As significant innovations in computing continue to be fueled by the developments in Machine Learning (ML) and Artificial Intelligence (AI), ScalPerf'25 will again focus on the interaction between these disciplines and High Performance Computing, in continuity with the theme of ScalPerf'24.
Given the enormous investments planned worldwide in computing infrastructures for AI and ML, an interesting question is whether the overall cost of computing could be reduced should one design, possibly from scratch, machines fully devoted to specific tasks, such as neural network and transformer computations. Part of the discussion sessions of the workshop will be devoted to this theme, with the objective to collectively imagining the broad structure and the salient features of such a machine. The rest will focus on the issues proposed (in advance) by the participants, as customary in ScalPerf.