Alvis System Now Available for AI/ML Research

Alvis is a new system for Artificial Intelligence and Machine Learning (AI/ML) research which is located at the Chalmers Centre for Computational Science and Engineering (C3SE). C3SE is hosted at Chalmers University of Technology in Gothenburg and is one of the six Swedish National Infrastructure for Computing (SNIC) centres for scientific and technical computing. The new Alvis system is a part of SNIC and is financed by SNIC together with the Wallenberg Artificial Intelligence, Autonomous Systems and Software Program (WASP).

Researchers from Swedish universities and research institutes may use Alvis. If you are interested in using Alvis in your research, see the instructions on the C3SE website about how to apply for access: www.c3se.chalmers.se/documentation/getting_access/ , or see the SNIC AI/ML rounds page on the SNIC User and Project Repository (SUPR) website: supr.snic.se/round . Note that, as with other SNIC resources, researchers need to be a member of a corresponding SNIC project in order to use the system.

Alvis is based on Graphical Processing Unit (GPU) accelerator cards, and the system contains several types of compute nodes, all of which have multiple NVIDIA GPUs. The system is being installed in two phases. Phase I of the project is an initial system that was installed at C3SE during the summer of 2020. Phase I of Alvis became fully operational at the end of August, after a preliminary test period. Applications for research projects to use Phase 1 of Alvis opened in mid-August 2020. SNIC has already been receiving applications for projects to use Alvis. Since Alvis is the only system within SNIC that is dedicated solely for AI/ML research, C3SE expects to see a lot more applications from AI/ML researchers wanting to use Alvis!

Phase II will actually be the major part of Alvis – this part of the system will be designed and built based on the experiences (and resultant recommendations) from operating and using Phase I of the system. The plan is that Alvis will be extended early in 2021, so Phase II is expected to be in operation by the end of summer next year.

Phase I of the Alvis system consists of a login node, a group of high-performance GPU compute nodes (Phase Ia) and a group of capacity GPU compute nodes (Phase Ib). Phase II is under discussion at the moment. It will most likely be in line with Phase I and consist of a group of capacity nodes and one (or several) groups of performance nodes, as it has been observed that dfferent fields of AI/ML research and research using AI/ML methods have varying needs in terms of the types of nodes. The specifications of the Alvis Phase I nodes are as follows.

Login node

• 4 × NVIDIA Tesla T4 GPUs with 16GB RAM

• 2 × 16 core Intel(R) Xeon(R) Gold 6226R CPU @ 2.90GHz (total 32 cores)

• 768GB DDR4 RAM

Phase Ia

12 high-performance GPU compute nodes with

• 2 × NVIDIA Tesla V100 SXM2 GPUs with 32GB RAM, connected by NVLink
• 2 × 8 core Intel(R) Xeon(R) Gold 6244 CPU @ 3.60GHz (total 16 cores)
• 768GB DDR4 RAM

5 high-performance GPU compute nodes with

• 4 × NVIDIA Tesla V100 SXM2 GPUs with 32GB RAM, connected by NVLink
• 2 × 16 core Intel(R) Xeon(R) Gold 6226R CPU @ 2.90GHz (total 32 cores)
• 768GB DDR4 RAM

Phase Ib

• 20 capacity GPU compute nodes
• 8 × NVIDIA Tesla T4 GPUs with 16GB RAM

• 2 × 16 core Intel(R) Xeon(R) Gold 6226R CPU @ 2.90GHz (total 32 cores)

• 576GB DDR4 RAM (1 node with 1536GB)

For more information about Alvis and how to use it, see the C3SE website ( www.c3se.chalmers.se/about/Alvis ) and, in particular, the section on “HPC and AI software”.