A controlled performance experiment

[nhnifong]

To get some reliable data on lag, lets do a controlled experiment!

Download the craft https://dl.dropbox.com/u/14567061/lagship2.craft

Run it on a fresh install of 0.16. default setting and default (windowed) resolution.

take off straight up and measure your framerate.

Post your computer specs, and the framerate you got, after launching, during the first stage.

(warning this rocket was not designed to operate after the first stage.)

each section contains exactly 100 parts, so feel free to take it up to absurd levels but please report the results without modification first.

My machine:

Intel i5-2500K @ 3.30 GHz

Nvidia GeForce GTX 570

framerate 10 FPS

If you know that your GPU does not have the latest drivers installed, please report that.

Edited September 2, 2012 by nhnifong

[_Aramchek_]

Use fraps to see your fps. http://www.fraps.com/

There isn't a huge mystery behind the slowdown, but this will be fun anyway, to get an idea how the game performs on different hardware.

[nhnifong]

Ah thanks, I thought fraps was just for recording video.

[_Aramchek_]

So what are your actual fps? and that thing just explodes every time for me.

I get between 1-5fps, it fluctuates a lot because the craft is basically disintegrating.

BD [email protected]

GTX 560

I think liquid fueled rockets would give a better control, also something slightly smaller, as there should be a wider variance of performance with less parts and less particle effects.

Interestingly enough the game was thrashing 5 of 8 cores, moderately using the other two and gpu usage was about 2% average, peaking at 34% when teh craft exploded.

I was getting 3fps at the time.

After crashing to the ground, and fps jump up to 13x's, gpu usage stayed at mid 30%.

The gpu is sitting around almost all of the time waiting for the cpu to catch up, and obviously with the large crafts it's almost 100% reliant on the cpu.

https://dl.dropbox.com/u/80346/kspderplag.jpg

Edited September 2, 2012 by _Aramchek_

[Clouds]

Core 2 Duo T9400 @ 2.53 GHz

Mobility Radeon HD3650 256MB

KSP in windowed mode at 1280x720, all shaders to minimal/disabled settings, no scatters, terrain at a custom Very Low level, 16 pixel lights and 4 shadow cascades.

I estimate 0.75 FPS and I believe this is not GPU-limited, for the framerate does not improve by switching to map view.

[nhnifong]

So what are your actual fps? and that thing just explodes every time for me.

I get between 1-5fps, it fluctuates a lot because the craft is basically disintegrating.

BD [email protected]

GTX 560

I think liquid fueled rockets would give a better control, also something slightly smaller, as there should be a wider variance of performance with less parts and less particle effects.

Interestingly enough the game was thrashing 5 of 8 cores, moderately using the other two and gpu usage was about 2% average, peaking at 34% when teh craft exploded.

I was getting 3fps at the time.

After crashing to the ground, and fps jump up to 13x's, gpu usage stayed at mid 30%.

The gpu is sitting around almost all of the time waiting for the cpu to catch up, and obviously with the large crafts it's almost 100% reliant on the cpu.

https://dl.dropbox.com/u/80346/kspderplag.jpg

Lol, I guess it's not a very good test craft. I didn't bother to fix the staging for anything except the first stage.

[_Aramchek_]

I believe this is not GPU-limited, for the framerate does not improve by switching to map view.

Oh there shouldn't be any debate about that, unless you have onboard or an extremely old or low end gpu, the game just doesn't use much gpu power.

Lol, I guess it's not a very good test craft. I didn't bother to fix the staging for anything except the first stage.

I'd chunk it down to half that say, or maybe a third of that size and it would probably be a better test, the goal is not to completely overwhelm everyone's pc, but to still stress it good. lol

[nhnifong]

Here's a better test craft.

https://dl.dropbox.com/u/14567061/lagship2.craft

It's liquid rockets only, and smaller at 404 parts

I'm getting 10 FPS with it, as measured by fraps.

[_Aramchek_]

I'll give that a try, that should give a better display of variance between setups I'd imagine.

[Shortsonfire79]

Alright here we go. On the new ship that nhnifong posted on a fresh KSP install. I can't get fraps (Mac) so I timed it with my phone and counted the frames.

2010 13" Macbook pro.

Mac OSX 10.6.8

Intel Core 2 Duo 2.66 GHz

8 GB 1067MHz DDR3

Nvidia Geforce 320M

Framerate: 1fp 2-4s.

Total ingame flight time was 1:27.00. Time on my iPhone 3:40.00. For the first few seconds, I had one frame for every 2-3 seconds. At the one minute of realtime mark, I was at thirty seconds of ingame time. Two minute realtime mark, 51 seconds ingame time. At the end, the engine sound cut off four seconds before the fuel bars disappeared.

[nhnifong]

Thanks for the good test.

I've also noticed that there can be a huge delay when fuel tanks run out. I wonder what it's doing...

[vexx32]

It'll be updating the physics again for the whole craft. With so many parts, when a fuel tank runs out, the CoM changes somewhat, and it is forced to recalculate it. Imagine how slow it'd be if it calculated the CoM in realtime as fuel drained xD

[EndlessWaves]

Total ingame flight time was 1:27.00. Time on my iPhone 3:40.00.

You're not measuring the same thing there, that's simulation speed rather than framerate and there's no guarantee their lower limit is tied together (their upper limit obviously isn't, the game doesn't run twice as fast when the graphics are rendering at 60fps as opposed to 30fps).

It may actually be a better test as it seems to be generally agreed that CPU is the limitation for large rockets rather than graphics.

[Shortsonfire79]

You're not measuring the same thing there, that's simulation speed rather than framerate and there's no guarantee their lower limit is tied together (their upper limit obviously isn't, the game doesn't run twice as fast when the graphics are rendering at 60fps as opposed to 30fps).

On a lag-less rocket, the simulation speed is generally in sync with real time speed. Eg: 3 minutes of simulation time is 3 minutes of realtime. (granted you don't timewarp). The frame rate for my rocket was 1frame for ~3 seconds (counting manually, don't have Fraps), and it was apparent with the simulation time being almost 1/3 the realtime.