developed and ready to solve more and larger research tasks.
Alvis is a national computer resource within the Swedish National Infrastructure for Computing, SNIC, and started on a small scale in the autumn of 2020, when the first version began being used by Swedish researchers. Since then, a lot has happened behind the scenes, both in terms of use and expansion, and now it's time for Chalmers to give Swedish research in AI and machine learning access to the full-scale expanded resource. The digital inauguration will take place on February 25, 2022.
What can Alvis contribute to, then? The purpose is twofold. On the one hand, one addresses the target group who research and develop methods in machine learning, and on the other hand, the target group who use machine learning to solve research problems in basically any field. Anyone who needs to improve their mathematical calculations and models can take advantage of Alvis' services through SNIC's application system – regardless of the research field.
"Simply put, Alvis works with pattern recognition, according to the same principle that your mobile uses to recognize your face. What you do, is present very large amounts of data to Alvis and let the system work. The task for the machines is to react to patterns - long before a human eye can do so," says Mikael Öhman, system manager at Chalmers e-commons.
How can Alvis help Swedish research?
Thomas Svedberg is project manager for the construction of Alvis:
"I would say that there are two parts to that answer. We have researchers who are already doing machine learning, and they get a powerful resource that helps them analyse large complex problems.
But we also have those who are curious about machine learning and who want to know more about how they can work with it within their field. It is perhaps for them that we can make the biggest difference when we now can offer quick access to a system that allows them to learn more and build up their knowledge."
Facts
Alvis, which is part of the national e-infrastructure SNIC, is located at Chalmers. Chalmers e-commons manages the resource, and applications to use Alvis are handled by the Swedish National Allocations Committee, SNAC. Alvis is financed by the Knut and Alice Wallenberg Foundation with SEK 70 million, and the operation is financed by SNIC. The computer system is supplied by Lenovo. Within Chalmers e-commons, there is also a group of research engineers with a focus on AI, machine learning and data management. Among other things, they have the task of providing support to Chalmers’ researchers in the use of Alvis.
Voices about Alvis:
Lars Nordström, director of SNIC:
- "Alvis will be a key resource for Swedish AI-based research and is a valuable complement to SNIC's other resources."
Sara Mazur, Director of Strategic Research, Knut and Alice Wallenberg Foundation:
- "A high-performing national computation and storage resource for AI and machine learning is a prerequisite for researchers at Swedish universities to be able to be successful in international competition in the field. It is an area that is developing extremely quickly and which will have a major impact on societal development, therefore it is important that Sweden both has the required infrastructure and researchers who can develop this field of research. It also enables a transfer of knowledge to Swedish industry."
Philipp Schlatter, Professor, Chairman of SNIC's allocation committee Swedish National Allocations Committee, SNAC:
- "Calculation time for Alvis phase 2 is now available for all Swedish researchers, also for the large projects that we distribute via SNAC. We were all hesitant when GPU-accelerated systems were introduced a couple of years ago, but we as researchers have learned to relate to this development, not least through special libraries for machine learning, such as Tensorflow, which runs super fast on such systems. Therefore, we are especially happy to now have Alvis in SNIC's computer landscape so that we can also cover this increasing need for GPU-based computer time."
Scott Tease, Vice President and General Manager of Lenovo’s High Performance Computing (HPC) and Artificial Intelligence (AI) business:
- “Lenovo is grateful to be selected by Chalmers University of Technology for the Alvis project. Alvis will power cutting-edge research across diverse areas from Material Science to Energy, from Health care to Nano and beyond. Alvis is truly unique, built on the premise of different architectures for different workloads.
- Alvis leverages Lenovo’s NeptuneTM liquid cooling technologies to deliver unparalleled compute efficiency. Chalmers has chosen to implement multiple, different Lenovo ThinkSystem servers to deliver the right NVIDIA GPU to their users, but in a way that prioritizes energy savings and workload balance, instead of just throwing more underutilized GPUs into the mix. Using our ThinkSystem SD650-N V2 to deliver the power of NVIDIA A100 Tensor Core GPUs with highly efficient direct water cooling, and our ThinkSystem SR670 V2 for NVIDIA A40 and T4 GPUs, combined with a high-speed storage infrastructure, Chalmers users have over 260,000 processing cores and over 800 TFLOPS of compute power to drive a faster time to answer in their research.”
Contact
Sverker Holmgren
- Research Professor, E-commons, Physics
Thomas Svedberg
- Researcher, E-commons, Physics
Mikael Öhman
- Systems Technician, E-commons, Physics