“G42”, the specialized technology holding company in the UAE, and “Cerebras Systems”, the leading company in accelerating artificial intelligence solutions in the UAE, have partnered to develop and accelerate technology solutions in the field of artificial intelligence. The collaboration resulted in launching the “Condor Galaxy” network, which aims to enhance AI computation by utilizing nine supercomputers.
This network introduces a new methodology that contributes to reducing the training time of AI models. The launch of “Condor Galaxy” represents a revolution in the field of AI and contributes to addressing global challenges such as energy and sustainability. Additionally, this network supports Abu Dhabi’s efforts in climate work and digital transformation initiatives.
The “Condor Galaxy 1”, the first AI-powered supercomputer on this network, has a computational power of up to 4 exaflops and a processor with 54 million cores. The two companies plan to launch the “Condor Galaxy 2 and 3 supercomputers” in the United States in early 2024. The network represents a leading supercomputing infrastructure that will provide processing power of up to 36 exaflops, a significant leap in the development of AI technologies globally.
Talal Al Qaisi, CEO of “G42 Cloud”, a subsidiary of “G42”, said: “We are pleased to collaborate with Cerebras to provide one of the fastest AI-powered supercomputers in the world and establish a foundation for connecting a group of these supercomputers worldwide. This partnership combines Cerebras’ unique computing capabilities with G42’s extensive and specialized experience in various sectors. The shared vision of both companies is to use the “Condor Galaxy 1″ supercomputer and other network devices to address important challenges facing society in healthcare, energy, environmental and climate work, and more.”
The new computer, launched in Santa Clara, California, connects 64 systems of “Cerebras CS-2” to a user-friendly AI-powered computer capable of training AI models at a rate of 4 exaflops. The device relies on high-quality support services provided by Cerebras to offer excellent performance for AI-powered supercomputers, without the need to manage or distribute training models across multiple devices.
The launch of “Condor Galaxy 1” marks Cerebras’ first partnership in building and managing high-capacity AI supercomputers. This partnership aims to empower G42 and its cloud computing clients to quickly and easily train and create AI models to accelerate the innovation process. The strategic partnership between the two companies has enabled the development of the latest AI models in areas such as Arabic language bilingual chat, healthcare studies, and climate research.
Andrew Feldman, CEO of Cerebras Systems, said: “The “Condor Galaxy 1” computer contributes to reducing the time required to train AI models and simplifying distributed computing efforts. Several cloud solution companies have announced the launch of large sets of graphics processing units costing billions of dollars, in addition to their complexity. Distributing a single model across thousands of small GPUs takes months and significant efforts from experts, highlighting the importance of the new device and its ability to overcome this challenge. It can create an AI model in minutes instead of months with no more than one person. The “Condor Galaxy 1″ is considered the first model of the three AI-powered supercomputers the company intends to launch nationwide in the United States next year.”
The United Arab Emirates University is developing AI and computing technologies.The company “JG 42” is working on developing efficient solutions for businesses, organizations, and governments in the field of digital transformation, and is considered a leader in this field thanks to its commitment to innovation, superb design, and investment in research and development. The establishment of the Mohamed bin Zayed University of Artificial Intelligence is considered as a pioneering university in focusing solely on artificial intelligence. The process of preparing large artificial intelligence models requires a significant amount of computing power, massive datasets, and specialized AI experts. “Cerebrus” and “JG 42” have formed a partnership that provides the three necessary elements for training large artificial intelligence models. The goal of both companies in launching the Condor Galaxy supercomputing network is to provide access to artificial intelligence for as many communities as possible, allowing for access to cutting-edge models in the field of AI computing. “JG 42” aims to utilize diverse datasets in areas such as healthcare, energy, and the environment, enabling system users to train new and advanced models. These models and applications aim to advance humanity. A distinguished team has been assembled, including hardware engineers, data experts, AI specialists, and industry professionals to deliver a comprehensive artificial intelligence model that provides solutions to customer problems. This model contributes to achieving excellent results and substantial support for hundreds of artificial intelligence projects worldwide. Forrest Norrod, Senior Vice President and General Manager of the Data Center Solutions Group at AMD, stated that the company is committed to advancing the field of artificial intelligence using high-performance computing processors and flexible computing products, in addition to collaborating with innovation sponsors like Cerebrus. He emphasized their shared vision with Cerebrus for a comprehensive application of AI technologies. The Cerebrus Condor Galaxy-1 device provides massive computing resources to researchers and companies in order to enhance reliance on artificial intelligence, with over 70,000 AMD EPYC processor cores. The computer “Condor Galaxy-1” directly supports training with large sequence lengths of up to 50,000 tokens without the need for additional software. These computers are fully programmed without the need for complex distributed programming languages, making it easy to run various models regardless of their size and providing the necessary time for distributing work across thousands of GPU units. Cerebrus produces Condor Galaxy computers in high-performance facilities for grid delivery in Santa Clara, California. The CS-2 systems are designed, assembled, manufactured, and tested by Cerebrus in the United States. Cerebrus is the only company specializing in AI systems that integrate and manufacture processors in the United States. The “Condor Galaxy-1” represents the first model of the three supercomputers that rely on artificial intelligence, alongside “Condor Galaxy-2” and “Condor Galaxy-3” produced locally by Cerebrus in the United States in collaboration with JG 42. The two companies aim to launch the largest artificial intelligence-based supercomputer consisting of 192 Cerebrus CS-2, by integrating these three supercomputers to achieve a capacity of up to 12 exaflops, distributed over 162 million cores and more than 218,000 high-performance AMD EPYC CPUs. The two companies plan to launch six additional high-performance computers of the Condor Galaxy type and link them to the network by 2024, increasing computing power to 36 exaflops.
