Summit Meets TaihuLight at ISC18 HPC Connection Workshop
On June 26th 2:00-6:00PM, the 19th HPC Connection Workshop will be held in Matterhorn 2, Movenpick Hotel Frankfurt City, Germany during the 2018 International Supercomputing Conference (ISC18). Experts from companies and institutions including IBM, China’s National Supercomputing Center in Wuxi, Tachyum, Atos-Bull, Shell International E&P Corporation, UberCloud, Shanghai Supercomputer Center and Inspur will discuss the latest progress and challenges of HPC. Participants will also explore the design and super-large scale applications of advanced supercomputing technologies including Summit and Sunway TaihuLight, the new BXI interconnection network, and the AI chip developed by Tachyum.
Today, the latest Top 500 Global Supercomputer List officially released at the ISC18. The International supercomputing Summit, the world’s first AI-based supercomputer, at the Oak Ridge National Laboratory is named the world’s fastest supercomputer, with peak performance reaching 200,000 trillion times per second (200 PFLOPS) Designed and built by IBM, Summit features 9,216 IBM Power9 processors and 27,648 Nvidia Volta GPU accelerators. At the HPC Connection Workshop, Fernando Pizzano from IBM CSM Development Department will explain in depth the architecture design and typical application of Summit, and reveal how scientists are using in-depth learning technology to study important research topics in the fields of cosmic astronomy, chemistry and biology using Summit.
Although Summit is the top spot on the Top 500 Global Supercomputer List, Sunway TaihuLight remains one of the fastest supercomputers in the world ranked No.2 right after Summit. Also the HPC Connection Workshop, Fu Haohuan, from the National Supercomputing Center in Wuxi, will deliver a presentation on the 2017 “Gordon Bell Prize” application completed by “Sunway TaihuLight” — a nonlinear earthquake simulation with high scalability. The software maximizes storage capacity, computing resources and other advantages of processors developed by China, to realize nonlinear earthquake simulations up to 18.9 PFlops. It marks the first time in the world that high-resolution and high-frequency nonlinear plastic earthquake simulations have been realized on such a large scale. The software enables, for the first time, high-resolution accurate simulation of the occurrence process of Tangshan earthquake (M7.8, 1976), which allows the scientists to better understand the impact of the Tangshan earthquake and provides important reference for future research on earthquake prevention and prediction.
If AI is the hottest technology today, AI chips are without a doubt the hottest AI product. Rado Danilak, the founder of Tachyum, will demonstrate an AI chip developed by Tachyum, the Tachyum Prodigy. The chip features 64 cores in the grid layout of 8 × 8, and can provide 72 PCIe 5.0 consistency support channels for local cluster interconnect. In deep-learning neural network training and reasoning, general AI, symbol AI, biology AI, big data, memory data analysis and most other large scale workloads, the chip achieves 131 Tensor TFLOPS/s.
The founder of numerous companies, Rado is the co-founder and chief technology officer of SandForce, acquired by LSI in 2011, and the founder and chief executive officer of Skyera, an ultra-dense solid-state storage system supplier acquired by WD in 2014. He is also an architect of the chipset and GPU in NVIDIA, an architect of CPU in Nishan Systems and Toshiba, the chief architect of the 64b × 86 CPU in Gizmo Tech, member of the IDC Technical Computing Advisory Panel, and writer for the Forbes Technology Council and TechTarget.
Maurice Araya Polyo, computer scientist from Shell International E&P Corporation, will share how to use machine-learning to better study exploration geophysics. He will share his experience in applying machine-learning at Shell, including a new method for fault detection and speed model establishment, which provided a new way to detect patterns in unprocessed seismic data. Mauricio Araya Polyo currently plays a leading role in multiple fields, including seismic imaging, modeling and compressed sensing, and machine-learning. Prior to joining Shell, he was a member of the famous Kaleidoscope Project at the Barcelona Supercomputing Center.
Jean-Pierre Panziera from Atos-Bull will explore the interpretations of the new internet technology BXI (Bull eXascale Interconnect), developed by Bull and designed and optimized for large-scale HPC workloads. Based on the Portals 4, BXI allows complete uninstallation of communication primitives in hardware, realizing an independent process of computation and communication. Panziera will discuss BXI ASIC modules, network interfaces and switchboards, BXI software environment, and how BXI is integrated to build a massively parallel system.
The HPC Connection Workshop co-sponsored by the Asia Supercomputer Community is held annually during the ISC Supercomputing Conference in Germany, the SC Supercomputing Conference in the US and the ASC Student Supercomputer Challenge, which aims to explore the application innovation and development trend of supercomputing in frontier science and engineering and emerging fields, promoting exchanges and cooperation between China and the world in supercomputing.