None
EN
Scale Up, Scale Out Challenges Amplified For Clusters
['Liz Allan', 'Technical Paper Link', 'Tom Katsioulas', 'Alex', 'Matt Bailey', 'Alexis R. Ware', 'Yağız Boz', 'Murugavel Ganesan', 'F. Chen', 'B.S. Deepaksubramanyan']
Semiconductor Engineering
SE: Compared to the scale up and scale out of racks in a data center, can clusters help reduce power consumption, or do they use more power and are therefore a part of the AI data center’s power problem?
But the software allocation and utilization of those pieces, or the slicing of that hardware cluster into software or virtual groups, is a different problem.
To solve the [power] problem we just discussed, you need to know and see the history of previous workloads at run time.
Emmer: Clusters don’t solve the power problem by themselves.
SE: When it comes to interconnects, I/Os, and data flow, are there different challenges between rack scale up, scale out, and compute clusters, or are they similar?