Condor Galaxy 1 (CG-1), the first supercomputer IA of this network, is characterized by its 4 exaFLOPs and 54 million core processors. Cerebras and G42 plan to deploy two more supercomputers of this type, CG-2024 and CG-2, in the United States in early 3. With a total programmed capacity of 36 exaFLOPs, this unparalleled computer network will revolutionize the advancement of AI around the world.
“It has been incredibly exciting to partner with Cerebras to rapidly deliver the training supercomputer of AI fastest in the world and lay the foundation for the interconnection of a constellation of these supercomputers around the world. This partnership combines Cerebras' impressive computing capabilities with G42's multi-sector expertise in AI. According to the shared vision of G42 and Cerebras, Condor Galaxy will be used to address society's most pressing problems in health, energy, climate action and beyond,” said Talal Alkaissi, CEO of G42 Cloud, a subsidiary of G42.
Located in Santa Clara, California, CG-1 brings together 64 Cerebras CS-2 systems by linking them into a single supercomputer IA easy to use, with a training capacity IA from 4 exaFLOPs. Cerebras and G42 propose CG-1 as a cloud service that allows customers to obtain the benefits guaranteed by the performance of a supercomputer IA without the need to manage or deploy models on physical systems.
With CG-1, Cerebras for the first time collaborates not only in the creation of a dedicated AI supercomputer, but also in its management and operation. CG-1 was designed to enable G42 and its cloud customers to quickly and easily train large innovative models, thereby speeding innovation. The strategic alliance between Cerebras and G42 already has state-of-the-art advanced AI models in bilingual Arabic chat, health and climate studies.
“Providing 4 exaFLOPs of processing IA at FP 16, CG-1 dramatically reduces AI training time while eliminating distributed computing fatigue,” said Andrew Feldman, CEO of Cerebras Systems. “Many cloud companies have announced large GPU clusters that cost billions of dollars to build, but are very difficult to use. Deploying a single model to thousands of tiny GPUs takes months of the time of dozens of highly skilled people. CG-1 solves this problem; setting up a generative AI model takes minutes, not months, and can be done by a single person. CG-1 is the first of three 4-exaFLOP AI supercomputers to be deployed in the United States. Together with G42, we plan to expand this implementation and build incredible AI processing capabilities from 36 efficient, purpose-built exaFLOPs over the next year.”
G42, leader in the AI and in cloud computing headquartered in the United Arab Emirates (UAE), it is at the forefront of large-scale digital transformation initiatives globally. The UAE was the first nation to appoint a Ministry ofArtificial intelligence, with subsequent major investments such as the creation of G42's research partner, the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), the world's first fully focused postgraduate university on AI.
Training large models requires massive computational volumes, large datasets, and specific AI expertise; the collaboration between G42 and Cerebras meets all three of these requirements.
With the Condor Galaxy supercomputer network, these two companies are democratizing AI by offering simple, seamless access to the industry-leading AI computer. G42 works on diverse datasets from health, energy, and climate studies to enable system users to train new and advanced fundamental models.
These models and their derivative applications represent a powerful force for good. Finally, Cerebras and G42 bring together a team of hardware and data engineers, AI scientists, and industry specialists for a comprehensive AI offering aimed at solving customer problems. It's a combination that will produce game-changing results and speed hundreds of AI projects around the world to the max.
Optimized for large language models and generative AI, CG-1 features 4 16-bit exaFLOPs for AI computation, with standard support for up to 600 billion parameter models and optional support for up to 100 trillion parameter models via extensible configurations. With 54 million AI-optimized compute cores, 388 terabits per second fabric bandwidth, and 72.704 AMD EPYC processor cores—unlike any GPU cluster—CG-1 delivers near-linear performance scaling from 1 to 64 CS-2 systems using simple data parallelism.
“AMD is committed to accelerating AI with cutting-edge high-performance computing processors and adaptive computing products, as well as through partnerships with innovative companies like Cerebras that share our vision of pervasive AI,” said Forrest Norrod, executive vice president and general manager of AMD's Data Center Solutions Business Group. "Based on more than 70.000 AMD EPYC processor cores, Cerebras' Condor Galaxy 1 will make ample computing resources available to researchers and enterprises committed to advancing AI."
CG-1 offers native support for training with extended sequence lengths and up to 50.000 tokens out of the box, with no need for special software libraries. Programming CG-1 is done entirely without complex distributed programming languages, so even the largest models can run without spending weeks or months distributing tasks across thousands of GPUs.
Located in Colovore, a high-performance colocation facility in Santa Clara, California, CG-1 is operated by Cerebras in compliance with US regulations, ensuring that state-of-the-art AI systems are not used by adversary states. Each Cerebras CS-2 system is designed, packaged, manufactured, tested and integrated in the United States; Cerebras is the only manufacturer to package processors and produce AI systems in the United States.
CG-1 is the first of three 4 exaFLOP AI supercomputers (CG-1, CG-2 and CG-3) built and installed in the United States as a collaboration between Cerebras and G42. These three AI supercomputers will be interconnected into a 12-exaFLOP, 162 million-core distributed AI supercomputer comprised of 192 Cerebras CS-2s and powered by more than 218.000 high-performance AMD EPYC CPU cores. G42 and Cerebras plan to bring six more Condor Galaxy supercomputers online in 2024, bringing total processing power to 36 exaFLOPs.
The Condor Galaxy, also known as NGC 6872, spans 522.000 light-years from tip to tip, making it 5 times larger than the Milky Way. Distant 212 million light-years from Earth, it is visible in the skies of the southern hemisphere as part of the constellation Peacock.
Cerebras Systems is a team made up of innovative IT architecture specialists, computer scientists, researchers deep learning and engineers of all kinds. Our goal is to build a new class of computing systems designed for the sole purpose of accelerating the work of generative AI. Our flagship product, the CS-2 system, based on the world's most powerful and fastest AI processor, simplifies training large models and avoids the complexity of distributed computing. Cerebras solutions are available in the cloud, via Cerebras AI Model Studio, or on-premise.
G42 is a global leader in building visionary AI capabilities for a better tomorrow. Founded in Abu Dhabi and active around the world, it champions AI as a powerful force for good. G42 teams continually reimagine the possibilities of technology, applying advanced thinking and innovation to accelerate progress and address society's most pressing problems. G42 works with nations, companies and people to create the infrastructure for the world of tomorrow. From molecular biology to space exploration, and every field in between, G42 realizes exponential possibilities, starting today.
BlogInnovazione.it
An ophthalmoplasty operation using the Apple Vision Pro commercial viewer was performed at the Catania Polyclinic…
Developing fine motor skills through coloring prepares children for more complex skills like writing. To color…
The naval sector is a true global economic power, which has navigated towards a 150 billion market...
Last Monday, the Financial Times announced a deal with OpenAI. FT licenses its world-class journalism…
