Kyro 2 - Is it a crutch?

[ummwhatever]

Why does the quake 3 engine run so much faster than others?

Here's my question, feel to argue or contradict anything I say - but please let's talk about engines not what game is better.

Many will argue that Tribes 2 and Serious Sam have many computations to make and has to render large sceneries.

This is very true.

But is it wasted effort?

Is it pre-rendering when it's not necessary; i.e. too much overdraw?

Is it slower because Quake is an 'indoor environment' and Tribes 2 and Serious Sam have an 'outdoor' environment?

What about Team Arena? It has Terra Maps that are outdoors.

Why do Team Arena Terra Maps run so well with the Quake 3 Engine?

Is it due to the following?

"PQ: id has said that the Quake III engine has always been able to support large maps like the terrain maps. Why haven't we seen them until now?

Jaquays: All the maps done for Q3A had to fit inside the memory allowed to us by a computer with 64 meg of memory. During production 0f Q3A, it became obvious that really large maps, and Q3DM12 was the "envelope pusher" in that product, ran into memory issues and did not play as well as smaller maps. During the production of Q3:TA, the programming team found numerous ways to optimize, improve and streamline memory handling to the point where it was possible to do much more with the game"

http://www.planetquake.com/features...s/jaquays.shtml

Where the Kyro 2 is spouted to boost the Tribes 2 (hearsay) and the Serious Sam (Anandtech) engine and it very well may with its Tile Based Rendering:

http://www.anandtech.com/showdoc.html?i=1435&p=3

It offers little improvement with Quake 3 over the Geforce GTS and none at all over the Pro:

http://www.anandtech.com/showdoc.html?i=1435&p=10

Why is this?

It's probably because the Quake engine does the same thing, in the engine itself:

http://www.bluesnews.com/abrash/chap64.shtml

It's called Visible-Surface Determination.

It's nearly the same process, yet built into the engine itself.

Hence, any engine developed with VSD - whether it's based on outdoor or indoor environments, will simply surpass an engine that is not, especially if the video card is simply faster - memory and rendering wise. It will also make a Tile Based Rendering card less beneficial.

Do other engines incorporate VSD or similar?

Team Arena has outdoor (terra) maps that pretty much remain consistently the same FPS in most locations, they don't jump around from 30 to 100 to 50 (such as in Tribes 2) standing in the same spot - this is due Visible-Surface Determination within the engine.

Since Quake engine, for the most part, is only drawing what you see, from what I've been reading - does it take the place of a need for Tile Based Rendering?

Is this true?

Is the Kyro 2 HSR-TBR merely a compensation for bad coding and lack of VSD in engines?

Does it help due to lazy programming?

Is the Kyro 2 a crutch?

 

[ahandinthebush]

hmm... interesting. i am awaiting ben's reply to this post...

 

[BenSkywalker]

I'll try not to get too long winded

VSD is actually used in all games, the question is the level of effectiveness. Using VSD/HSR requires CPU cycles, it is possible to nearly eliminate overdraw if you want it to, UnrealTournament is a good example of this, but that game is also extremely CPU limited. Carmack's engines tend to have a very good blend of reducing overdraw without the heavy CPU hit that games like UT suffer, however the overdraw is still definately there.

If you break down the raw fill of the Kyro2 it should be roughly equal in performance to a V3, and handily bested by a GF2MX in Quake3 if there was absolutely no OD(even UT has ~1.5, Quake3 is probably closer to 2.5-3 most games would likely fall in the ~3-4 range).

If the question is can you do the same thing in software the Kyro does in hardware the answer is yes, but it won't be anywhere near as fast. Determining the particular visibility of each pixel would absolutely throttle your CPU, so much so that it is definately not worth it.

The Kyro2's approach certainly isn't cared for by everyone(Sweeney most notably and vocaly), because the way in which it works can lead to potential problems(PVR advocates say that isn't so, but none of them will say how to get around it). Because you are binning all of the geometry data in advance as your geometry loads rise, your tile buffer needs to either increase significantly in size or you need to reduce the amount of pixels that a single tile will handle. This definately isn't a good idea for massive poly loads. As of right now, no game is close to pushing the limits of what the K2's throughput seems to be(based on actual geometry benches, not theoretical specs), but upcoming titles like Unreal2 and Doom3 will be in that area and it may become a problem for the K2. Perhaps not, but as of right now benches are showing the K2 stuck in the ~4M Poly/sec range which is definately well below that of any recent traditional save the V5(which was AGP limited to ~5-6Mil polys/sec) with or without hard T&L.

There are other approaches however. Right now, the true performance potential of the GF3 isn't close to being realized, and that is ignoring all of its' new vertex/pixel shaders and everything about nFnite FX. The GeForce3 is built so it can perform the VSD calculations in hardware as long as the application renders front to back. It does this very early in the pipeline and while it won't match the effectiveness of the TBR, it also has ~500% more raw texel fill, it can be half as effective and it will absolutely obliterate anything on the market(think, two to three times faster then the Ultra running the same settings). This isn't a proprietary nVidia technique, it is free to use for any of the companies and likely will be for the traditionals that we will be seeing this year.

So then software support becomes much easier. You still want some VSD or you'll chew up too much bandwith and you won't be able to run on older hardware, but you can lessen the amount of time spent on it considerably as a decent hardware base will have support for it(expect mainstream games support by mid '02).

Tribes2 I'm iffy on. I really have to get around to picking up the game, I'm not certain that it is a flaw in the game code(ie- it may not be sloppy coding, though with the amount of bugs I wouldn't rule it out). With outdoor games it can go two ways, if you have relatively flay terrain then a traditional can clean up because it has more raw MTex/MPix fill. If it is a very hilly terrain then the amount of OD removed by a TBR can be very significant.

 

[TravisBickle]

why not let every single game developer rewrite tile based rendering.
let's reinvent the wheel. sucks to OOP too.
stupid.