The LOWAIN project has the following four goals:
Supercomputer simulations (CFD, NWP, etc.) represent the main application of supercomputers (e.g., 10 of the Top100 supercomputers are exclusively used for meteorology). This gives sufficient economical justification for a program of making processors with substantially reduced computing power, provided the reduced power is used efficiently for simulations. In this way it could be possible to build processors/accelerators that are substantially simpler and cheaper.
The motivation section showed that the memory bandwidth of supercomputer processors/accelerators is typically utilized to about 50 % when running the HPCG benchmark and it can be speculated that the same happens with other simulation programs.
Moreover, the present supercomputer memory systems do not offer very high memory bandwidth and very high memory capacity at the same time: HBMs have limited capacity and high capacity DRAM blocks have limited bandwidth.
One of the main goals of LOWAIN is to develop a simulation specific memory architecture (called ORPA) that would guarantee optimal use of
high memory bandwidth simultaneously with large memory capacity.
It is clear that the key parameter of the recent systems is the memory bandwidth. Its increase would enhance the performance of a supercomputer. The LOWAIN consortium has no skills that are necessary to build processors and memories with higher communication bandwidth. However, preliminary contacts exist in this direction with a major global society.