Can we just say thank you for Squad's latest attention to performance tweaks?

[Franklin]

FPS hasn't left 60 since the update. Game runs as smooth as butter now, with all the settings maxed. I may even be able to re-add Universe Replacer.

edit: nope, seems UR still drags my FPS below 30, ah well

Edited December 18, 2013 by Franklin

[regex]

I haven't noticed any FPS improvements but the scene changes are much faster, and that alone is worth it.

[bsalis]

Have not had a chance to get my mitts on .23 yet. However I noticed that the platform you have matters a lot with this game... Probably more a Unity thing though.

What platform are you guys on? Win, Unix, Mac, .. Intel, AMD?

[Whackjob]

I've gotten a five-to-six-fold increase in FPS when watching this.

[Franklin]

What platform are you guys on? Win, Unix, Mac, .. Intel, AMD?

windows/amd

[TNM]

Did anyone else notice that the

default: max physics delta-time per frame

is now 0.04 and the slider only can do so much in comparison to older versions?

Do i remember wrong?

[Franklin]

No, I think it was .10 prior. And you can't seem to warp past 1000x on the launchpad now.

[Madrias]

Noticed mild performance increases. Windows, Intel build, nVidia cards. I won't mention system stats, other than that there's more CPU behind this thing than needed for almost any game out there.

[vexx32]

-snip-

I've gotten a five-to-six-fold increase in FPS when watching this.

I have come to the conclusion that if we ever need a really good metric for FPS improvements, we should ask Whackjob to load up his most monstrous creations.

[Santoron]

Will be the standout feature of .23 for me. Ship and station designs I avoided/abandoned due to poor game performance have been reopened to me. Thanks Squad!

[Spatzimaus]

I've gotten a five-to-six-fold increase in FPS when watching this.

When you're measuring framerate in seconds per frame, instead of the reverse, it's easy to see an improvement.

But really, yes, it's a tremendous improvement if you're using huge constructions. I'm not Whackjobian, but I've got some big rockets, and they're much easier to control now. Once we get Unity multithreading, things'll be even better, but who knows when that'll happen.

[pa1983]

Did anyone else notice that the

default: max physics delta-time per frame

is now 0.04 and the slider only can do so much in comparison to older versions?

Do i remember wrong?

I have not checked what the new default is but if its 0.04 (remove settings.cfg to verify) its lower then before. It was 0.10 before.

You cant set it in the game to higher then 0.12 and I think it could go a lot higher before. Have to check with an older version of the game later.

Lowering it will trade Physic time for Frame rate so in practice that means more CPU time is spent on rendering and less on physics.

Advantage is that High part count models will run at a smooth FPS with pretty much the same Real time Rendering.

The down Side is that low part count models might run at less then real time, lower physic time that is with out any benefit if your FPS is already at say 60 FPS with Vsync.

I personally set Physic time to 0.03-0.01 with 1000+ parts and with say 500-600 parts I run it at 0.04-0.03.

Running it at 0.01 for example with 500 parts will give me a physics time of 0.6x but and a steady frame rate but running at 0.04x will give me close to 1.0x Physics time so real time rendering but still maintain a good smooth FPS.

Running a 1300 part ship with 0.10 will run at about 0.3x real time but with a few FPS at best. Running at 0.03x will give me more or less the same 0.3x real time physics but a lot better frame rate tough not smooth its not still frames at least. 0.01 will reduce physics time to 0.25x or so at a more steady frame rate.

So from my testing since 0.18 its preferable to lower Physics Delta with increased part count. Exactly whats optimal is hard to say, it depends on GPU and mostly CPU, what ever limits you. But there is a balance between CPU, GPU and part count for the optimal Physics Delta. 0.10 is good fore some that run low part count models but others like 0.03 and others like something in between or higher then 0.10.

Suggest using the debugging menu to show Physics time and some tool to measure FPS.

But if they lowered Physics Delta to 0.04 as default they did not realy do anything new to tweak the game becuse the user could do this before in the settings menu and I have ever sense I started playing back in 0.18.

Might be Why my i7 3930K @ 4Ghz and GTX570 did not get any performance difference in 0.23 VS 0.22 with any of my test crafts either I ran 0.10, 0.04, 0.03 or 0.01 I still got about the same Physics Time compared to previous version and the same perceived frame rate. I did not see the 30% or so boost on and i7 Mu was talking about.

Tried both the windows version in wine and the native Linux port. I cant find any sign of performance optimization.

Changing the default in the settings.cfg cant realy be called improving performance because it all depends on how many parts you run. People using few parts would probably prefer higher Physics Delta while those that use hundreds of parts would prefer a lower Physics Delta.

If they lowered Physics delta to 0.04 one should run in to the yellow clock a lot sooner when it comes to part count sens less CPU time is spent on physics rather then rendering so the game will run slower a lot sooner but maintain a higher frame rate even with more parts then with the previous Physics Delta at 0.10 in 0.22 and older if thats the case.

I think we need Mu to confirm if that Performance update was implemented in 0.23 that was suppose to be in 0.22 but was left out. I cant replicate his 30% boost on an i7 that he claimed and I have an i7. Cant even get close to a noticeable boost.

Edited December 18, 2013 by pa1983

[manskal]

Massive performance increases might also be related to the fact that many people are currently running full stock and clean installs, whereas previously they might have a metric c**-ton of mods installed.

Still, scene changes are noticeably faster now, thanks to Squad. Now, where's my 64-bit Unity for Windows, Unity Dev Team?

[Palleas]

thank you squad

[Spatzimaus]

Massive performance increases might also be related to the fact that many people are currently running full stock and clean installs, whereas previously they might have a metric c**-ton of mods installed.:

Nope. I transferred my mods directly from 0.22 to 0.23; only five needed any sort of updating to be compatible with the new version. There really was just a substantial improvement in the engine itself.

[pa1983]

If you dont run the same physics Delta in settings as previous version aka 0.10 a comparison is invalid. Any one could set0.04 before in settings and get the same results as 0.23 do by default now.

And yes 0.23 breaks the previous 0.10 as Default delta physics settings and its now 0.04 and that will give better FPS at the cost of Lower real time rendering as soon as one of you CPu cores is maxed out by the main thread.

So no its not odd that people get better frame rates at the cost of real time rendering and the later is hard to notice at first but can be pretty evident especially if you turn on debugging and also starts to check the clock if its yellow or not as you increase part count. With lower Physics Delta it will go in to the yellow sooner with less part then at a higher Delta Physics setting.

I have been running like that sense 0.18. Changing default settings is hardly a performance improvement just a tweak to rebalanced the game to work optimally with the part count people might use. Thats why I set it manually depending on what craft I use.

All the testing I did to day on my i7 showed no improvement realy over 0.22 or any other version with the same physics delta compared to 0.22.

So no no improvement only thing Physics Delta do is prioritize if you spend CPU time on Rendering or on Physics. They still need to speed up the code and get som proper multi threading or else where stuck with the same crappy performance.

Im not convinced Mu's improvement was put in to 0.23. He talked about as much as 30% improvement on an i7 and more for slower processors.

Changing Physics Delta alone is not much of a optimization its just a tweak the user can do.

People can easily pull back Physics Delta in settings to 0.10 and try that and see if the game runs as "before" pre 0.23.

Edited December 18, 2013 by pa1983

[keoki]

This is really nice.... I can't wait till I have all of my research unlocked and can try my heavy lifters again.... But so far it feels really smooth with all of the settings maxed out.

Thanks Squad! The best Christmas present of the season!!

[Spatzimaus]

If you dont run the same physics Delta in settings as previous version aka 0.10 a comparison is invalid. Any one could set0.04 before in settings and get the same results as 0.23 do by default now.

Nope. In one of the other threads, someone did the inverse of that by setting both versions to 0.10 physics delta. v0.23, at 0.10, still had much higher frame rates than v0.22 did at the same setting.

Face it, it's improved.

[pa1983]

Not seeing that on my i7 3930K @ 4Ghz and GTX570. Actually no version sense 0.18.2 have had any measurable improvement in performance loading screens aside but thats pretty irrelevant In my opinion.

So no I dissagre I get the exact same results 0.22 or 0.23. But I dont care mush for Squads promises any more. Just another disappointing release.

And frame rates alone is not enough to measure KSP preference. People tend to forget that the game wont run in realtime if the CPU cant keep up. If the game only traids that for better FPS the real problem is not fixed. So just because some on gets better FPS wont necessarily mean the game runs better.

I have run 0.01 for 6 months just because it gives a smooth frame rate preferably at 60fps V-Sync and even with 1300+ parts one can get like 30 fps but real time rendering due to physics slowing down is a side effect. But 20-30 fps at some what lower real time is better then higher real time at 2fps.

I ran 0.04 with my 5-600 part crafts because it gave more or less smooth frame rates at more or less 1.0x real time in 0.22.

So fare I cant measure that the sum of the two (fps and real time rendering delta) has gone up in 0.23 VS 0.22 meaning the game would runs more efficiently for me with 0.23.

So fare I cant tell any real difference between the two versions in performance.

Edited December 18, 2013 by pa1983

[Kasuha]

I might be wrong about it but I think the "max physics delta time per frame" value is irrelevant to FPS.

The game already works and even before worked as this: render a frame - do physics step - render a frame - do physics step - ...

But the size of the physics step was modulated by how long time has passed between frames and by this maximum cap.

CPU consumed on doing the physics step is the same regardless how big the delta t is, it's just that the physics is more accurate when the step is smaller.

With value of 0.1, the game started to slow down the game time versus real time below 10 FPS

With value of 0.04, the same starts to happen below 25 FPS

[Demon_82]

With "max physics delta time per frame" at 0.03, as I had it before, my 5 years old Intel CPU (soon to be replaced) is crying of happiness seeing the increase in framerrate. A lot more of the time I'm getting the mighty 60fps, and when I load my biggest monstruosities, I'm going up from 8fps to 15fps. Middle flying monsters that gave around 15fps went up to 30fps or more.

[Daishi]

Copied from another thread.

Did a few experiments. Turned 0.23's "Max Physics Delta Time" back up to 0.22 stock value (0.10) and ran some framerate tests on a 250 part vessel with fraps.

Repeated the same tests on a 0.23 0.04 (version stock) setting, a 0.23 0.10 setting, and a 0.22 (version stock) 0.10 setting.

0.23 "Fenrir C00". 251parts (w\o clamps), MPDT 0.04 (stock)

27FPS on launchpad

23FPS on ascent

25fps on 1st stage seperation, debris out of physics range. 159 parts

25fps on 2nd stage seperation, debris out of physics range, 126 parts

0.23 "Fenrir C00". 251parts (w\o clamps), MPDT 0.10 (old .22 stock)

29FPS on launchpad

20FPS on ascent

20fps on 1st stage seperation, debris out of physics range. 159 parts

20fps on 2nd stage seperation, debris out of physics range, 126 parts

0.22 "Fenrir C00". 251parts (w\o clamps), MPDT 0.10 ( stock)

9FPS on launchpad

8FPS on ascent

15fps on 1st stage seperation, debris out of physics range. 159 parts

16fps on 2nd stage seperation, debris out of physics range, 126 parts

0.23 still preforms superior, irregardless of what physics delta time you set it at. I wonder why they decided to change it.

(system specs - i7 970 (12 cores at 3.20gHz), 12gigs of ram, GTX580 - game antialiasing offloaded to gpu with the nVidia control panel on all game versions)

Very noticable differences for me, physics-time-per-frame changes or not. Both versions running the same mods, universe replacer being one of them, barring Flight Engineer cause that broke.

So yah. Big thanks to Squad indeed!

[TMS]

I run on an i7 and I see improvements.

Space Centre scene changes are no longer lagging (0.22 helped this, but didn't solve it)

SOI transitions are much more silky (again 0.22 didn't solve this fully)

LKO rendering is smoother, even with modifications

That bodes well for higher part counts, but yeah, this is all very welcome as-is.

[pa1983]

Well I have to test more later but sence people get goodcresults I supose I have some tweaking to do. Tough I had good fps before so I ddon't expect to much. Over 16+ fps with 534 parts in 0.22 with 0.10 delta.

Usually ran 0.03 or lower doubeling that fps.

[KasperVld]

It has been stated that the performance increases would be most noticable with large craft and lower end machines, and that's where I think your confusion comes from: the people who are noticing very little impact all seem to have high end i7 processors.