Data & Analytics

Petrobras’ Atlas Supercomputer Joins List of World’s Largest

Atlas and Fênix, also from Petrobras, are the two largest supercomputers in Latin America.

supercomputer.jpg

About 1.5 million smartphones or 40,000 latest generation laptops—the processing capacity of Petrobras’ Atlas supercomputer is equivalent to these devices, combined. In operation since April at the Center for Information Processing and Treatment in Rio de Janeiro, Atlas occupies the 57th position in processing capacity of the Top500.org list, released on 22 June. The supercomputer is the best ranked in Latin America in the world ranking of high-performance computers (HPCs).

In the previous Top 500 list, released in November 2019, the title of the largest supercomputer in Latin America was Santos Dumont, which returned to the top position after its processing capacity was multiplied by five with investments. Before that, in the list released in June last year, the supercomputer Fênix, also from Petrobras, occupied that position. It underwent an update this year and now ranks as the second most powerful in Latin America, at number 83 on the list.

Byte Oceans

Atlas and Fênix are part of a team of several supercomputers that perform different functions in the company. The two HPCs are responsible for processing geophysical data generated during exploration and development activities for gas and oil production. Together, they have the processing capacity equivalent to 2.5 million smartphones or 67,000 new laptops.

Vast amounts of processing power are needed to create the representative images of geology below the seabed. The seismic images, fundamental to the findings on oil and gas, cover hundreds of square kilometers and reach thousands of meters deep. That is why the algorithms that process them involve complex mathematical equations, with an immense volume of data, generating images that geologists and geophysicists can interpret. The volume of data related to a single seismic project may reach tens of terabytes. The investment in HPCs reduces the costs of this type of processing.

Data Tsunamis

Since 2018, Petrobras has reinforced its investment in HPCs. This process has been occurring in three waves. The first, last year, resulted in the acquisition of Fênix. This year, the company also started to count on Atlas and Guaricema, the latter being dedicated to the simulation of data generated from oil and gas reservoirs. With the completion of the third wave in 2021, the company will have 10 times more HPC processing capacity than it had in 2018.

This investment resulted in a significant reduction in the processing time of geophysical data in the geological areas of interest to the company and the application of complex, state-of-the-art algorithms.

Nicolas Simone, Petrobras’ director of digital transformation and innovation, said the processing capacity of these computers allows the development of strategic programs for Petrobras, accelerating the increase in efficiency in the company.

“The intensive use of artificial intelligence and large computational capacities guarantee an expansion of the level of data processing that would be impossible to achieve without these technologies,” he said.

Technical Specifications

Atlas

  • Number of servers: 136
  • Mathematical Accelerator (GPU): Nvidia V100 / 32 GB / SXM2
  • Accelerators per server: 8
  • Processor: Intel Xeon Gold 6240
  • Processors per server: 2
  • Cores per processor: 18
  • Velocity: 2.6 GHz
  • Memory per server: 768 GB
  • Internal network: InfiniBand EDR 100 gbps
  • Operational system: CentOS 7.7
  • Theoretical performance: 8.0 PFLOPS Rpeak DP
  • Measured performance: 4.4 PFLOPS Rmax (or 4,376 TFLOPS Rmax)
  • Consumption (at maximum capacity): 54 6KW

Fênix

  • Number of servers: 360
  • Mathematical Accelerator (GPU): Nvidia V100 / 32GB
  • Accelerators per server: 2
  • Processor: Intel Xeon Gold 5122
  • Processors per server: 2
  • Cores per processor: 4
  • Velocity: 3.6 GHz
  • Memory per server: 192 GB
  • Internal network: InfiniBand EDR 100 gbps
  • Operational system: CentOS 7.6
  • Theoretical performance: 5,372 TFLOPS Rpeak DP (or 5.4 PFLOPS Rpeak DP)
  • Measured performance: 3,161 TFLOPS Rmax (or 3.2 PFLOPS Rmax)
  • Consumption (at maximum capacity): 390 KW