Original title: Europe released the most powerful AI supercomputer LEONARDO, the pattern of supercomputing system will change

Lei Feng net(official account: Lei Feng net)Leonardo, unveiled this week at the CINCA Research Center in Italy, uses 14000 Nvidia GPU,Nvidia to call it "the most powerful AI system in the world", which provides 10 exaflops of AI performance in some semi-precision floating-point (FP16) applications and about 200 petaflops of computing power when tested against Top500. The launch of this system will also change the pattern of European supercomputing system from the end of this year.

For those who have been engaged in supercomputing for several years, they should be very familiar with Bull systems, especially in Europe. The company was acquired by Atos in early 2014, but from an outside point of view, the nature remained the same until HPC products, including the "Sequana" series, used the name Bull.

For many years, Bull has been a well-known but small-scale HPC integrator in Europe, with several major achievements and only a relatively small number of systems in research and academia. But times are changing, especially after Bull's important acquisition.

The first was the sale of IBM's x86 server business to Lenovo because of IBM's key position in the high-performance computing market. European HPC centres also have no choice because Lenovo is so far away. IBM has enough native resources to be used in many centers, so over the years, many European research institutions, government agencies, meteorological agencies, and even large enterprises have used IBM products for their HPC.

Cray, another company, hopes to win many victories in Europe, a trend that continued to grow until the recent acquisition of HPE, especially in the weather and large computing devices. Although we have not yet seen the impact of this supplier shift on the market, we expect changes in Europe to show up in the next few years. This is not because Cray or IBM are inappropriate, or because their technology has changed, but what Europe has always wanted-its own ecosystem of HPC suppliers.

As an independent company, Bull can't be like today's CINECA in Italy, which we'll discuss later. But with the support of this system, and many other systems that will enter the supercomputing Top 500 next year, Atos is an absolutely noteworthy HPC system, at least in Europe.

In terms of system share in the supercomputer Top 500 rankings, Bull / Atos did not show much growth even if it was acquired. One might think that a larger parent company could help Bull secure a deal, but this is not the case. Bull maintained 3.6 per cent of the system share from 2012 to 2014, jumping to 4 per cent in 2016 and slightly to 4.4 per cent in 2018.

According to the latest figures for this year, they have risen to 5.2%, with 26 overcalculators on the list, the largest of which is the French CEA system, which ranks 20th. Add that without CEA, Bull as a supercomputing company might not exist, and CEA needs a local French company to achieve its highest value national priority.

Overall, the share of the system will change, starting with the global top 500 list to be released in November.

It seems that Atos is building a "Baby Summit" supercomputer. The so-called "high density" refers to a GPU-intensive system similar to the Oak Ridge National Laboratory in the United States. The difference is that instead of using IBM Power-based processors, they avoid Power altogether, choosing only Intel and AMD (which is not surprising). But then again, when it comes to systems like the upcoming CINECA, floating-point functionality is provided almost entirely by Nvidia's Ampere A100 GPU.

In Europe, Atos has developed with supercomputers equipped with GPU. Last week, they announced that they would introduce their Bull Sequana XH2000 super to CSC in Finland. Before that, they announced two large systems, including the "largest AI supercomputer" in Sweden and the UK.

In addition to large systems in Sweden and the UK, they have also strengthened their partnership with Nvidia to provide (possibly customized) SuperPOD computers to other universities. Moreover, there are signs that they are preparing to split the Cray / HPE and IBM markets in the lucrative and vital HPC market (weather forecast). Bull also signed agreements with ECMWF this year for four systems (a total of 27 petaflops) and new systems, which is the meteorological agency's AI and quantum research center.

Today, Atos is in the most powerful position in the history of Europe, which shows that it can provide systems that are as powerful as Cray / HPE and IBM around the world for multidisciplinary scientific and technological computing. In fact, some might argue that today's announcement of the "LEONARDO" supercomputer will bring Atos into a truly "big era" in the world of high-performance computing.

It's hard to say where the LEONARDO supercomputer will rank when it enters the Fortune 500, but it is expected to achieve peak semi-precision floating-point performance of 200 petaflops (Nvdia calls it the top 10 EB AI supercomputer, which is based on FP16).

In a conference call with Nvidia's Ian Buck, the GPU maker said it would be the most powerful AI supercomputer in Europe, with more than 14000 A100 GPU and decomposed into a single network with Intel Corp Rapids processors and Nvidia Mellanox HDR 200Gb / s on four A100 GPU on the Atos Sequana node.

LEONARDO is procured by EuroHPC, a joint partnership between European governments and the European Union to develop a world-class supercomputing ecosystem and 10 billion supercomputing in Europe, funded by the European Commission through Italian universities and research departments.

Marc Hamilton, Vice President of Architecture and Engineering, NVIDIA Solutions, said: "Europe's 10 billion euro EuroHPC technology roadmap opens the door to rapid growth and innovation for HPC and AI. We are working with CINECA and Atos to accelerate scientific exploration in a wide range of applications, providing a platform for the 10-billion-level computing era. "

Interestingly, how many HPC hubs will choose SuperPOD for AI-centric computing next year, and how many will adopt the more standard HPC workload-oriented architecture provided in Sequana? We want to be similar to what we see elsewhere in the AI / HPC world, where AI and accelerated scientific computing applications are GPU-intensive, and there are some AI-focused R & D centers that require a full DGX or SuperPOD.

Nvidia and partners build multiple SuperPOD-based systems across Europe while maintaining their Sequana beyond hardware. The software and supporting ecosystems of these systems are so large that profits may be negligible in rebuilding SuperPOD, and their composition in terms of supporting and strengthening ecosystems will be priceless, especially since Europe's role is more ambitious than ever before.

Original link: https://www.nextplatform.com/2020/10/15/with-another-key-supercomputer-win-atos-looks-stronger-than-ever/Lei Feng net

The original article of Lei Feng net is forbidden to reprint without authorization. For details, please see the reprint instructions.