When not to enable RTX execution strategy

OptiX 6.0.0 gives the option to disable the RTX execution strategy using RT_GLOBAL_ATTRIBUTE_ENABLE_RTX. My question is, when is it necessary to do this?

Some calls such as rtContextSetStackSize are not supported with RTX. Does this mean that we should not call rtContextSetStackSize, or are there cases in which RTX will not be enabled and the code will fall back to using rtContextSetStackSize?

There are a few corner cases when running on pre-Turing hardware where the pre-rtx mode can be a little faster than the RTX mode. However, even on older hardware, RTX-mode is usually faster – so unless you have a specific case where you are seeing perf benefits to the legacy mode, you should just go with the default (RTX) mode.

We will likely faze out non-rtx mode in the near future as we improve these few remaining perf cases.