[ad_1]
Condor Galaxy 1 will be used to solve society’s most pressing challenges in healthcare, energy, climate action and more
read more…
Abu Dhabi-based technology holding group G42 and California-based artificial intelligence chip startup Cerebras Systems have launched the world’s largest supercomputer for artificial intelligence training.
Condor Galaxy is a network of nine interconnected supercomputers that provides a new AI computing method that is expected to significantly reduce AI model training time.
The network’s first AI supercomputer, Condor Galaxy 1 (CG-1), has 4 exaFLOPs (a measure of computer speed equivalent to 1 trillion floating points per second) and 54 million cores. Two more such supercomputers, CG-2 and CG-3, are expected to be deployed in the United States early next year. This unprecedented supercomputing network has a planned capacity of 36 exaFLOPs and will completely change the progress of global artificial intelligence.
Talal Alkaissi, CEO of G42 Cloud, a subsidiary of G42, pointed out that Condor Galaxy will be used to solve society’s most pressing challenges in health care, energy, climate action, etc.
“It is extremely exciting to work with Cerebras to rapidly deliver the world’s fastest artificial intelligence training supercomputer and lay the foundation for the interconnection of these supercomputer clusters around the world. This collaboration combines Cerebras’ extraordinary computing power with G42’s Multi-industry AI expertise comes together,” Alkaissi said.
Cerebras’ flagship CS-2 system is powered by the world’s largest and fastest AI processor, making training large models simple and easy by avoiding the complexity of distributed computing. Located in Santa Clara, California, CG-1 connects 64 CS-2 systems into an easy-to-use AI supercomputer with an AI training capacity of 4 exaFLOPs. Cerebras and G42 offer CG-1 as a cloud service, allowing customers to enjoy the performance of AI supercomputers without having to manage or distribute models on physical systems.
Accelerate innovation
CG-1 is Cerebras’ first collaboration to build, manage and operate a dedicated artificial intelligence supercomputer. CG-1 is designed to allow G42 and its cloud customers to quickly and easily train large, groundbreaking models. The Cerebras-G42 strategic partnership is already advancing state-of-the-art artificial intelligence models in Arabic bilingual chat, healthcare and climate research.
Cerebras Systems CEO Andrew Feldman said: “CG-1 provides 4 exaFLOPs of AI computing on FP 16, greatly shortening AI training time while eliminating the pain of distributed computing.”
“Many cloud companies have announced massive GPU clusters that cost billions of dollars to build but are extremely difficult to use. Distributing a single model across thousands of tiny GPUs requires dozens of people with rare expertise. Months. CG-1 eliminates this challenge. Setting up a generative AI model takes minutes, not months, and can be done by a single person. CG-1 is one of three 4 exaFLOP AI supercomputers deployed in the U.S. Next year, we plan to expand this deployment with G42 and support up to 36 exaFLOPs of efficient, dedicated AI computing.”
G42’s collaboration with diverse datasets in healthcare, energy and climate research will enable system users to train new cutting-edge underlying models. This collaboration brings together a team of hardware engineers, data engineers, artificial intelligence scientists and industry experts to provide a full range of artificial intelligence products to solve customer problems. This combination will produce breakthrough results and drive the development of hundreds of artificial intelligence projects around the world.
CG-1 Explanation
CG-1 is optimized for large language models and generative AI, providing 4 exaFLOPs of 16-bit AI computing power, standard support for up to 600 billion parameter models, and expandable configuration to support up to 100 trillion parametric model. Unlike any known GPU cluster, CG-1 features 54 million AI-optimized computing cores, 388 terabits per second of fabric bandwidth, and is powered by 72,704 Advanced Micro Devices (AMD) EPYC processor cores. Provides near-linear performance scaling from 1 to 64 CS-2 systems using simple data parallelism.
“Powered by more than 70,000 AMD EPYC processor cores, Cerebras’ Condor Galaxy 1 will provide researchers and enterprises with massive computing resources available to drive AI,” said Forrest Norrod, executive vice president and general manager of AMD’s Data Center Solutions Business Unit. forward.”
CG-1 provides native support for training with long sequence lengths, up to 50,000 tokens, out of the box without the need for any special software libraries. CG-1 is programmed without the need for complex distributed programming languages, meaning even the largest models can be run without spending weeks or months distributing work across thousands of GPUs.
[ad_2]
Source link