Updating the Energy Model for Future Exascale Systems

摘要

The 2008 DARPA Exascale report had as its goal determining if it were possible to achieve 1000X the computational power of the then-emerging peta-scale systems at a system power of no more than 20 MW. The main conclusion was that there was no such path with technology and architectures as projected at that time. Key to this conclusion were architecturally-tailored models as to how projected advances would translate into system performance. This paper introduces a major update to the "heavyweight" (modern server-class multi-core chips) model, with a detailed discussion on the underlying projections as to technology, chip layout and microarchitecture, and system characteristics. The model is run over the same time period as the 2008 model to verify its accuracy.