Abu Dhabi, UAE – G42, a technology holding group based in the UAE, and Cerebras Systems, a pioneer in accelerating generative artificial intelligence (AI), have launched Condor Galaxy, a network of nine interconnected supercomputers. The network is designed to significantly reduce AI model training time and revolutionize the advancement of AI globally.
The Condor Galaxy network is based on the Cerebras CS-2 wafer-scale engine, which is the world’s largest AI accelerator. Each CS-2 chip has 400,000 AI cores and 1.2 terabytes of on-chip memory. The nine supercomputers in the Condor Galaxy network have a total of 36 exaFLOPS of computing power, making it the world’s largest AI training supercomputer network.
The Condor Galaxy network is designed to address several challenges in AI training, including the need for more computing power, memory, and faster data transfer speeds. The network’s large scale and high performance will allow researchers to train larger and more complex AI models, which will lead to new advances in AI applications such as healthcare, transportation, and manufacturing.
The Condor Galaxy network is also designed to be energy efficient. The Cerebras CS-2 chip is designed to be very power efficient, and the network’s interconnect technology is designed to minimize data transfer latency. This will help reduce the environmental impact of AI training, a major concern for many researchers.
The launch of the Condor Galaxy network is a significant milestone in the advancement of AI. The network’s large scale, high performance, and energy efficiency will allow it to train larger and more complex AI models, leading to new advances in AI applications.