| Abstract |
MN-Core developed by Kobe university and PFN is an accelerator for deep learning, but it can handle general-purpose computation. It won first place in the Green500 ranking on three occasions because it delivers high performance per power consumption. So it is expected to become an accelerator beyond GPUs. Currently, MN-Core series processors are under consideration as the accelerator in Post-Fugaku FS: the research project of the successor supercomputer of Fugaku operated by the Japan government. However, MN-Core does not yet have programming environment for general-purpose computation, even though it is expected to be used as an accelerator in HPC. In this presentation, I would like to talk about the consideration of using OpenACC as the programming environment for general-purpose computation on MN-Core, the design philosophy of it, use cases based on the behavior of MN-Core considering computational science users in particular, the programming model, specifications, language processing system such as compilers, and runtime libraries. |
