The Swedish National Infrastructure for Computing (SNIC) is investing in a new high-performance computing (HPC) system for Swedish researchers. The new flagship system is an HPE Cray EX supercomputer, called Dardel, that is being installed at PDC during 2021.
Dardel is a new supercomputer that will replace Beskow, which is the current flagship system at PDC that is being retired later this year. The new Dardel system will have several partitions, including
- a large CPU partition (that will be suitable for a broad range of computational applications), and
- a partition based on graphics processing units (GPUs), which is intended for the most computationally demanding applications.
Dardel will be installed in two phases. Installation of the CPU partition (providing about 65,000 CPU cores) and the storage system will start in April this year. This first phase of the system will be available for research from July 2021. The GPU partition will be installed later this year in the autumn, and will be available to researchers from January 2022.
Academic researchers may apply for allocations to use Dardel through SNIC . (The next round of SNIC applications for large allocations includes Dardel and is opening on the 1st of March - for more information, see snic.se/allocations/compute/large-allocations .) Business and industrial researchers should contact PDC directly to discuss possibilities as there are several options, such as applying for free allocations for research projects via the PRACE SHAPE programme, or purchasing CPU hours from KTH.
Each node will be a dual AMD EPYC™ 2.25 GHz 64 core processor.
There will be several sizes of compute nodes in the CPU partition.
- 488 X 256 GB (SNIC thin nodes)
- 20 X 512 GB (SNIC large nodes)
- 8 X 1024 GB (SNIC huge nodes)
- 2 X 2048 GB (SNIC giant nodes)
- 36 X 256 GB (KTH industry/business research nodes)
This gives a total of 554 nodes.
This partition will be equipped with GPU nodes, each of which will be configured with:
- one AMD EPYC™ processor with 64 cores (which is currently under development)
- 512 GB memory
- four AMD Instinct™ GPUs (which are currently being developed)
- HPE Slingshot (ethernet-based) using Dragonfly topology
- 100 Gb/s for the first phase, which will be upgraded to 200 Gb/s in the second phase
The expected LINPACK performance of the first phase of the system is above 2 Petaflops. The overall peak performance of the whole system is expected to increase to about 13.5 Petaflops when the second phase is operational.
- available for academic research via SNIC and PRACE
- available for business and industrial research projects
- operation currently funded till Q2 2026