Raw power has hit its upper limit already, thats why we've seen a shift to parrelism, and programmable shaders doing neat effects for us taking up transisters rather than a real icnrease in power. Raw poly pushing power isn't gonna improve at the same rate it has historically, and to be honest I'm disapointed. It's like we gave up too early. Shaders et al have their place, but gimmie juice!
wtf is up with poly counts
- Thread starter GwymWeepa
- Start date
And also people seem to forget that all the "things flying around when i smash my car into a building" are all geometry. And the more you want, the more geometry you need. Not to make them look pretty, but to make things move and break and fly around and bounce more realistically.
Lots of games need LOTS of characters or THINGS on screen at once. All geometry.
Realistic cloth, hair and water simulation... All geometry. Not to make them look pretty, but to make them move properly.
Organic models such as humans and trees, you only get the non-videogame-look by adding geometry. Not necessarily to make them look pretty, but to make them move properly.
Same difference between Tomb Raider1 with Lara's head made up of a bunch of polygons with a "face" texture there, and FFTSW. Obviously we get into "overkill" kind scenarios, and no one will see the difference from a 1M poly model and a 1.2M polys one, but we're still VERY far off from that level, and spiky knees TODAY are embarrassing.
Lots of games need LOTS of characters or THINGS on screen at once. All geometry.
Realistic cloth, hair and water simulation... All geometry. Not to make them look pretty, but to make them move properly.
Organic models such as humans and trees, you only get the non-videogame-look by adding geometry. Not necessarily to make them look pretty, but to make them move properly.
Same difference between Tomb Raider1 with Lara's head made up of a bunch of polygons with a "face" texture there, and FFTSW. Obviously we get into "overkill" kind scenarios, and no one will see the difference from a 1M poly model and a 1.2M polys one, but we're still VERY far off from that level, and spiky knees TODAY are embarrassing.
I'd rather have the 40k poly model over one that is 4k and normal mapped. Have you seen the pics of that one creature, the one featured from the very beginning for UE 3.0, the large dinosaur looking things with guns and chains? WTF, why are the chains cheaply painted on? That looks so f'ing terrible. I need lots of polys if a character needs it. I imagine that a fighting game will look stupendous, but a lot of the other screens are disappointing me and I do not believe I'll feel any different until a generation of software goes by...which is sad I want to be wowed immediately...I mean these machines are monstrous after all.
Now, a year or two from now things will be different...I can't wait to see what Square is doing, we will see polys galore...call me selfish but I wanted them first gen (well, we'll see, since most of the shots of so called next-gen games have been running off UE 3.0, I'm still crossing my fingers for other first gen titles to surprise me).
Now, a year or two from now things will be different...I can't wait to see what Square is doing, we will see polys galore...call me selfish but I wanted them first gen (well, we'll see, since most of the shots of so called next-gen games have been running off UE 3.0, I'm still crossing my fingers for other first gen titles to surprise me).
Yeah i agree. Chains are chains, and not only they don't look right on the UE3 monster, being 2D, but because they're just painted on, they don't MOVE. At all. That's just not good enough.
If something is 3D in reality, it should be 3D in a game, or people will notice quickly just by moving either the object or the point of view.
If something is 3D in reality, it should be 3D in a game, or people will notice quickly just by moving either the object or the point of view.
pc999 said:Remember that in motion, low-polys meshs dont look so bad, in a ss is easy to see things that way and we already see that in corrent gen games.
BTW everyone lost the faith in MRM, processural detail for next gen (high-polys near to you and low-polys far from you)
LOD? It's being used forever and will be used for a long time too.
london-boy said:pc999 said:Remember that in motion, low-polys meshs dont look so bad, in a ss is easy to see things that way and we already see that in corrent gen games.
BTW everyone lost the faith in MRM, processural detail for next gen (high-polys near to you and low-polys far from you)
LOD? It's being used forever and will be used for a long time too.
has the Xenon GPU tile base rendering(defferend render like Dreamcast)??
gamepower said:
Not as far as i know. Besides, these days, GPU seem to be a minestrone of different old technologies, among which deferred renderers. Why?
For what reason? So that when there are hundreds/thousands of such models in the scene, you get a warm, inner glow knowing that when your card drops to 10fps it's finally doing some real work? :?GwymWeepa said:
Seriously though, your comment was a bit of a sweeping statement. I'm sure there are plenty of models that will look just fine with the 4k+normal map system.
Simon F said:
That's not the point.
You can have LOD for those kind of heavy scenes.
Personally, i'd prefer my VF5 models to have as many poygons as possible, mostly because whatever's 3D will move like it's 3D.
I'd want my model to have real chains, even simple chains, not made by 8000 polygons each, just enough to make them look and move like chains. NOT a shiny normal map attached to the body like it was painted over it.
ShootMyMonkey
Veteran
Polycounts are still going up, mind you, even though it may not be that much. In current gen games, with certain genre-specific exceptions, characters rarely go above 5200 polys... especially on multi-SKU titles. With UE3, 6000-8000 is not out of the question.
One thing you have to realize, (and while this is fairly certain on Xbox2's part, I don't really know about the other two), the shift is less towards increasing raw fillrate massively, but more focus is put on shader power. Being able to execute longer, more complex shaders is basically where the power is going. More importantly, that increased range means more customizability of the overall "look." That's not to say we won't be able to process huge numbers of polygons anymore, but more that the increase in fillrate over current GPUs is not as large as you might expect.
The downside is, of course, if you're going to be loading up on shader instructions, you can't process a million polys. If it was something like DOA3 where all the characters are basically using vertex lighting and completely fixed-function rendering... well, then, you can just load up on polygons as much as you want so long as you're not saturating bandwidth.
Also as far as the whole on-the-hardware support for subdivision surfaces. No real clue what to make of that. Nobody I know of has beta hardware, so there's no real way to test a feature like that. I'm not even entirely sure if that rumor is worth taking seriously. So even if some developer out there knew for sure that the support would be there, the best you can do for now is render screenshots without it. For that matter, if you knew for sure that there would be SS support, I don't see why you'd build control meshes that are even close to current-day polycounts -- you could afford to go far, far lower...
One thing you have to realize, (and while this is fairly certain on Xbox2's part, I don't really know about the other two), the shift is less towards increasing raw fillrate massively, but more focus is put on shader power. Being able to execute longer, more complex shaders is basically where the power is going. More importantly, that increased range means more customizability of the overall "look." That's not to say we won't be able to process huge numbers of polygons anymore, but more that the increase in fillrate over current GPUs is not as large as you might expect.
The downside is, of course, if you're going to be loading up on shader instructions, you can't process a million polys. If it was something like DOA3 where all the characters are basically using vertex lighting and completely fixed-function rendering... well, then, you can just load up on polygons as much as you want so long as you're not saturating bandwidth.
Also as far as the whole on-the-hardware support for subdivision surfaces. No real clue what to make of that. Nobody I know of has beta hardware, so there's no real way to test a feature like that. I'm not even entirely sure if that rumor is worth taking seriously. So even if some developer out there knew for sure that the support would be there, the best you can do for now is render screenshots without it. For that matter, if you knew for sure that there would be SS support, I don't see why you'd build control meshes that are even close to current-day polycounts -- you could afford to go far, far lower...
On the grounds NURBS have been tooted for ages and feature on PSP, it'd be daft not to include them on next-gen for all their benefits if you've got the number crunching power to abuse such technology.
I think bandwith is undersold and long shaders are oversold ... sure for PCs you have to go that way, because eDRAM isnt possible. For consoles though with potentially plenty of bandwith the balance between shaders and raw throughput should be different (unfortunately Sony of all the players seems to disagree with me for the next gen, at least if they forego eDRAM ... ironic).
All renderes today are deferred (They build up a scene in steps and then display it when it's finished). And almost all renderes have some kind of fillrate saving and polygon culling/early Z check either in software or hardware or both.gamepower said:
But tile based rendering? If the 10Mbs of eDRAM on the VPU in xbox 360 holds true, then probably not. Maybe it can split the screen, SLI style, to reach higher resolutions. But if it has good AA, then 720P should be more than sufficient.
Simon F said:
london-boy pretty much covered my thoughts on the matter. There are times where lower poly models with normal mapping are just fine, and other times where I'd rather have polys up the ying yang. For example, if you model chains on a dino, then have them be real, moving chains. In certain games, if handled in a traditional manner, then I want tons of polys. For instance, an FPS like Halo, games like this have a dozen or two characters on screen at once tops. I want polys up the ying yang, after all, normal mapping is trying to fake the look of more polys...just give me more polys please. Now, if they were to take the Halo engine and make some massive RTS with hundreds of units, naturally a normal mapped low poly model is the way to go and I won't notice, because the individual units probably won't be focused on in the game.
It's all a trade off, mostly a memory tradeoff on Nextgen consoles. But a dev can still choose to use a lot of polys and limited normal maps. Or less variety and more detail in each model.
I personally don't think that Normal mapped mid poly count models are the way to go, but the UE3 screen shots seem to have a lot of people sold, so I'm probably in the minority.
I don't think that all games will go the UE3 low res model route, but with all the web sites screaming about how good it looks, it's hard to make a decision to focus on polygonal detail instead.
I'm sure the look will evolve and become more varied over the lifetime of the consoles.
I personally don't think that Normal mapped mid poly count models are the way to go, but the UE3 screen shots seem to have a lot of people sold, so I'm probably in the minority.
I don't think that all games will go the UE3 low res model route, but with all the web sites screaming about how good it looks, it's hard to make a decision to focus on polygonal detail instead.
I'm sure the look will evolve and become more varied over the lifetime of the consoles.
The way I understand it, if anyone disagrees with you on the issue it would/will be NVidia, not Sony.Mfa said:
Though I do agree, eDram is a nice asset to have for many things. It's just a question of how much of your featureset are you willing to give up for it (because we ARE giving up something for it, even in Xenon).
It makes sense if it's done properly. On PS3, stuffing this to GPU looks pointless, since you have CPU that can do it better. Especially if GPU will have VertexShaders, giving CPU even more leeway.ShootMyMonkey said:
But assuming it's flexible enough, this option would be usefull for nicer things then just plain Nurbs.
ShootMyMonkey
Veteran
eDRAM is possible on the PC... it's just that even with it, the bus between CPU and GPU is still going to be your bottleneck. But you're right in that it makes more sense to go that route with PC chips where a number of standards are set in place which you can't entirely have a whole lot of say on. Fact of the matter is that an ATI or nVidia wouldn't make as much money off of X2 or PS3 as they would off of PC chips, so it makes sense that most if not all of their research prior to making a console deal heads in that general direction. I mean, they're both selling these GPUs on raw power bases, so shader power is about the only thing you can give to really set it apart since that's really what utilizes that raw power.
It's easy for a programmer or someone outside the industry to say that NURBS will solve everything. People tend to forget that from an artist's standpoint, animating and especially part-welding with NURBS is a major pain. It's well-suited to rigid things like vehicles, but for something organic... eh... Subdivision surfaces are a better choice in that respect.
The thing is that the playing field is still not quite equal there. I mean PSP has a little bitty 480x272 screen, where detailed shader effects and normal mapping is probably not going to be all that noticeable. So raising the polygon count (which again, because of the small screen doesn't need to be raised that much), is perfectly feasible. On a big HDTV screen (which is what the 3 hope you'll ultimately play on), little details and shader effects will be more noticeable.
I agree that having SS support would be terrific because it could save a boatload of bandwidth. But it's definitely not a universally applicable solution. Sure it could make it easier to see "Ruby"-like characters in game (although if you ask me, Ruby with Monty Burns' face was a lot more amusing
)... but it still greatly increases the number of vertices going through the vertex shaders -- which if you apply complex shaders, is a problem. SS/NURBS won't replace things like normal mapping because it doesn't create new detail.Yes it is the point .
First off there has been no screens of a next gen fighting game released yet . Its alot easier to optimize and push your hardware to the limit when you know there will only ever be 2 characters on screen at once .
The game shots released last night are from a mmorpg . Compare it to another mmorpg from this current gen .. something like final fantsy and you will see the huge diffrence in quality. Expect that jump from a fighting game over a ps2/xbox fighting game
Iron Tiger
Regular
Did Huxley stir this rant up? If so, I'd ease off it because it's an MMOFPS, and there's likely to be a lot more characters on screen than we'd see in a deathmatch or campaign-type UE3 based game.GwymWeepa said:
I can't get down on normal mapping, because it really expands the potential for creativity. Combined with instancing, you could have one mesh in memory for several models, and they all look radically different because of varied textures and normals.
We also have to remember that normal mapping works best in motion, not in screenshots. If you've got a guy running down a corridor with recessed lighting fixtures overhead, you're gonna see the way his model interacts with those lights much differently than if you just grabbed a still frame of it. Even Doom 3's blockheaded, stubby fingered imps and zombies don't look all that bad in motion.
