The Leibniz Supercomputer Center (LRZ), Cerebras Systems and Hewlett Packard Enterprise have announced the joint development and delivery of a new system using next-generation artificial intelligence technologies to significantly accelerate research.
The new system is funded by Bavaria through the Hightech Agenda program, which is dedicated to strengthening the technological ecosystem of the region.
The new system is expected to be delivered this summer and will be housed at LRZ, an institute of the Bavarian Academy of Sciences and Humanities (BAdW). The system will be used by local scientific and engineering communities to support a variety of research – natural language processing (NLP) and medical imaging, diagnostic and prognosis acceleration, and computational fluid dynamics (CFD) to improve understanding in areas such as aerospace engineering.
The supercomputer will be specially designed to process large data sets to solve complex research. The system consists of an HPE Superdome Flex server and a Cerebras CS-2 system, making it the first solution in Europe to use these chips.
The HPE Superdome Flex server provides a modular solution and has special capabilities for targeted processing of large amounts of memory required to process large amounts of data. In addition, the specific capabilities of the HPE Superdome Flex server (before and after data processing for AI model learning and output) are ideal for supporting the Cerebras CS-2 system, which provides deep learning performance of 100 GPUs, with easy single-node programming. Thanks to the largest processor ever created – Cerebras Wafer-Scale Engine 2 (WSE-2), which is 56 times larger than its nearest competitor, CS-2 provides more computing cores optimized for artificial intelligence, faster memory and more bandwidth than any other existing deep learning chip.
“Currently, we see that the demand for AI computing increases every three to four months with our users. With high integration of processors, memory and embedded networks on a single chip, Cerebras provides high performance and speed. This promises much greater efficiency in data processing. and thus achieve scientific discoveries faster, “says Prof. Dr. Dieter Kranzlmüller, Director of LRZ.” As an academic computing and national supercomputer center, we provide researchers with advanced and reliable IT services. To ensure optimal use of the system, we will work closely with our customers and our partners Cerebras and HPE to identify ideal community use options and help deliver innovative results. ”
AI and machine learning methods require computing power. Now the complexity of neural networks used to analyze large amounts of data will double in a matter of months. However, such programs have so far run mostly on shared or GPUs.
“We founded Cerebras to revolutionize computing,” said Andrew Feldman, CEO and co-founder of Cerebras Systems. “We are proud to be working with LRZ and HPE to give Bavarian researchers access to extremely fast artificial intelligence, enabling them to test new hypotheses, teach large-scale language models and, ultimately, advance scientific discoveries.”
Cerebras WSE-2 is 46,225 square millimeters of silicon, contains 2.6 trillion transistors and 850,000 AI-optimized computing cores, as well as evenly distributed memory that holds up to 40 gigabytes of data and has fast connections at 220 petabyte per second. This allows the WSE-2 to store all the parameters of multilayer neural networks on a single chip at runtime, which in turn reduces computation and data processing time. Today, the CS-2 system is used in a number of research institutions and enterprises in the United States and is particularly effective in image and image recognition and natural language processing (NLP). Additional efficiency is also provided by water cooling, which reduces electricity consumption.
It should be noted that the HPE Superdome Flex server contains a software package with software for creating AI procedures and models.
“We are excited to expand our collaboration with the Leibniz Supercomputer Center to provide its scientific community with next-generation computing technology,” said Justin Hotard, executive vice president and general manager of HPC & AI, HPE. – “Through our work with LRZ and Cerebras, we are pleased to support the next wave of scientific and engineering innovation in Germany. As artificial intelligence and machine learning become more widespread as we enter an era of understanding, highly optimized systems such as the new LRZ system will accelerate scientific breakthroughs for the benefit of humanity. ”