The GPU app uses one cpu core at 100%. I know, it is default behaviour for CUDA apps to spend time in busy wait loop, however in this case it looks like a waste of resources. For example I throttled the app down to 10-20% using TThrottle and it didn't cause anby noticeable slowdown, i.e. GPU still loaded at 100% and run time of the app remained more or less the same. Is it possible to get rid of busy wait? It's worth doing, because there are other more useful things you can do with cpu.

You need a developer that understands exactly what the code in the app is doing. Then you need to have them optimize the app with them being a CUDA wiz.

Only developer I know of capable of this is Petri33. And he is busy right now optimizing the Einstein CUDA apps. Maybe afterwards, he might be persuaded to take a crack at the Minecraft app.