"Supercomputer" digest: 4 news from the world of high-performance computing
Today we decided to prepare for you an overview of the latest news about supercomputers. Under the cut, we’ll tell you about the most powerful industrial supercomputer and the most powerful supercomputer in Germany, as well as how scientists from the Los Alamos National Laboratory saved $ 2 million when building an HPC system.
/ photo by NASA CC
The American IT company CSRA and the organization Edge Solutions & Consulting have increased the performance of the Discover supercomputer installed at the NASA Climate Modeling Center. A new unit with a capacity of 1.56 petaflops was added to the machine, which raised the overall system performance to 5 petaflops.
')
Discover “built” of separate hardware modules called SCU, or Scalable Compute Units. The new unit for Discover was named SCU14, and it is based on 528 Supermicro FatTwin server nodes. Each node has two Intel Xeon Gold 6148 (Skylake) processors with twenty cores and Intel 4600 solid-state drives.
The architecture of the FatTwin SuperServer F619P2-RT system is made in frame size 4U and contains 8 nodes. Each node has 12 slots for memory cards and 6 SSD-drives. The cluster uses Intel Omni-Path 100G ToR-switches with 48 ports.
NASA believes that the fastest processors, switches, and an impressive amount of memory per core (4.8 GB) will enable the SCU14 module to be the best solution for weather and climate modeling.
Previously, the Discover system could build climate models with a spatial resolution of 1.5 km. SCU14 will allow to achieve a resolution of 1 km or less, which will provide an opportunity to more accurately predict the weather. You can see how Discover models the physical processes in the Earth’s atmosphere at this link .
On January 18 of this year, Eni launched a new HPC4 supercomputer at its Green Data Center located in Ferrera-Erbognon. And it is the most powerful industrial supercomputer in the world.
The maximum performance of HPC4 reaches 18.6 petaflops, and in combination with the already existing HPC3, the total Eni capacity has increased to 22.4 petaflops.
The new cluster runs on 1600 HPE ProLiant DL380 nodes, each of which is equipped with two 24-core Intel Skylake processors (76,800 cores in total) and two NVIDIA Tesla P100 GPUs connected by InfiniBand technology. HPC4 works with a 15 petabyte storage subsystem.
The HPC4 supercomputer will be used by Eni to build 3D seismic images, simulate petroleum systems and industrial plants. According to Claudio Descalzi, CEO of the company, improving supercomputer infrastructure is an important part of the Eni digital transformation process.
/ photo Vitalij Fleganov CC
On 22 January 2018, the French IT company Atos signed a contract to create a high-performance Bull Sequana X1000 system for the Juelich research center in Germany. The machine with a capacity of 12 petaflops will be the most powerful German supercomputer.
Bull Sequana will find application in the field of computer modeling and will be used to process unstructured data. In addition, the supercomputer will use the Human Brain Project to simulate the workings of the human brain in order to better understand the processes occurring in the human head.
The new machine will replace JUQUEEN - a supercomputer that has long been considered the most powerful in Europe. JUQUEEN will “retire” in the spring of 2018, and the installation of its successor, Bull Sequana, is scheduled for the second quarter of 2018.
One of the “fastest” supercomputers in the United States — the Trinity project — was completed last summer and began working normally two months ago.
Trinity is a supercomputer on the Cray XC40 architecture with a capacity of 41.5 petaflops. It runs on 19,000 nodes that use two processor architectures: Intel Xeon Haswell and Intel Xeon Phi Knights Landing. Trinity is equipped with a water cooling system that consumes 10 MW of electricity.
During the construction of this machine and after the approval of the system plan, the team at the Los Alamos National Laboratory (LANL) decided to change the wiring plan for the power supply of the supercomputer racks. Engineers decided to use new aluminum cables instead of copper, which reduced the cost of materials by 20% and reduced the weight of the whole structure by three times. In total, the laboratory saved $ 2 million in materials and labor costs. The project will be used to ensure US national security.
PS Other materials from our blog on Habré:
PPS A couple of fresh materials from our corporate blog:
/ photo by NASA CC
The upgrade of the supercomputer Discover
The American IT company CSRA and the organization Edge Solutions & Consulting have increased the performance of the Discover supercomputer installed at the NASA Climate Modeling Center. A new unit with a capacity of 1.56 petaflops was added to the machine, which raised the overall system performance to 5 petaflops.
')
Discover “built” of separate hardware modules called SCU, or Scalable Compute Units. The new unit for Discover was named SCU14, and it is based on 528 Supermicro FatTwin server nodes. Each node has two Intel Xeon Gold 6148 (Skylake) processors with twenty cores and Intel 4600 solid-state drives.
The architecture of the FatTwin SuperServer F619P2-RT system is made in frame size 4U and contains 8 nodes. Each node has 12 slots for memory cards and 6 SSD-drives. The cluster uses Intel Omni-Path 100G ToR-switches with 48 ports.
NASA believes that the fastest processors, switches, and an impressive amount of memory per core (4.8 GB) will enable the SCU14 module to be the best solution for weather and climate modeling.
Previously, the Discover system could build climate models with a spatial resolution of 1.5 km. SCU14 will allow to achieve a resolution of 1 km or less, which will provide an opportunity to more accurately predict the weather. You can see how Discover models the physical processes in the Earth’s atmosphere at this link .
The most powerful industrial supercomputer launched
On January 18 of this year, Eni launched a new HPC4 supercomputer at its Green Data Center located in Ferrera-Erbognon. And it is the most powerful industrial supercomputer in the world.
The maximum performance of HPC4 reaches 18.6 petaflops, and in combination with the already existing HPC3, the total Eni capacity has increased to 22.4 petaflops.
The new cluster runs on 1600 HPE ProLiant DL380 nodes, each of which is equipped with two 24-core Intel Skylake processors (76,800 cores in total) and two NVIDIA Tesla P100 GPUs connected by InfiniBand technology. HPC4 works with a 15 petabyte storage subsystem.
The HPC4 supercomputer will be used by Eni to build 3D seismic images, simulate petroleum systems and industrial plants. According to Claudio Descalzi, CEO of the company, improving supercomputer infrastructure is an important part of the Eni digital transformation process.
/ photo Vitalij Fleganov CC
The creation of the most powerful supercomputer in Germany began
On 22 January 2018, the French IT company Atos signed a contract to create a high-performance Bull Sequana X1000 system for the Juelich research center in Germany. The machine with a capacity of 12 petaflops will be the most powerful German supercomputer.
Bull Sequana will find application in the field of computer modeling and will be used to process unstructured data. In addition, the supercomputer will use the Human Brain Project to simulate the workings of the human brain in order to better understand the processes occurring in the human head.
The new machine will replace JUQUEEN - a supercomputer that has long been considered the most powerful in Europe. JUQUEEN will “retire” in the spring of 2018, and the installation of its successor, Bull Sequana, is scheduled for the second quarter of 2018.
Trinity supercomputer started working
One of the “fastest” supercomputers in the United States — the Trinity project — was completed last summer and began working normally two months ago.
Trinity is a supercomputer on the Cray XC40 architecture with a capacity of 41.5 petaflops. It runs on 19,000 nodes that use two processor architectures: Intel Xeon Haswell and Intel Xeon Phi Knights Landing. Trinity is equipped with a water cooling system that consumes 10 MW of electricity.
During the construction of this machine and after the approval of the system plan, the team at the Los Alamos National Laboratory (LANL) decided to change the wiring plan for the power supply of the supercomputer racks. Engineers decided to use new aluminum cables instead of copper, which reduced the cost of materials by 20% and reduced the weight of the whole structure by three times. In total, the laboratory saved $ 2 million in materials and labor costs. The project will be used to ensure US national security.
PS Other materials from our blog on Habré:
- Step to quantum supremacy: Intel's 49-qubit quantum computer
- “Course on hyperscale”: there are almost 400 hyper-scalable data centers in the world
PPS A couple of fresh materials from our corporate blog:
Source: https://habr.com/ru/post/347674/
All Articles