What is the most powerful supercomputer in the country, which is among the 100 best in the world, and what is it for?

The national government announced in mid-December the purchase of a supercomputer that ranks among the 100 most powerful in the world. This is what it is known as “High Performance Computing“(HPC extension), or high performance computing, and has multiple uses at the federal level.

“A high-performance computing system is a system designed for tasks that an ordinary computer cannot perform, either because they are very large or because they would take many hours to complete,” explains Pablo Minnini, coordinator of the National High Performance Computing System (SNCAD) of the Ministry of Science and Technology.

The equipment, installed in the Data Center of the National Meteorological Service, will be financed by the Ministry of Defense at a cost of 265 million pesos for its installation. But to get to this there was a competition that was won by Intel and that allowed the country take a billionaire teamat a reduced price.

However, as Minnini says, “it is not enough just to have the equipment”: high performance computing also implies “the necessary infrastructure and human resources with the knowledge to maintain these systems and to be able to use them: high performance computing is the set of very large computers, the data centers that host them and human resources trained to use them”.

To give you an idea, last year the Province of Córdoba launched Serafín, a supercomputer equivalent to 300 desktop PCs (on average), with a computing power of 156 trillion operations per second. However, the new team 15390 TeraFLOP, or more than 15 thousand trillion.

​Here Minnini talks about the new equipment, which is even more powerful, what it is for and what it means for Argentina in terms of computational advance.

─Why is HPC called “supercomputer”?

─We use equipment that can be thought of as many computers together, but have special technologies to connect to each other and work efficiently. Additionally, these computers often use specific technologies in processors, accelerators (usually called GPUs, because they are derived from the graphics processing units that our computers use to generate 3D graphics), and cooling methods years later they reach our computers or the commercial datacenter.

─What are these devices for?

─In the world these teams are used to solve scientific problems, industrial problems, the oil industry, for defense problems, for weather forecasting and for many other things.

─What does it mean to be Top 100?

─It means this machine will be among the 100 largest in the world, in a list where countries report their largest computers for public use, defense and private use. The estimate is that today this system, due to its size, it would be in 82nd place. The final location will depend on the measurements that will be made when the equipment is installed in a few months.

─What specific uses will it have in Argentina?

─It has multiple uses: training in artificial intelligence problems, solving complex physics and chemistry problems, drug design, bioinformatics and genomics studies. These computers are also used for inverse problems that need to be solved during oil exploration. And for the weather forecast.

─What would this weather forecast app look like?

─The National Meteorological Service (SMN) already has a 370 TFLOPS equipment that is used for forecasting, and which has significantly improved its quality by being able to make more “copies” of the forecast every 3 hours, and thus be able to better calculate the probability that different phenomena occur in the atmosphere.

The new team: a computer beast

─What calculation capacity does the new equipment have?

─The equipment has a computing capacity of 15390 TeraFLOP. A FLOP is a floating point operation (that is, an addition, subtraction, or multiplication of two decimal numbers), and FLOP is a measure of operations per second. Therefore, 15390 TeraFLOP means this computer can work 15,390 billion addition, subtraction or multiplication in every second.

─How did the tender go and why did you decide to work with Intel?

─System vendor is Lenovo, using Intel processors and accelerators (GPUs), along with technologies from Lenovo and other brands. An international public tender has been carried out e 3 companies who have industry experience and equipment in the TOP 500 (the list of the 500 largest supercomputers in the world).

─What did you take into account in the tender?

─The economic offer was considered, but also the quality of the equipment, the calculation capacity, the energy efficiency, the after-sales service, and the results of the efficiency tests that each bidder had to carry out. Lenovo was the winner for submitting the best offer in most points evaluated. The offer of Lenovo uses processors and accelerators (GPU) brand new from Intel, which are not yet on the market, which has allowed them to offer a very large computing capacity.

─How much did it cost?

─The computer cost 5 million dollars, and this price includes 3 years of maintenance by the company that supplies the equipment. It is the largest investment to date in the High Performance Computing area in Argentina, but it is also the lowest price paid per FLOP (that is, for computing power) from our country for a scientific calculation system.

─What processors does it use?

─The computer has processors and accelerators. The processors are Intel Sapphire Rapids from HBM, which have very fast memory integrated into the processor, with technology similar to that used by video cards used for video games. This allows them to access the data needed to perform any type of calculation very quickly, and they are the first processors available with this type of memory. Accelerators (which provide most of the computational power) are Intel Ponte Vecchio. Right now they are the fastest GPUs out there. In the US, Argonne National Laboratory is buying a supercomputer with the same technology as Intel, albeit in a smaller version. different supplier and with a much larger size.

─What is this equivalent to, in terms of mass consumer equipment?

─For comparison, a Playstation 5 has a computing capacity of 9.2 TeraFLOPs and a top-of-the-line MacBook Pro laptop has a computing capacity of 5.3 TeraFLOPs. This computer is equivalent to 1672 Playstation 5s, up to 2903 MacBook Pros or more than 5000 average laptops.

─And yet, it is much more than that.

─Of course, it’s not enough to stack 1672 Playstations or 5000 laptops in a room, because you also need to connect them efficiently, have a storage system (i.e. hard drives) for all the data this system can generate, and do it efficiently from an energy point of view. While 1672 Playstation 5 needs 585 KW to run, this system has a peak power consumption of 233 kW.

The place of HPC in Argentina

─How is the day with a team like this? How many teams work on it and what do they do, specifically?

─Users connect to the computer in the form at a distance. It’s not much different than connecting to a portal that allows you to see what resources are available and which ones you can use. And they submit their jobs to the system which has a queue that runs jobs as they arrive, similar to waiting to be seen at a bank or doctor’s office. This way, at any time of the day there can be many groups working simultaneously with the team, on very different problems. And the equipment is on and available 24 hours a day, 7 days a week.

To access the team, projects will be requested, which will be evaluated by a jury of experts, who will decide how many calculation hours each project can use.

─What does having this team mean for Argentina?

─Argentina had a significant deficit in its scientific computing capacity. Brazil has several computers in the TOP 500. Mexico had them (although today it has none on that list), and Chile is working to increase its computing capacity. In the case of Brazil, many of the supercomputers are used to search for oil deposits. In Argentina having these tools is essential, for example, to have a forecast weather for the whole country. This has a economic impact important, both for agricultural production and for warning systems against natural disasters. The SMN acquired a fully dedicated forecasting team 4 years ago, which has had a huge impact. And a team that is open to use across the scientific community, technology sector and industry will have an even greater impact.

─What impact does it have on the local knowledge industry?

─The acquisition of this equipment is also part of a plan, which has assessed the country’s needs, and which provides for the training of human resources and collaboration with various players. A system of computer centers for scientific use has existed in the country for more than a decade. In 2019, the science and technology system has identified the need to have a team larger than the existing ones, and a strategic plan was written. This assessment found that installed computing capacity had to be multiplied by at least 10 to meet the country’s needs. The newly purchased equipment multiplies the capacity of the SMN equipment by 40 and by 100 that of the largest scientific equipment available today in Argentina. It will have a great impact in science and technology, allowing new problems to be solved, and in the NMS where it will serve experiment with new technologies and train human resources.

SL