James Ross, on 06 March 2021 - 03:09 PM, said:
Partial investigation (more to follow)
I have done some analysis of the performance of the Stable Version, Testing Version, and NewYearMG version on my two computers: a laptop with integrated graphics, and a desktop with discrete graphics.
- Discrete desktop, default* settings: Stable 172, Testing 370, NewYearMG 298
- Discrete desktop, high* settings: Stable 163, Testing 371, NewYearMG 270
- Integrated laptop, default* settings: Stable 118, Testing 121, NewYearMG 133
- Integrated laptop, high* settings: Stable 61, Testing 55, NewYearMG 78
What's interesting here is that NewYearMG is performing better on integrated graphic, but the Testing Version is ahead on discrete graphics.
My guess at this point is that NewYearMG is doing something better on the GPU/graphics side, while the Testing Version is doing something better on the CPU/computation side (it may even be one thing that changes the CPU/GPU balance).
Settings used:
- Default settings are /MultisamplingCount=2
- High settings are /MultisamplingCount=2 /ConditionalLoadOfDayOrNightTextures /DistantMountainsViewingDistance=60000 /DynamicShadows /ModelInstancing /PreferDDSTexture /ReduceMemory /SaveCommands=False /SimpleControlPhysics /ViewDispatcher /WindowGlass
I will do further analysis tomorrow into the differences.
Knowing that ORNYMG does support 64bit I was wondering if that was impacting the results, but seems both on Carlos tests were done in 32bit mode, and also doing some testing here there seems minimal impact on performance from running ORNYMG in 32 or 64bit mode. My tests are running on integrated graphics as well.
Also both OR testing and ORNYMG do use the same version of Monogame, so there is something else causing this behavior.