Thank you for sharing your tests and results.
I’ve been doing some more flying yesterday with the A320 over EDDF in trying to spot whether there would be something else to tackle. I’ve for example disabled PostProcess > Sharpen to see whether I’d gain more fps if everything is set to 100% (TAA + SS). It is marginally giving a little more and this was more noticeable to me when trying TAA70+SS150 where, although a little more blurry, it appeared smoother sometimes.
During the A320 @ EDDF test though, I’ve noticed something odd: in doing fast bank left/right, when looking outside the left, it was really nearly butter smooth always but when looking to the right, it was updating something like 10fps (no fpsVR, just from the look of it). This made me wondering what is the difference besides scenery so I turn 180 deg and again same thing.
It turns out when looking to the right you have a lot of EFIS in view whereas to the left there is none. And here came a surprise: changing the Gauge Update rate from Medium to Low exhibited even more stutters! changing it to high nearly eradicated these stutters and restored an even fps.
My understanding is this: in lowering the gauge refresh rate, FS2020 is rendering all EFIS at once say at 5Hz. This means it is rendering the whole 3D fine and every so often a burst of EFIS rendering intervenes and maybe because this is badly scheduling into the FS2020 task scheduler it is taking a whole lot more ms than the frame budget. The renderer is then waiting for the EFIS rendering completion until submitting a new frame to OpenXR and this is a missed (or more) frame(s). However when set to HIGH, the EFIS is constantly overbudgeting the frame time budget and the renderer is always missing every other frame mostly, with the result being you get a more even stream of frames. In effect, this would resemble this:
#: frame, E: EFIS, !: missed frame budget (= stutter), *: display
LOW : 0, 1, 2, 3, E, !, 4, 5, 6, 7, E, !, !, 8, 9, 10, E, !, 11, ...
VIEW: *, *, *, *, ., ., *, *, *, *, ., ., ., *, *, *, ., ., *
HIGH: 0, E, !, 1, E, !, 2, E, !, 3, E, !, 4, E, !, 5, ...
VIEW: *, ., ., *, ., ., *, ., ., *, ., ., *, ., ., *, ., This explains to me why you always get better performance with analogue cockpits in general because there is less load due to rendering using the Coherant GT tech (see my comments below), but what matters most in VR is constant fps rather than high fps as we can see FS2020 is fluid enough at 30fps in VR as long as there is no stutters.
In choosing Coherant GT for gauge logic and display, FS2020 is having a capable universal solution for UI which is scriptable and easy to deal with for creating new gauges more easily but this makes me wondering this is not as well impairing the performance we can really get out of the AAA rendering engine Asobo has created: https://coherent-labs.com/posts/coherent-gt/
For example the way I’d do implement EFIS rendering would be using a round-robbin fashion so that you’d only refresh 1 screen at a time, not the whole 6 in burst in the A320. This would spread out the load over multiple frames and would be unnoticeable from the user. In LOW, it would be 1 at a time, in MED it would be 2 at a time and in HIGH 3 at a time with ULTRA all at once. The problem is I’m not sure the Coherant GT tech would permit them doing so easily or even if the JS code they’ve built upon Coherant GT and serving as the foundation block of all other gauges and systems is even coded with this kind of optimizations in mind.
I’m not much concerned with the ability to use add-ons, but I doubt the implementation from being able to offer much more than this and I hope I’ll be proven wrong. For airliners, Asobo is focusing much exclusivity to a single 3rd party vendor for the time being who is most likely concerned to offering the first airliner to the market place, and Asobo is using this vendor expertise as their main, if not sole, source of knowledge about what it is needed in the gauge and system SDK. I’m certain all add-ons will be as capable as their airliner, both for its good and its bad, no more no less either. Kind of a “we believe every vendors is doing like them, because they’re doing this for a long time and selling a lot, therefore, this is how all other vendors must and should do their aircraft”.
Having said this, I’m about to test the 446.14 again and see what it gives though. This is unrelated to Coherant GT limitations I’d like to really see whether the NVidia VR only stutter bug in more recent drivers is really that much of a problem with FS2020 now that I can reduce the simulator stutters with a HIGH gauge refresh rate.