Cerebras and G42 begin building CG-2
Second 4 exaFLOP system to be added to the new AI supercomputer network
#UAE #USA #supercomputing - Sunnyvale-based AI chip and systems company Cerebras Systems and Abu Dhabi technology holding group G42 have begun building the Condor Galaxy 2 (CG-2), their second four exaFLOP AI supercomputer with and 54 million AI-optimized compute cores. The two companies announced their first supercomputer, Condor Galaxy 1, in July, as the first system in a massive nine supercomputer network offering cloud services for AI model training,
SO WHAT? - Condor Galaxy 2 is the second step in a plan to create the world's most powerful commercial supercomputer network for training AI models. The new addition will bring the Condor Galaxy network up to a total of eight exaFLOPs and 108 million cores. Like CG-1, CG-2 is designed to train the largest AI models, with long context, out of the box. The supercomputer cloud service allows developers to work on the largest models as if they had the simplicity of a 1B parameter model, streamlining development and shortening time to solution.
The Condor Galaxy constellation enables simple and easy access to the largest and fastest AI compute in the world, whereover your developers are.
Training large AI models requires a massive amount of compute, vast datasets, and specialised AI expertise. G42 and Cerebras aim to deliver across all three.
Condor Galaxy 1 was already used to train G42 healthcare group M42's Med42 70B large language model, and G42's 30B Arabic LLM.
The full nine AI supercomputer network will deliver 36 exaFLOP performance, requiring 576 Cerebras CS-2 systems and feeding the cluster with more than 654,000 AMD CPU cores.
ZOOM OUT - High performance computing systems are expensive, requiring high price and difficult to procure hardware, facilities and skilled engineers, not to mention immense amounts of energy. Add to that the fact that supercomputers are only really cost-justifiable if the organisation is an intensive user of HPC. The global AI chip shortage and the export licencing rules introduced by the US government this year, add further obstacles to those requiring high performance computing. Against this backdrop, buying access to AI supercomputing cloud services is going to provide an increasingly attractive option for public and private organisations.
IMO - Cerebras Systems and G42 seem to have created a magic combination for an AI supercomputing cloud service that few will be able to try and match. The demand for compute capacity for the development of AI models is sure to increase globally, and outpace the availability of AI chips. Whilst certainly ambitious, the Condor Galaxy network appears to make perfect sense.
Also read:
Cerebras & G42 build massive cloud supercomputer network (Middle East AI News)
G42 merges Injazat with G42 Cloud and Inception (Middle East AI News)
Will GenAI champion the Arabic language? (Middle East AI News)