Powered by Epyc Processors and AMD Instinct GPUs, producing 552 petaflop/s!
Taken from TPU … AMD has won a contract to empower the LUMI supercomputer, designed for the EuroHPC Joint Undertaking (EuroHPC JU) in conjunction with 10 European countries. The contract will see AMD provide both the CPU and GPU innards of the LUMI, set to be populated with next-generation AMD Epyc CPUs and AMD Instinct GPUs. The supercomputer, which is set to enter operation come next year, will deliver an estimated 552 petaflop/s – higher than the world’s current fastest supercomputer, Fugaku in Japan, which reaches peak performance of 513 petaflop/s – and is an Arm-powered affair.
The contract for LUMI’s construction has been won by Hewlett Packard Enterprise (HPE), which will be providing an HPE Cray EX supercomputer powered by the aforementioned AMD hardware. LUMI has an investment cost set at 200 million euros, for both hardware, installation, and the foreseeable lifetime of its operation. This design win by AMD marks another big contract for the company, which was all but absent from the supercomputing space until launch, and subsequent iterations, of its Zen architecture and latest generations of Instinct HPC accelerators.
Supercomputers enable the fight against pandemics and help resolve unforeseen research questions
LUMI’s top-notch computing resources are needed in leading-edge research in a wide range of data- and computing-intensive fields. Examples include climate, pharmaceutical, and artificial intelligence.
LUMI will also have a fast-track for urgent computing in time- and mission-critical simulations. This kind of simulation might be, for example, related to a large epidemic or pandemic disease. The current COVID-19 pandemic has largely benefitted from supercomputers: supercomputers have been used for example to simulate studies related to vaccine research and defeat the spread of the virus. With its vast computing resources, LUMI can address different research challenges even faster than before. In addition, it will enable addressing totally new types of scientific challenges combining multidisciplinary research and artificial intelligence.
World-class environmental sustainability and energy-efficiency
As a carbon-neutral data center, LUMI helps the European ICT sector in becoming greener and more cost-efficient, which is a necessity for reaching EU’s ambitious climate targets and paving the way for the green transition. CSC’s data center in Kajaani is among the world’s most eco-efficient: it uses 100% renewable energy produced with hydropower. LUMI’s waste heat will be used in Kajaani’s district heating network: 20% of the area’s yearly district heating needs will be covered with LUMI’s waste heat.
LUMI system architecture explained:
- The LUMI system will be supplied by Hewlett Packard Enterprise (HPE), based on an HPE Cray EX supercomputer.
- The peak performance of LUMI is an astonishing 552 petaflop/s meaning 552 *1015 floating point operations per second. This figure makes LUMI one of the world’s fastest supercomputers. For comparison, the world’s fastest computer today (Fugaku in Japan) reaches 513 petaflop/s and the second fastest (Summit in the US) 200 petaflop/s (more information: www.top500.org). If LUMI’s computing power was compared to normal laptops, it would require 1.5 million laptops together to reach the performance of LUMI. If these laptops were piled up, they would form a tower of over 23 kilometers high!
- LUMI will also be one of the most advanced platforms in the world for artificial intelligence (AI). With LUMI, it will be possible to combine AI, especially deep learning, and traditional large scale simulations combined with massive scale data analytics in solving one research problem.
- The number crunching capability of LUMI is accelerated by the GPU (Graphics Processing Unit) partition. It is based on the future generation AMD Instinct GPU.
- LUMI will be complemented by a CPU (Central Processing Unit) partition, featuring 64-core next-generation AMD EPYC CPUs.
- LUMI’s data analytics partition has 32 aggregated terabytes of memory and 64 visualization GPUs. This partition is used e.g. for visualization, heavy data analysis, meshing, and pre/post-processing.
- LUMI’s storage system will consist of three components. First, there will be a 7-petabyte partition of ultra-fast flash storage, combined with a more traditional 80-petabyte capacity storage, both based on the Lustre parallel filesystem, as well as a data management service, based on Ceph and being 30 petabytes in volume.
- In total, LUMI will have astounding storage of 117 petabytes and an impressive aggregated I/O bandwidth of 2 terabytes per second
- LUMI will also have an OpenShift/Kubernetes container cloud platform for running microservices.
- All the different compute and storage partitions are connected to the very fast Cray Slingshot interconnect of 200 Gbit/s. The global bandwidth of the LUMI-GPU partition is 160 TB/s. The global Internet traffic would fit therein, in fact, two times!
- LUMI takes over 150m2 of space, which is about the size of a tennis court. The weight of the system is nearly 150 000 kilograms (150 metric tons).