The fastest supercomputer (HPC) in the Czech Republic, which will cost more than 400 million crowns, will be from the American company Hewlett Packard Enterprise (HPE) and will run on processors from AMD and Nvidia graphics cards. Tender in the national supercomputer center IT4Innovations in Ostrava, according to information from Lupa, it has already been completed and should have been approved by the European Union, which participates in co-financing.
Next year, a supercomputer will be put into operation in Ostrava, which will become part of a European network of similar machines associated under the company. EuroHPC. Through these activities, the European Union wants, among other things, to catch up with the United States and China and become more technologically independent. Supercomputers are crucial for demanding scientific or industrial calculations and simulations. Therefore, as part of a supercomputer project a European HPC processor using ARM and RISC-V technologies is created.
In the first phase, 1.5 billion euros, ie almost 41 billion crowns, will flow into the EuroHPC. The goal is to get HPCs with petascale, pre-exascale, and in the future also exascale performance (at least one exaFLOPS, which is ten to eighteenth floating point operations per second). The first supercomputers, including the Czech one, will start operating in 2021. The program will then be expanded, with the addition of counts on investments of eight billion euros. There is also talk of acquiring an HPC with ten exaFLOPS.
One of the most powerful in Europe
The Ostrava supercomputer falls into the petascale category. Today, it would be the sixth most powerful machine in Europe and the twenty-first in the world. The Czechia has a budget of 14.8 million euros, with about a third paid by the European Union and the rest coming from the state budget, specifically from the Ministry of Education. In IT4Innovations, new HPCs would be procured anyway as part of a new generation renewal, but membership in EuroHPC allows more money to be used.
“I can confirm that HPE is becoming the supplier of the latest system to be located in IT4Innovations. The 15.2 petaFLOPS system will be the most powerful supercomputer in the Czech Republic and at the same time one of the most powerful in Europe. The contract is currently being inspected by EuroHPC JU, which is one of the parties and at the same time the provider of 35 percent of the cost of acquiring the entire system. The commissioning of the supercomputer with the working name EURO_IT4I is planned in the first half of 2021, “confirmed a spokesperson for IT4Innovations Zuzana Červenková.
According to Lupa, HPE won over the French company Atos, which became a major player in HPC thanks to the previous acquisition of Bull. In Ostrava, after all, since last year running the second and weaker Barbora supercomputer with a power of 826 teraFLOPS. The balance of the two suppliers is thus maintained in IT4Innovations.
The tender in Ostrava was prestigious on the market. This is partly due to the fact that it is part of the European network of supercomputers and at the same time it was one of the few local really large HPC projects. Both companies, including other suppliers such as Huawei, IBM or Lenovo, are competing in other EuroHPC tenders. The Czechia is one of the first EuroHPC countries to complete the competition. Machines have already been selected in Luxembourg a Slovenia. Timeline of the European project is available here.
Another success for AMD
According to Lupa, HPE also managed to win thanks to the fact that Intel was in the process of intergenerational renewal of its processors at the time of the tender, and AMD thus hit the ideal window. HPE does not want to comment on the trade or the exact parameters of the machine before the official signing of the contract. The contract is to be signed on October 15.
IT4Innovations already provides some technical data. The supercomputer will have 720 servers (3.8 petaFLOP / s), the acceleration part will count 70 servers with eight graphics cards in each (11 petaFLOP / s for standard HPC simulations and up to 150 petaFLOP / s for artificial intelligence calculations), part for processing large data files will have up to 24 TB of shared memory, 36 servers will be in operation to provide cloud services (131 teraFLOP / s), 200Gb network and storage of one PB with a speed of storage of one TB / s.
The Ostrava supercomputer will use processors from AMD, but not graphics cards. These will be supplied by Nvidia. The tender stated that the graphics must support the CUDA technology, which Nvidia is behind.