Condor Galaxy-1: The world's fastest supercomputer with a processing power of 4 exaflops launched

Image credit : Reddit G42, a technology holding group based in Abu Dhabi, has revealed the Condor Galaxy-1 (CG-1), the world's fastest supercomputer with an impressive processing capacity of four exaflops and 54 million cores. The CG-1 is situated in Santa Clara, California, and will be operated by Cerebras, a US-based AI firm, adhering to US laws. The demand for supercomputers has surged with the growing prominence of artificial intelligence (AI) technology, as businesses seek powerful infrastructure to train their AI models. G42 aims to create futuristic infrastructure, collaborating with nations and corporations to address significant global challenges. To address the need for efficient AI model training, Cerebras, in partnership with G42, has developed interconnected AI supercomputers instead of relying on a single central supercomputer. This approach significantly reduces AI model training time, making setting up generative AI models a quick and manageable process, achievable by a single individual. The entire Condor Galaxy System, comprised of multiple interconnected supercomputers, will boast an unprecedented combined processing power of 36 exaflops. [caption id="attachment_188713" align="aligncenter" width="2500"] Image credit : SiliconeANGLE[/caption] The CG-1 supercomputer features 64 of Cerebras' flagship CS-2 AI processors, complemented by AMD's EPYC processor cores, providing 54 million AI-optimized compute cores, 388 terabits per second of fabric bandwidth, and 82 TB of memory storage. Operating at 16-bit computation, CG-1 delivers four exaflops of computing power, making it four times faster than the current fastest supercomputer globally. It has the capability to train 600 billion parameter models, extendable to support 100 trillion parameter models, surpassing the OpenAI GPT-4 model, which uses 1.7 trillion parameters. Furthermore, Cerebras and G42 plan to introduce two more installations of the supercomputer, CG-2 and CG-3, in the US in early 2024 while offering CG-1 as a cloud service to customers. Also read : Sphere Entertainment Introduces Next-gen Concert-grade Audio System Notably, CG-1 operates natively with 50,000 tokens and does not require any special software libraries, allowing programming without complex distributed programming languages. This eliminates the need to distribute workflows across GPUs, saving valuable time in the process.