Skip to main content
gamedev.net
🔒 Locked

Performance Question

Started by fathom88 Oct 14, 2005 at 12:57 PM 13 replies 750+ views
Original Post
fathom88
fathom88
I'm drawing a 2D scene. According to my simple bench marks, I'm drawing five million triangles to the screen. They are only RGB colored triangles. There is no special light effects or anything like that. I'm using vertex arrays to draw my image. It has vertex arrays and color arrays enabled only. I'm not doing any transforms. Simply drawing the triangles to the screen. It takes roughly 20 milliseconds to draw the scene. Is this a good value? Or is there a lot more room for improvement. I've tried display lists and VBO already. Are there any other performance tricks out there?
Sneftel
Sneftel
That's a pretty good framerate. VBOs may help out if properly hinted, though it depends on what else you're doing in your program.

What is the source of the triangle data? The specifics of your application will influence which optimizations are open to you.
evanofsky
evanofsky
You do realize that 5 million triangles is a beepin' lot of triangles, right? [grin]
etodd.io | @etodd_
vNistelrooy
vNistelrooy
Quote:
Original post by fathom88
I'm drawing a 2D scene. According to my simple bench marks, I'm drawing five million triangles to the screen. They are only RGB colored triangles. There is no special light effects or anything like that. I'm using vertex arrays to draw my image. It has vertex arrays and color arrays enabled only. I'm not doing any transforms. Simply drawing the triangles to the screen. It takes roughly 20 milliseconds to draw the scene. Is this a good value? Or is there a lot more room for improvement. I've tried display lists and VBO already. Are there any other performance tricks out there?


Good or bad is relative to your system. What CPU, GPU?
"C lets you shoot yourself in the foot rather easily. C++ allows you to reuse the bullet!"
fathom88
fathom88
>>What is the source of the triangle data? The specifics of >>your application will influence which optimizations are open >>to you.

It's a flat file whose data I turn into x,y coordinates. I get snipets of it every so often. Hence, I can't do it all at once and load it into the VBO. I tried the VBO, and it was slower because I had to constantly update my data to the video card (even with the DYNAMIC load flag).
fathom88
fathom88
I'm running a 1.7 Pentium M with a GForce Go 6200 (it's a laptop as you tell from my specs). How many triangles is a high end game like DOOM 3 pushing? Just curious. DOOM looks a lot better than a bunch of colored triangles :)
evanofsky
evanofsky
It just occured to me that you might mean it's drawing 5 million triangles per second, in which case it would be pretty slow. Like vNistelrooy said, we need some more information before we can help you.

[edit:] Whoops, ya beat me to it.
etodd.io | @etodd_
fathom88
fathom88
Oops. I made a big mistake. It's 1.4 million triangles in 20 millisecond. Sorry about that. That's what 70 million triangles in one full second. Please feel free to correct me. As you can see, I'm accident prone :)
zedzeek
zedzeek
impossible to tell without knowing what youre doing
(guesstimate figures which i believe are pretty close to reality)
on the same system
doom3 does 5mil tris/sec
quake3 does 50mil tris/sec

fathom88
fathom88
>>doom3 does 5mil tris/sec
>>quake3 does 50mil tris/sec

Shouldn't it be the other way around?

DOOM 3 => 50; Quake 3 => 5

Or do you mean Quake 4? I figured the models in DOOM 3 would be composed of more triangles because it looks so much better. Or is this an effect of the bump mapping etc. ?
evanofsky
evanofsky
Doom 3 used bump-mapping to the point of exhaustion. The models themselves were actually pretty low-poly. And zedzeek was right, because Doom 3 does more computations per triangle (what with all the bump-mapping) than Quake 3 does, because Quake 3 didn't have dynamic lighting and all the nifty features Doom does.
etodd.io | @etodd_
fathom88
fathom88
Is it just me or do the models in D3 look better than the ones on HL2?
evanofsky
evanofsky
I don't think so. HL2 had much better models and lighting, except for the lack of real shadows. grrrr....
etodd.io | @etodd_
zedzeek
zedzeek
Quote:
Shouldn't it be the other way around?

DOOM 3 => 50; Quake 3 => 5

no though i might of exaggerated the difference
doom3 is pretty low polygon count
most of the characters seem to be around the 1000tri count
i forget what quake3 was but they wouldnt be much less than that
the diffence is your computer runs quake3 at >5x the fps of doom3
hence the higher tris/sec

all im trying to say is its not the number of tris u use but more what u do with them that impacts most on the visual quality

(passes per character will be roughly similar)

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.

Sign in to reply to this topic.