The performance of the Spring engine

[reivanen]

I decided to do some testing with spring since i have been wondering for a long time that spring seems to be running slower then what you would expect from a game like this. Could some dev explain what is up with spring, and why does it perform so bad ?

I made a setup with crysis: 28 units moving + some more tanks just standing there as i was running out of space on the spot i chose. Remember that this is a FPS engine and is not designed for a view like this. At ground level i got steady 60fps at my test scenario.

Then i opened spring, spawned some 30 bulldogs and made em move and measured the fps. Remember that this is a RTS engine.

I will let the pictures speak for themselves.





Crysis
14 tanks, 3 cars, 11 hovercrafts moving with AI + 22 tanks onscreen standing standing still. FPS = 37


Spring
29 tanks moving FPS = 44

And as a last image here is what crysis looked when you zoom up-close:

[Vadi]

Nice one there, but buddy, think a moment about what are you comparing (and why, and how).

Also come back to test the next release, which'll have multi-threading in and such will perform a lot better on 1+ core systems.

[Hobo Joe]

It's not secret that the Spring engine is very poorly optimized.


However, from what I hear, the next release is supposed to make big steps in that direction, so keep your fingers crossed!

[Vadi]

a) You're comparing a game that's reported to have a budget of $27 million, against a game with a budget of $0. That's a 27million difference, right there!

b) You don't play Spring with 29 tanks. OK? You just don't. You play Spring with 500 tanks all moving, fighting, getting build, and flying about. Do you play Crysis with 500 tanks? Give that comparison a try, because that's what Spring is designed for. Don't forget the 16 players you can have at once, each with their own army.

c) Why are you comparing a mostly RTS game with a mostly FPS one?

d) Unless you look at the code, you can't say how is it "optimized". Multi-threading support isn't considered an "optimization", it's a feature.

[Acidd_UK]

Looking at that image, what is your ground decals setting at?

[hrmph]

a) You're comparing a game that's reported to have a budget of $27 million, against a game with a budget of $0. That's a 27million difference, right there!

b) You don't play Spring with 29 tanks. OK? You just don't. You play Spring with 500 tanks all moving, fighting, getting build, and flying about. Do you play Crysis with 500 tanks? Give that comparison a try, because that's what Spring is designed for. Don't forget the 16 players you can have at once, each with their own army.

Those two points sum it up perfectly.

I assume crysis becomes horribly lagged if you spawn anywhere close to as many units as a typical spring game. Even 1v1s usually have 50+ units.

[Hobo Joe]

I assume crysis becomes horribly lagged if you spawn anywhere close to as many units as a typical spring game. Even 1v1s usually have 50+ units.

Well, naturally. The poly counts are in excess of 5k on those tanks, I'm guessing. Probably around 10k though.

The poly count on bulldogs is probably like 2 or 3 hundred.

[reivanen]

First of all, i was not complaining i was asking.

And no, it won't start lag even if i set 100 tanks in there.. probably not with 1000 either. There was no performance difference before or after placing the models or making them move. If you will keep missing the point, i could redo the test with a few hundred tanks...

[FLOZi]

Another thing to note is that Crysis was developed with 8000 series cards in mind, whereas spring consistantly has had efficiency problems with the new cards.

[rattle]

i set 100 tanks in there.. probably not with 1000 either.

Without LuaUI I have a constant 100-170 fps with 500+ low poly units on simple maps without water and fairly high settings (no grass though). About 30-50+ fps with water 2. Really depends on what is visible.

You can't really compare a game that was made with DX10 and next gen cards in mind to Spring... well what Flozi said.

[NKrypt]

d) Unless you look at the code, you can't say how is it "optimized". Multi-threading support isn't considered an "optimization", it's a feature.

I would love to see most people here attempt to write multi-threaded applications. In fact I'd challenge most to write an incredibly simple single threaded application!

[Satirik]

Looking at that image, what is your ground decals setting at?

+1

[Felix the Cat]

So Spring isn't as pretty as Crysis?

Uh... very few games are as pretty as Crysis.

[AF]

Simply put:

Its not that spring wasnt designed for 8800 cards, spring wasnt designed at all, it was just hacked together until it worked.

Even SJ the creator almighty said its just a nehe GL tutorial he went crazy with.

[SwiftSpear]

Your comparison doesn't really show anything. The spring scene has more terrain being shown. Spring has a very inefficient terrain mip handler, hence the reason why it was rewritten for the SM3 format. You've also got thousands of decals being rendered in the spring scene, each of which forces the renderer to run more passes, decreasing FPS, and finally your spring units are pathfinding, effectively, taking a chunk out of your CPU performance as well. Your crysis units aren't doing anything at all. It's impressive that spring is able to beat the framerate of that crysis scene considering we don't have hundreds of programmers and artists to endlessly re optimize the models and mip coding.

The crysis engine might be a performance hog. but it's also build to handle large scenes. It has very good teiring of model and terrain quality as it gets far out and doesn't need to be seen any more.

[AF]

You'll also note a recent svn commit ( IIRC made by kloot ) made changes to the terrain drawing code that cleaned up the formatting and introduced some speed-ups, however I cannot comment on whether this is a conceptual minor speed up, or something that actually changes the frame rate.

[smoth]

currently for me, without anything moving I get between 20-10 fps

[AF]

My ATI integrated laptop graphics x1100 express gives 15-25 fps in spring yet homeworld 2 looks fluid.

[Vadi]

There's also an amazing thing called 'video settings' that can be observed to increase performance.

That, and the rest comes down to renderer, drivers & the OS you're using.

[FLOZi]

currently for me, without anything moving I get between 20-10 fps

I get 80ish.

YOUR DOIN' IT WRONG