Atos, for a long time a key IT services provider in Europe, stepped onto the HPC stage in a big way in 2014 when it bought systems maker Bull’s hardware business in $840 million, instantly making it a major supercomputer vendor on the continent. In more recent years, the company has looked to expand the capabilities of its HPC systems into such areas as the cloud and its global reach beyond Europe.
As we detailed recently, its acquisitions over the last year of Nimbix, Visual BI, and Ideal GRP gave Atos much of the fuel it needed to push those ambitions, an important step as the company navigates its way into a future that could see the company splitting in two.
But as Atos looks to compete more directly with the likes of Hewlett Packard Enterprise, Dell Technologies, IBM and others in the growing global HPC market, it has found a landing spot in Brazil. For each of the past three years, Atos has delivered a supercomputer to Petrobras, the country’s state-owned national oil company that last year pulled in $84 billion in revenue.
In 2019, Petrobras launched “Fenix,” an Atos Bull system powered by Intel Xeon SP-5122 Gold processors and Nvidia Tesla V100 GPU accelerators that now sits at 161 on the Top500 list with a peak performance of 5.37 petaflops. A year later came “Atlas,” number 116 on the list with a peak performance of 8.85 petaflops and also running on Xeon SP-6240 Gold CPUs and Tesla V100 GPUs. In 2021, Atos delivered “Dragao,” ranked 60 on the list with a peak of 14.01 petaflops with Supermicro systems powered by Intel Xeon SP Gold chips and Nvidia GPUs. It’s the most powerful system in South America.
Now Atos is building a fourth supercomputer for the company, with plans to get the “Pegaso” system up and running by the end of the year. When it goes online, Pegaso – Portuguese for “Pegasus” – will deliver almost 21 petaflops of peak performance, almost as much as Atlas and Dragao combined and will bump the overall supercomputer power at Petrobras’ disposal from 42 petaflops to 63 petaflops. (However, the chart below from Petrobras indicates the goal is to increase the overall compute capabilities to 80 petaflops by the end of 2022.) The massive system will include 2,016 GPU accelerators, hold 678 terabytes of memory, and run on a 400 Gb/s network.
Pegaso, at almost 21 petaflops of peak performance, would find a place in the top 50 systems on the Top500 list, Nelson Campelo, Atos’ CEO in South America, told news site BNamericas in an interview in May. The system has yet to be benchmarked, but Campelo said that “we’re going to benchmark it, and that benchmark will certainly put it in the top 50 in the world, maybe even in the top 30.”
It took 32 trucks to transport the 30 tons of Pegaso components to the site in Rio de Janeiro, where it will be pulled together. When all the racks are filled, they will create a line that stretches out 35 meters, or more than 38 yards. Assembling Pegaso will take three months and will be followed by final adjustments and the implementation of the operating system and other software.
Atos also will offer services to help Petrobras run and maintain the system, as per the tender for the project that went out in early 2021.
The oil and gas industry ranks right up there with other sectors like government and national security, big pharma, financial services, aerospace and automotive as users of supercomputers and HPC. As we noted last year, there are a number of systems on the Top500 list that serve the petroleum industry, including the HPC5 supercomputer at number 12 (used by Eni in Italy) and Dammam-7 at number 18 and Ghawar-1 at 28 (both for Saudi Aramco).
The investments made to build up its HPC capabilities will enable Petrobras to accelerate the massive amounts of geophysical and geological data it produces to reduce risks – both geological and operational – as well as the time between when an oil field is discovered and when production can begin, according to the company.
“Expanding data processing capacity allows Petrobras to generate increasingly clearer subsurface images of areas mapped for oil and natural gas exploration and reduce the processing time of this information,” the company said in a statement. “This contributes to optimizing production, increasing the recovery factor of current reserves and maximizing the efficiency of the company’s exploratory projects.”
Accelerating the processing of geophysical data and simulations of flows in Petrobras’ reserves will drive company programs like EXP100, which was launched to reach 100 percent use of data and knowledge in exploratory projects, and PROD1000, whose goal is to reduce the time it takes to start production in a field.
The Pegaso system is the latest bit of good news for Atos. Last month the company learned that it was selected by the EuroHPC Joint Undertaking, which oversees the development of pre-exascale and exascale systems in Europe, to build the MareNostrum 5 system at the Barcelona Supercomputer Center in Spain. It likely will be based on Atos’ BullSequana XH3000 system.
Atos earlier this year previewed the XH3000 supercomputer family, noting that they can run on chips froms Intel, AMD, Nvidia and SiPearl, which is developing the microprocessor for European supercomputers.
Still, what Atos will look like by this time next year is unclear. The company’s financial numbers from last year included a 3.1 percent year-over-year decrease, to less than $11 billion, and a turnaround plan being bandied about would include splitting Atos’ business into two companies, one called Evidian responsible for its digital, big data and security units and the other – which would keep the Atos name – to include its legacy businesses, such as datacenter, digital workplace, unified communications and business process outsourcing, according to a proposal outlined in June.