It's about AI performance
Nvidia and Amazon Web Services (AWS) announced the expansion of their strategic cooperation, which will also include the construction of the fastest artificial intelligence supercomputer, Ceiba.
Ceiba will have 16,384 GH200 Superchips and will have a performance of an incredible 65 ExaFLOPS. True, it’s worth saying right away that this is performance specifically in AI tasks, that is, we are not talking about calculations with double or single precision. There is no data on classical indicators.
But this is not all that expansion of cooperation will consist of. Amazon will also use the latest Nvidia adapters for its Amazon EC2 service. We are talking about accelerators GH200, H200, L40S and L4. Overall, the press release said the companies will combine the best of their technologies—from Nvidia's latest multi-node systems with next-generation GPUs, processors, and AI software, to the advanced virtualization and security of the AWS Nitro system, Elastic Fabric Adapter (EFA) interconnect, and UltraCluster.