Today,the biggest decision that IT has to make is how much per-node performance to make available for an AI workload, taking into account these aforementioned components and how they are connected and optimized together, because that’s how performance gains are achieved —not with just a bunch of GPUs but with a platform that maximizes the utilization of those GPUs. IDC believes that IBM’s Power System AC922 is an excellent choice to achieve maximum per-node performancefor AI training. After all, the Power System AC922is the building block of the world’s fastest supercompute