Finland’s LUMI supercomputer has hit a brand new milestone, efficiently finishing the pilot section of its GPU partition that extends the processing energy of the system.
LUMI is the quickest supercomputer in Europe and the third quickest globally, in keeping with the Top500 list printed in November 2022. It was inaugurated in June final yr, however the system’s GPU partition had not absolutely been put in on the time.
Now a group of Norwegian researchers have accomplished the GPU pilot section for LUMI, testing out the capabilities of the system with the GPU partition (LUMI-G) put in and operational.
The testing comprised three pilot tasks protecting photo voltaic atmospheric modelling, pure language processing (NLP) and imaging of seismic information, in keeping with the LUMI group.
For the photo voltaic atmospheric modeling venture, researchers on the Rosseland Centre for Photo voltaic Physics on the College of Oslo work with simulations of the Solar’s environment, and examine these with observations from photo voltaic telescopes to be taught extra concerning the workings of the Solar and different stars.
This entails in-house purposes referred to as Bitfrost and Dispatch, that are coded to be extremely parallel, in order that they use many processor cores to carry out calculations sooner. The analysis group is engaged on adapting the code to make use of GPUs.
The NLP venture on the Division for Informatics on the College of Oslo covers analysis into automated translation, human-computer interplay through spoken language, and content material suggestions and contextual promoting. This requires superior deep studying fashions, the coaching of which requires huge computing energy.
In the meantime, the seismic information venture on the Division of Geoscience and Petroleum on the Norwegian College of Science and Know-how (NTNU) goals to get detailed pictures of Earth’s inside utilizing seismic information recorded on land and within the sea.
To extend the extent of element, a brand new modeling approach is getting used the place seismic information is simulated utilizing a complete numerical mannequin. The simulations are repeated for a lot of sources, and the massive variety of calculations concerned requires the assets out there with LUMI-G.
LUMI, put in on the IT Heart for Science (CSC) datacenter in Kajaani, Finland, is a pre-exascale system provisioned underneath the auspices of the EuroHPC Joint Enterprise.
It’s primarily based on HPE’s Cray EX structure, just like the Frontier exascale system at Oak Ridge Nationwide Lab within the US. The CPU-only partition contains 1,536 dual-socket CPU nodes, every that includes AMD “Milan” Epyc processors and between 256GB and 1,024GB reminiscence. The LUMI-G GPU partition has 2,560 nodes, every that includes a single customized AMD “Trento” Epyc chip and 4 AMD MI250X GPUs.
Andrey Kutuzov, a member of the Language Know-how group utilizing LUMI, stated that onboarding was “comparatively straightforward for these already aware of HPC environments,” and that usually, the code designed for Nvidia GPUs was discovered to run flawlessly on LUMI’s AMD GPU methods.
Helmi improve
In November final yr, LUMI was additionally upgraded with the addition of a quantum accelerator from the VTT Technical Analysis Centre of Finland. Often known as Helmi, the quantum processor contains 5 qubits and is being touted as a quantum accelerator or co-processor to work alongside LUMI’s classical CPU nodes.
The EuroHPC JU has introduced a call for tenders to pick a vendor for JUPITER, which is meant to be the primary European exascale system. This shall be put in on the campus of the Forschungszentrum Jülich research institution in Germany, however shall be owned by the EuroHPC JU and operated by the Jülich Supercomputing Centre. The cut-off date for tenders is February 13.
The EuroHPC JU will conclude a contract with the profitable tenderer to be able to purchase, ship, set up and preserve this supercomputer system. It can fund half the full price of the brand new system, with the opposite half coming from the German Federal Ministry of Schooling and Analysis (BMBF) and the Ministry of Tradition and Science of the State of North Rhine-Westphalia (MKW NRW). ®
Source link