Cerebras & G42 build massive cloud supercomputer network
Cerebras & G42 build massive cloud supercomputer network
Cerebras will deliver the world's fastest supercomputer for training AI models
Sunnyvale-based artificial intelligence company Cerebras Systems has announced a strategic partnership with Abu Dhabi technology holding group G42 to create a massive supercomputer network offering cloud services for AI model training, together with a $100 million deal to supply ‘the world's largest AI supercomputer’ as the first system on the new cloud network.
The first system is the 4 exaFLOP, 54 million core Condor Galaxy 1, or CG-1. The two companies are planning to deploy two more such supercomputers, CG-2 and CG-3, in the U.S. in early 2024, taking planned capacity up to 36 exaFLOPs in total.
To put this into a Middle East perspective, the most powerful supercomputer order announced from the region to-date was the Shaheen III (an HPE Cray EX) ordered by Saudi Arabian research university KAUST (King Abdullah University of Science and Technology) last year. Equipped with more than 2,800 NVIDIA Grace Hopper Superchips, coupled CPU/GPU accelerators, Shaheen III is expected to be 20 times faster than KAUST’s previous supercomputer (Shaheen II) and to deliver 100 petaflops of performance.
Located in Santa Clara, G42’s CG-1 links 64 Cerebras CS-2 systems together into a single, easy-to-use AI supercomputer. CG-1 will have an AI training capacity of 4 exaFLOPs, which is the equivalent of 4,000 petaflops.
Cerebras and G42 will offer CG-1 as a cloud service, allowing customers to enjoy the performance of an AI supercomputer without having to manage or distribute models over physical systems. G42 plans to sell the cloud supercomputer services on to its commercial, government and institutional customers, whilst wholesaling excess capacity via Cerebras to customers in the open-source AI community.
Find out more about this story