FPS hit

[mjrhealth]

Not sure what happened could be a driver update but last version of SMP killed my FPS was down as low as 15 above , 20 when in clouds. Previous version was fine in the 30s. Ill have another go when its cooler and see what happens

Log.txt

[sundog]

Performance can vary greatly depending on the weather conditions, the location of the camera relative to the clouds, the cloud draw area setting in SMP, your resolution, and your anti-aliasing settings. You'll probably want to tweak some of those to find the sweet spot for your system.

That said we do expect the fix in 5.0.4 that resolved the issue of mountains drawing on top of the clouds to come with a slight performance cost. But I wouldn't expect it to cut your performance in half. Might just be that the weather was different when you flew most recently.

Edited January 25, 2021 by sundog

[FlyAgi]

I also noticed there is severe performance impact with thunderstorm/towering clouds involved. If I understand correctly those are still classic style clouds so I don't understand why there is a massive hit with them - even if there is low visibility, massive fog present (about 10 sm, so low clouds draw distance) and very few volumetric style clouds present (just one or two small volumetric clouds not touching the storm clouds), the towering clouds can reduce performance heavily (this is more severe than the volumetric clouds when flying through or close above/below them).

[benjaprud]

For me so far I had the worst performance hit in some low overcast conditions, I'm still not sure which set of conditions lead to that much performance impact yet. One was low overcast (4000 ft) with two broken layers underneath, another one was two successive layers of low overcast with low visibility.

[FlyAgi]

Low volumetric cloud layers will have more impact than higher ones as the rays passing through the clouds create more work for the GPU the closer you get to the clouds because of the angle the camera looks at the cloud formation. In low clouds situations the clouds come closer to you so low clouds cause higher FPS impact.

But, as the thunderstorms are non-volumetric I wonder why the have a more dramatic impact on my FPS than flying closeto the volumetric clouds (with full classic style clouds the impact is not that heavy).

 

[sundog]

25 minutes ago, FlyAgi said:

But, as the thunderstorms are non-volumetric I wonder why the have a more dramatic impact on my FPS

They are just really huge clouds that are composed of a large number of individual cloud billboards. They result in a lot of depth complexity which taxes the fill rate of your GPU.

[FlyAgi]

Yes, but why is this impact only there when using volumetric clouds? I never experienced this impact in SMP 4.x, there was a noticable hit but it was not that severe (as least I don't remember that).

[Cameron]

35 minutes ago, FlyAgi said:

Yes, but why is this impact only there when using volumetric clouds? I never experienced this impact in SMP 4.x, there was a noticable hit but it was not that severe (as least I don't remember that).

Well, simply because your GPU is already working hard enough to process volumetric ray casting. Adding in another element to have that fill rate issue is just compounding on top of it.

It's a balance. The GPU ray-casted clouds are the fastest processing clouds we've ever developed. But, they also will work your GPU hard. If you already have settings that are reasonable to you, mixed with numerous add-ons, it may suddenly become unreasonable with the ray casting. On the flip side, if you're more or less like myself who only had minimal add-ons (save for some orothos and maybe ORBX, plus reasonable GPU settings), then you can notice less an impact and the improvements discussed.

Finding the balance can be difficult. And, as we get more feedback we'll try to work through that on our end too. We are somewhat at the cusp of GPU's really being able to handle this stuff too. For instance, we've had the ability to do this for years, but hardware just wasn't there. The 1x series of Nvidia cards are especially at the lower end of that spectrum, as is probably my 5700xt, but even I can get it to acceptable numbers.

[FlyAgi]

10 minutes ago, Cameron said:

Well, simply because your GPU is already working hard enough to process volumetric ray casting

Even if there are about no volumetric clouds visible around the thunderstorm cell? Well... if I think further this might make sense as the ray casting has to be done even if there is no cloud reflecting them... then the ray casting creates some kind of 'base GPU load' even if there are no clouds visible which the classic only clouds don't and everything adds up to this, is this right?

Also I know my graphics card is lower end in this regard but I also have a need for 60 FPS to feel comfortable so it's another special situation (if I was fine with 30 fps I think I would have no issues at all with performance).

 

[Cameron]

4 minutes ago, FlyAgi said:

is this right?

Essentially, yes. @sundog can further clarify that.

5 minutes ago, FlyAgi said:

I also have a need for 60 FPS to feel comfortable so it's another special situation (if I was fine with 30 fps I think I would have no issues at all with performance).

Ya, you're in-between a rock and a hard place! That FPS requirement is probably not realistic on ray casting with a 1080.

[FlyAgi]

Well... it works to some extend because I'm mostly flying small helicopters so I can stay away from the clouds, this works fine in most situations but might get hard with thunderstorms or towering cumulus clouds if they are very low above the ground. Further, I'm only with 1070, the 1080 with about 25% more processing power would help in many situations already.

 

[benjaprud]

I'm not sure if dedicating 80% of the GPU frametime to just clouds and 20% to all the rest is a sound expectation to have. It leads to huge variablility in performance and basically asking users to compromise on all the rest just to make some room for the clouds.

[Cameron]

Just now, benjaprud said:

I'm not sure if dedicating 80% of the GPU frametime to just clouds and 20% to all the rest is a sound expectation to have. It leads to huge variablility in performance and basically asking users to compromise on all the rest just to make some room for the clouds.

Not really. Perhaps from a VR perspective if you need it super high.

Otherwise, I seem to do just fine running ORBX HD NORCAL, objects to medium, ORBX airports mixed in, and something like the IXEG 737. 35-40 fps on my 5700xt is absolutely acceptable and reasonable to me.

But, of course, I can't tell you you're opinion is wrong either. I can only counter with mine.

[dirmer3]

I noticed today that even without volumetric clouds I had lower fps than with V4. Might switch back and compare.

[mjrhealth]

The funny thing om 5.01 same flight had no FPS hit at all, even V5.0 was fine

[FlyAgi]

Well... if that is true with exactly the same weather situation there is some hope this might be a bug and finding this one would speed up things.  Have you saved the weather files or do you still know time and date from the weather files for testing? Using ASXP or FSGRW historical weather it should be possible to test exactly the same weather situation with different versions of SMP.

[Cameron]

20 minutes ago, mjrhealth said:

The funny thing om 5.01 same flight had no FPS hit at all, even V5.0 was fine

5.0.4 introduced a method to work around mountains showing through clouds, like seen here: 

 

In order to achieve this, a performance hit is unavoidable, though I am slightly puzzled why some see it harder than we did. It cost me about 3-4 fps.

[mjrhealth]

6 minutes ago, Cameron said:

5.0.4 introduced a method to work around mountains showing through clouds, like seen here: 

 

In order to achieve this, a performance hit is unavoidable, though I am slightly puzzled why some see it harder than we did. It cost me about 3-4 fps.

Not sure but im running same flight, see how it goes today. It was at sunrise but  5.01 had no issues at all.

[Cameron]

Just now, mjrhealth said:

Not sure but im running same flight, see how it goes today. It was at sunrise but  5.01 had no issues at all.

Be sure to give the new 5.0.5 a go.

[mjrhealth]

4 minutes ago, Cameron said:

Be sure to give the new 5.0.5 a go.

Oh ok didnt know there was one I will let you know. Cheers

[benjaprud]

I think the performance hits were there from the start. In my case I've noticed my first serious hits with 5.0.1 after several hours of flight. The fact that some people are noticing it after 5.0.4 is probably because of the short lifespan of previous builds since it takes particular sets of conditions to trigger worst performance scenarios.

[mjrhealth]

@Cameron installed, going for bike ride before gets to hot, let you know how it goes.

Thanks for all the good work

[Jakob Ludwig]

hi folks. I do have good frames with 5.0.5 together wir RWC and ASXP. Usually above 30 in every situation and even higher in some areas. But I noticed a lot of FPS spikes (FPS jumps constantly up and down by 5 fps and more) and thus resulting in micro stutters... My GTX1070 isn't that busy and there is still almost 1,5 - 2 GB of VRAM available. Turning SMP off in the plugin admin, gets rids of all stutters, indicating it must have something to to with SMP. 

[Cameron]

Hey @Jakob Ludwig

Can you tell me if you selected retail or beta Gizmo? Have you tried beta if not?

[Jakob Ludwig]

11 minutes ago, Cameron said:

Hey @Jakob Ludwig

Can you tell me if you selected retail or beta Gizmo? Have you tried beta if not?

Beta