poly counts for next-gen games

[Headless]

There's a thread on the modelling forums about next gen poly counts, and another one on general chat.

The PGR3 counts (which are confirmed in the Visuals video here (http://media.xbox360.ign.com/media/741/741362/vids_1.html)), seem a little high to me, so either they have some good level streaming technology or something, or they're giving us pre-normal map counts. If someone works at or knows someone at Bizzare Creations then if they could shed some light on this i'd be interested to hear what the story is with these.

[SHEPEIRO]

i think you might be right on the normal map front. 40,000 per car seems sooooo high,
but who knows.

this is the time when wild exaggerations will be thrown around

[SHEPEIRO]

actually looking at the video perhaps 40,000 is correct if thats exteroir/interoir, and driver.
i think the pre normal map model would be closer to 400,000 with the written indents. like the one on the wing mirror at begining of vid.

[Headless]

I'm more surprised by the environment than I am by the car counts. 40,000 may seem alot compared to the 5000-15,000 poly's that people are touting for net gen game characters, but if you think about it, you're unlikely to get more than say, 4-5 cars on screen at any one time, if that, where in an action game you might be pushing closer to 10-15 characters on screen at once.

[AdamAtomic]

Actually the model for the bridge probably is 600,000 triangles, but it is probably constructed in segments, only one of which would be displayed at the highest or full LOD. The rest will probably be much less. 10% and 50% are common LOD sizes in most next-gen game engines I think? So while there may be a city full of 5,000 triangle buildings, they are usually only about 500 triangles until you get right up close. Likewise, say the bridge is made up of 10 60,000 triangles segments, the most likely case is you'd have one segment at 60,000, a couple at 30,000, and the rest at 6,000 or less. So instead of 600,000 triangles just for one bridge, you'd have more like 130,000. I'm sure they cut it down more than that, but that's just a sample of how even simple, conservative LOD usage can cut down what you draw on screen by a HUGE amount (80% or so in our little example).

[kaylon]

As stated in all the other threads on this topic...it's all dependant on the genre of game.

PGR3 has 80000 per car (40k on the outside, 40k on the inside) and upto 3k to 4k per average building...the stars of this game are the cars...large objects like the bridge, a location people know will have a high count at LOD1 as they will basically take up the same space as a few blocks of buildings.

You cant get out of your car and look around...so the backs of buildings etc will be huge polys and probably look like crap. Most of the detail in PGR3 comes from the 1024 photos they are using for thier base materials...it looks photoreal cos they ARE using Photos...how clever huh (sarc off)

On foot action games will be very different. They still will have to have semi low counts for peds and filler vhehicles etc. It all depends what are the features of the product and what tech is being used.

If people are still using renderware then they still have to be very clever cos renderware has not gone next gen. Even source is not really next gen, Unreal 3 is getting close to next gen.

That's the main prob at the moment when asking about stats and "My character is 20000 polys, he's for an PS3 game" ... sure the PS3 can probably handle that (btw next gen consoles still cry like a little girl about large textures)...but the majority of engines can't, and it's the engine that drives a game. You should forget about the hardware and look at what software you are gonna be using...cos the hardware is next gen now...but the software is not yet, it's still the same old stuff as it was last gen. Dev companies have not had enough time on the new hardware to do anything next gen yet...we only got our beta hardware a few weeks ago...maybe next year we will see the real next gen games.

K.

[Headless]

If people are still using renderware then they still have to be very clever cos renderware has not gone next gen.
I thought Renderware 4 was supposed to be next gen?

[kaylon]

Might not be known by many... renderware 4 is dead. Might not even be public knowlage in which case I may get into trouble...but what the hell, Criterion does not basically "support" RW 4 anymore ... it got broke so bad they wont fix it...I mean, it does not even do double sided materials or support morph target animation. I'm not sure what EA has in line for renderware but it sure aint a fix for 4 ... sigh ... and as far as next gen.... it's about as Next gen as Pac Man umm.. maybe Donkey Kong ....

K.

[TheShow]

Actually the model for the bridge probably is 600,000 triangles, but it is probably constructed in segments, only one of which would be displayed at the highest or full LOD. The rest will probably be much less. 10% and 50% are common LOD sizes in most next-gen game engines I think? So while there may be a city full of 5,000 triangle buildings, they are usually only about 500 triangles until you get right up close. Likewise, say the bridge is made up of 10 60,000 triangles segments, the most likely case is you'd have one segment at 60,000, a couple at 30,000, and the rest at 6,000 or less. So instead of 600,000 triangles just for one bridge, you'd have more like 130,000. I'm sure they cut it down more than that, but that's just a sample of how even simple, conservative LOD usage can cut down what you draw on screen by a HUGE amount (80% or so in our little example).

I agree with this - I'm doing some env modelling now for some PS2 / XBOX titles and we have to split all that stuff up into segments. The entire mesh poly totals are much higher than on screen totals.

[RmachucaA]

anyone ever think that the game engine may do auto lod's? i can tell you at least, that PGR2 had it.

[Confracto]

there are games that do that, I know Dawn of War did, it was described as being similar to Max's optimize modifyer.
...much easier than doing a whole bunch of extra artwork...

[SHEPEIRO]

what maxes optimise modifier.

i hope not thats a pile of pooh.

much better to have artist created LODS

racing games can have alot of lods, streaming off memory, because of the linear naturwe of the tracks, in an adventure game where you can travel in any direction the lods have to be much more flexible and therefore not as streamlined.

if you think about a gta city compared to gran tourismo you get the idea, it will be the same wid next gen, although the greater amounts of polys that are thrown around will affect the graphics quality of non linear games more than linear racing games. i.e the difference between pgr2 and pgr3 will be much less than say the difference between gta san andreas and gta5,6 or whateva

[inneractive]

Here are some PGR3 shots with wires in XSI. Plenty of polys.
http://projectgothamracing3.com/screenshots/default.htm?c00=1

Got the link from XSIBase.com news.

[SHEPEIRO]

those must be pre normals.

there are so many wasted polys.

i know these new machines will throw alot of polys around but there is still no need for wastage at that skale.

no neeeed

[Confracto]

much better to have artists create LOD's?!?
are you crazy?
I played a bit of hte game, and specifically looked to see if I could notice them getting less detailed, and I couldn't...it was really well done, and likley a LOT faster than making tons of LODs for all the units in the game...

if you think about it objectivly, like art vs. time, and time is money spent...it makes more sense to do it automatically.

[SHEPEIRO]

hmmm ok.
i was commenting more on the fact that the edge place ment on maxes optimise mesh modifier is pants at low res. but durrrr we are talking about high res where it will be alot lesss noticable will have to look at dawn of war.
was also thinking about flight unlimiteds awfull landscapes that used to look like teh sea when close to.
you are right about automatic the way forward for the fuiture. with 5million squillion poly meshes flying around you dont want to have to model all their LODs,

[AdamAtomic]

That's true to a certain extent - however, artist-generated LODs are usually very fast, maybe 1-2 hours tops per model even without special tools. And many games would rather use the cycles dedicated to dynamic LOD for something else! I think it probably depends on the game a lot - racing games, as people have already stated, are much more predictable than say a third person action game, you can buffer and pre-load and stream things much more easily, and I suspect that makes a big difference in how fast you can dynamically adjust the LOD on the scenery.

[SHEPEIRO]

there will be alot more stuff generated on the fly with the new consoles as theyre prossessing power completely outweighs theyre mem capacity. this means that fractal generated water, landscape and vegetaion could be much more viable than with the last gen. chech out the Avalanche engine from Avalanche studios. havnt got link to any images but seen theyre engine on mega pc, phew its nice.

[vrapp]

Funny you should mention the Avalanche Engine, since that's where I go to work everyday. Currently I work on our current gen game Just Cause though. :)

Also, I do belive that the PGR3-wires that were linked to earlier are in-game models and not pre-normalmapping models.

[SHEPEIRO]

cool man i would like to wotk there,im jealous. the engine looks cool. are you using a version of it for Just Cause it looks quite similar. looks like a really interesting project. esspeccially that next gen engines.
i was thinking when i saw it of a detailed wildlife simulation ive always wanted to play. sommit like a bird sim with flys/ fish/ rodents to catch. hot summers days, treehouses, paddling in ice cold streams...........and back to reality.

[Omita]

I think over all polycount will be lower ingame as normal mapping is very expensive GPU wise (passes). But until people have actual Xbox 360 and PS3 dev kits it's all theoretical numbers. The machines are so powerful that you can have crazy polycounts if you do sectional LOD...

In my opinion, no one should worry about how many polys, its only a worry about can we afford and get all that work done. What good is a million poly bridge is the game doesn't ship? Nada.

I think Nex-Gen is going to back-fire and really hurt the industry. Or at least... be the end of small studio's. My silver lining is that procedural art can be used more, but the reality is that only big studio's with huge art libraries are going to be able to stay on budget and afford the huge bills of NexGen titles. Publishers are already pretty concerned about the staggering budgets of NexGen titles. A NexGen title that flops could put a company 12 million in the hole or even more. I think this will be the end of small developers, at least on consols. Which will end up causing a smaller number of titles for audiences. Of course, casual games and smaller titles might spin off but they won't be in the same class as future Halo's, GTA's, and Half-Life's.

Now... I hope that Western society (which has always been obsesed with Photorealism) will realize that much cooler things can be done with art then mimic reality. Doubtful but my dream. I think Katamari Domaci should be the future. Fun games that are visually unique and less expensive to make. There isn't much game in plot line driven "play the movie" games. Cardinal sin games like GTA are also pretty un-imaganative. I think games should be games... not reality in a box.

-Hays

[membreno-david]

Buddy of mine @ EA Sports modeling heads for FIFA 06 for XBOX 360 says that they got 4k for the head, then 4k for the hair (depending if its bald or not)....

[Headless]

Omita: I don't necessarily think that small developers are done for, nor do I think that you have to sacrifice photorealism, I just think developers have to get smarter about how they go about making games. Using procedural content, as you mentioned, is just one example.

Sure if developers continue to make games as they're doing at the moment, then yes, they're going to get these massive budgets and long development times, but if you look at something like Will Wright's 'Spore', I don't see any reason why you couldn't do that game with 20-30 guys over a 2-4 year dev cycle.

Alternatively the industry might be able to solve quite a few of their content problems by looking into photogrammetry.

[membreno-david]

Omita: I don't necessarily think that small developers are done for, nor do I think that you have to sacrifice photorealism, I just think developers have to get smarter about how they go about making games. Using procedural content, as you mentioned, is just one example.

Sure if developers continue to make games as they're doing at the moment, then yes, they're going to get these massive budgets and long development times, but if you look at something like Will Wright's 'Spore', I don't see any reason why you couldn't do that game with 20-30 guys over a 2-4 year dev cycle.

Alternatively the industry might be able to solve quite a few of their content problems by looking into photogrammetry.

Wasn't Spore supposed to make artists a thing of the past, and rely on its game engine to produce content according to the choices the user makes? I hate the idea of having to play a game where it's all gameplay and no art. Where it's all programmers behind it all. I've seen some screens of Spore and it looks like crap. I plan on boycotting Spore, thank you very much. P.S. What does remote sensing have anything to do with poly counts? Or are you talking about about photo realism? Ah in any case, photo realistic graphics aren't going to be around til PS 6 (if that) my friends...

[SHEPEIRO]

the thing about photorealism is that it will be a itch that just has to be scatched until its acheived. once it is consumers/publishers will need to do other things to be different. as with the renaisance in Painting, it expressionism wasnot deemed important until realism had been acheived, once the market had been saturated with it, new ideas will emmerge as these will be the ones in demand.

dont worry too much our industry is unfortunatly in for difficult times but their will be light at the end of the tunnel.

its going through a particlarly expensive and moody puberty thats all.

[Headless]

Wasn't Spore supposed to make artists a thing of the past, and rely on its game engine to produce content according to the choices the user makes? I hate the idea of having to play a game where it's all gameplay and no art. Where it's all programmers behind it all. I've seen some screens of Spore and it looks like crap. I plan on boycotting Spore, thank you very much.
You seem to be getting your facts mixed up a little. They're not doing away with artists as such, they're essentially creating really easy to use tools and bundling them with the game as part of the game so that the users create the content, but it's done as more of a step up from how players created their own house in the sims, they're just doing it all with procedural methods. The idea is that players create their own little creatures or towns or whatever, using the in-built tools, then new races from other players are downloaded to fill in additional game content. They still need artists to create a core set of content to come preloaded on the disk, so that users who don't have internet connections can still play, and so that there's content there on day of release before players have created their own.

It's kind of tough to explain in a way that lets people get onboard with the idea, so watch this (http://www.pqhp.com/cmp/gdctv/) video (which you'll need to register for but registration is free), and i'm sure it'll change your mind.

Oh and if you're all pissy cause you feel threatened that procedural content is going to do away with game artists then don't be, cause 99% of games in this next generation are still going to use the usual methods.

P.S. What does remote sensing have anything to do with poly counts? Or are you talking about about photo realism? Ah in any case, photo realistic graphics aren't going to be around til PS 6 (if that) my friends...
I'm talking about potential ways to solve the issue of having to create masses of high detail content for next gen games. My reference to photogramettry was that I was suggesting that with some work, photogrammetry techniques could be used to create photorealistic game environments. Check this (http://www.debevec.org/Movies/facade97-campanile-prolog.avi) video to see what I mean, and check here (http://www.debevec.org/Movies/debevec-campanile.mov) for the results (bear in min these are from 8 years ago. There's been some advances since then).

[JaMo]

if games are going to be HD next gen which means more triangles in the cameras view to be render plus you would need i think at the very least least a texture map of 1024 for stuff the camera gets close to without it looking like crap at those higher resolutions.

so with all this extra stuff needing to be pushed for HD are next gen systems capable? i mean on a human type character you would need a 1024 map on the face and 1024 map map for the rest of the body. i just cant see pushing alot of these size texture maps around the screen but it seems they would have to not to lose the textures quality. so i guess my concern is less with poly counts and more with texture sizes which imo are more important. anyone know what sizes can be pushed and how many at a time?

oh another limitation i see would be loading the textures into the ram. 1024 maps are usually pretty big but maybe im just dumb about all this.:argh:

[Headless]

You're not being dumb at all; texture sizes are certainly going to be an issue with next gen consoles. Yes they'll take a step up but texturing is still going to be an area where developers are going to struggle, just as it is with current gen.

For Unreal Engine 3 (which I like to use as an indicator of next gen counts), Epic have said that they're creating texture and normal maps at 2048 x 2048 for PC's, and scaling that down by 2x (1024 x 1024) for next gen consoles, and 4x for low end PC's. They say that that depends on texture counts and scene complexity though.

Just to note: While Microsoft's marketing would have you believe that "the HD era", is about amping up the whole gaming experience, in reality the next gen push for HD is just about running games at a higher resolution, it's not about polycounts or texture size. Yeah those things will get better too, but really the only real difference of switching to HD is that you'll get less jaggies.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.