Cluster for cosmological data
20. 3. 2023
20. 3. 2023
The project Central European Institute for Cosmology and Fundamental Physics – CEICO is an institution established within the Institute of Physics of the CAS. CEICO is an international research group focused on the interplay between cosmology, gravity and string theory. CEICO was established in November 2016 with the financial support of the European Structural and Investment Funds and Czech state funds from the Ministry of Education, Youth and Sports, the European Research Council and the Grant Agency of the Czech Republic.
For the Institute of Physics of the CAS, resp. for the CEICO project, we delivered a high-performance computing cluster with a GPU – this became part of the HPC (high performance computing) system “Phoebe”, which will serve the international team of scientists from the CEICO project from summer 2022.
This HPC system interprets tens of terabytes of cosmological data from the European Space Agency’s Planck satellite. Phoebe also stores and analyses data from the telescope operated by members of the lab, and this data is securely provided to the wider scientific community after cleaning, thanks to its robust storage.
The most demanding class of tasks running on the system are the so-called. N-body simulations – these allow the study of the evolution of galaxies, their dynamics and generally provide the possibility of verifying models explaining the origin of the Universe and our galaxy.
An interesting task, using the powerful NVIDIA A100 graphics card, is the implementation of machine learning algorithms – a neural network, trained on data collected by generations of astronomers, learns to recognize different space structures on Phoebe – thus speeding up the processing of the ever-increasing amount of data from telescopes on the Earth’s surface or in orbit.
Challenging matrix computations and direct communication between processes are made possible by the low latency Infiniband computing network, advanced AMD EPYC™ processors and NVIDIA A100 accelerators.
equipped with CPU AMD Epyc generation Milan
(each NODe 64 physical cores and 512GB of RAM)
each has 64 physical CPU cores, 2TB of RAM and 8 NVidia A100 80GB SXM4 GPUs
that has the entire Compute Node cluster
these are NVIDIA A100 cards, which thus consist of 110,592 CUDA cores, 6912 Tensor cores, total GPU memory of 1280GB
total performance of all GPUs in FP64
each with powerful AMD CPUs with a base frequency of 3.5GHz and a total of 32 physical cores
AMD EPYC™ 7003 Series processors are designed to keep your applications running smoothly. EPYC™ processors run virtually all x86 applications, enabling seamless migration and seamless integration into existing x86 infrastructures. These processors are suitable for virtualization, hybrid cloud or software-defined infrastructure.
AMD processors, together with AMD 3D V-Cache™ technology, are the most powerful x86 server processors for technical computing – designed to accelerate product development and increase productivity. These are the first server processors to use true 3D die stacking using a copper die design with no outputs. This enables 200× higher interconnect density compared to current 2D technologies and more than 15× higher interconnect density compared to other 3D technologies using solder outputs.
Interested in more information about AMD EPYC™ processors? See if an AMD solution is right for your infrastructure.