Performance hit when flying through a cloud layer

[FlyAgi]

I figured out how to get reasonable performance with my old GTX 1070 in most conditions but there is still one situation breaking all of this: If I start climbing towards a cloud layer from below at some point when the clouds start to create some kind of white-out effect (the effect is fine) my GPU load goes up heavily while flying very close to the clouds layer or when trying to climb through a hole to get above the clouds. As soon as I get close enough from below the GPU load increases and keeps high until I get high enough above the cloud layer (will upload a video later on showing this effect, my upload takes some time).

 

[sundog]

3 minutes ago, FlyAgi said:

I figured out how to get reasonable performance with my old GTX 1070 in most conditions but there is still one situation breaking all of this: If I start climbing towards a cloud layer from below at some point when the clouds start to create some kind of white-out effect (the effect is fine) my GPU load goes up heavily while flying very close to the clouds layer or when trying to climb through a hole to get above the clouds. As soon as I get close enough from below the GPU load increases and keeps high until I get high enough above the cloud layer (will upload a video later on showing this effect, my upload takes some time).

 

Being inside a cloud is the worst case scenario for volumetric clouds, as every pixel on the screen needs to be ray-cast all the way through them all in this case. So it's expected that performance would be at its worst from this position.

Not much to be done except the usual performance tips; lower your cloud draw area or rendering settings. Anti-aliasing and high resolutions make matters worse.

 

[FlyAgi]

That's what I have expected. Sadly there is not much I can do as lowering AA below FXAA does nothing and rendering settings don't have that much of an effect to this load spikes (the objects slider only has a minor effect on GPU load, it's there but does not matter in most cases).

 

[FlyAgi]

Not sure if this can be done technically but maybe it's possible to stop rendering voxels which are not visible meaning the ray march stops if it reaches a point where this pixel/voxel won't change anyway as it's already fully opaque no matter how many stuff behind this point is added to it? Maybe I get the ray marching wrong but I can imagine that there is some space to stop the marching not only a the defined threshold but also in certain cases to improve performance.

[sundog]

3 hours ago, FlyAgi said:

Not sure if this can be done technically but maybe it's possible to stop rendering voxels which are not visible meaning the ray march stops if it reaches a point where this pixel/voxel won't change anyway as it's already fully opaque no matter how many stuff behind this point is added to it? Maybe I get the ray marching wrong but I can imagine that there is some space to stop the marching not only a the defined threshold but also in certain cases to improve performance.

We actually do that already (it's called early ray termination.) It helps, but not as much as we'd like.

[FlyAgi]

Well... that's weird because I would think that, if I'm looking at a cloud layer from the same altitude, the performance in theory should be better than looking from below as the rays only need to pass some few clouds, some meters into the clouds blocking almost everything behind them.

[sundog]

12 hours ago, FlyAgi said:

Well... that's weird because I would think that, if I'm looking at a cloud layer from the same altitude, the performance in theory should be better than looking from below as the rays only need to pass some few clouds, some meters into the clouds blocking almost everything behind them.

It's because we actually have to shoot the ray pretty far before its opacity reaches the point where we can actually stop it early, especially if there are gaps in the clouds. Shooting through a cloud layer from above or below is a much shorter distance to traverse.

[FlyAgi]

Thanks for clarification... and by no means I want to tell you how to do things, that kind of coding you are doing is far beyond my skills and it feels like magic to some extend.

:-)

[sundog]

1 hour ago, FlyAgi said:

Thanks for clarification... and by no means I want to tell you how to do things, that kind of coding you are doing is far beyond my skills and it feels like magic to some extend.

:-)

You have a pretty good understanding of how it works TBH!

[FlyAgi]

That's just the basic theory, I could not put that into math/code and this is where the magic begins for me.

[danFly]

I have the trio of SkyMaxx v5 + Real Weather connector + FSGRW. I set to historic weather on 24/01/2021 at 12pm (snow) from EGCC to EGLL. Well, x-plane crashed 5 times, and highly gpu bounded, especially crashing when the weather was refreshing. When it was working the fps was about 10fps (on GTX1080). No setting changes in x-plane or skymaxx solved the problem. Only changing overcast to "soft hd" improved the frame rate to 35fps. 

PS: every time I untick volumetric cumulus cloud, xplane crashes.

 

[sundog]

11 hours ago, danFly said:

I have the trio of SkyMaxx v5 + Real Weather connector + FSGRW. I set to historic weather on 24/01/2021 at 12pm (snow) from EGCC to EGLL. Well, x-plane crashed 5 times, and highly gpu bounded, especially crashing when the weather was refreshing. When it was working the fps was about 10fps (on GTX1080). No setting changes in x-plane or skymaxx solved the problem. Only changing overcast to "soft hd" improved the frame rate to 35fps. 

PS: every time I untick volumetric cumulus cloud, xplane crashes.

 

Sounds like you're running out of memory; volumetric clouds use a lot of it. Try reducing your cloud draw area setting, or freeing up memory in other ways (reducing your anti-aliasing, turning off HDR, lower resolution, removing ortho scenery, etc)

[danFly]

Hi, I did try reducing the cloud draw and texture quality. Skymaxx menu was showing more available gpu memory, but this this did not have any impact on the gpu frame time. I am running on 3x QHD resolution, however, I am comparing the performance relative to Skymaxx v4. From 30fps to 10fps is quite a performance hit.

I guess especially when using real-world weather there is a need for an adaptive quality mechanism which perhaps reduces the number of rays to speed up the gpu frame time. After all this is a dynamic simulation, we don't know when we are inside the cloud and when few clouds only!

[FlyAgi]

Well... this sounds like your GPU is just not fast enough - I'm in the same situation (GTX 1070 in a single screen 1080p) so welcome to the club. The only thing which helped my out with volumetric clouds  performance is reducing the SSAA in X-Plane options one notch (leaving me with only FXAA and a lot of scenery flickering but I can fly this way at least).

Further, SMP always had some adaptive quality mechanisms such as cloud draw distance dynamically changes depedning on weather situation and altitude, the higher you fly, the more clouds you can see for example, and this is still present in SMP 5 with the volumetric clouds but it seems this is not as effective with the volumetric clouds as with the classic ones.

Reducing the rays casted to give a sigificant increase in performance will most likely result in a more grainy or blocky look of the clouds, they are already looking a bit blocky on the edges when seen from certain distances.

But yes, I would appreciate any kind of performance increase myself but I think a (much) faster graphics card is the way to go here (and I know there almost nothing to buy on the market currently, at least not for a reasonable price tag).