zidane1strife
Banned
ShootMyMonkey said:
Who knows... maybe nvidia delivered the fillrate sony thought was necessary...
wtf is up with poly counts
- Thread starter GwymWeepa
- Start date
Who knows... maybe nvidia delivered the fillrate sony thought was necessary...
Okay, let's see how a fully modeled 3D chain would work on our dino monster.
First you have to model a single chain link, which is basically a torus primitive. Make the cross section a 6 sided circle, and the chain itself a 12 sided circle or elliptical whatever. That's 6 times 12 quad polygons, which is 144 triangles for a single chain link. Build the chains from about 5 dozen links; that makes 8640 polygons for the chain alone. So just by turning the chains from a normal map detail into geometry, we've doubled the polygon count of the creature.
Okay, so you want it to move with the character as it's limbs move around? You have 3 basic choices to implement this.
First is per keyframe vertex animation - for every movement of the character, store at least 15 positions per second for each of the ~8000 vertices. Oops, that's too much data; 8000 vertices * 15 keyframes * 3 32-bit floats for XYZ coordinates takes about 1.3 megabytes per seconds of animation.
Okay, then let's have our animator do the job using bones. You probably have to ad at least one bone per 2 chain links, for the ~60 links that makes an additional 30 bones for the character. An average skeleton requires 50-60 bones (with less than 5 fingers per limb), so adding the chains takes about a 50% performance hit. The bigger problem is that the animator now has to do a lot of extra work for every second of animation he creates, to make those chains rock and roll as the character turns, jumps and runs around.
So, can we do some rigid body physics instead? I'm not sure about the calculation times for Havok, but I'd expect it to take about as much processing time as an entire scene in Half-Life 2 (remember, we have 60 individual chain links colliding with at least 2 neighbors and the character's body). All for a single character walking around without any interaction with the enviroment.
Well, on second thought, let's just bake those chains into the normal map, and hope that we'll have displacement mapping soon...
First you have to model a single chain link, which is basically a torus primitive. Make the cross section a 6 sided circle, and the chain itself a 12 sided circle or elliptical whatever. That's 6 times 12 quad polygons, which is 144 triangles for a single chain link. Build the chains from about 5 dozen links; that makes 8640 polygons for the chain alone. So just by turning the chains from a normal map detail into geometry, we've doubled the polygon count of the creature.
Okay, so you want it to move with the character as it's limbs move around? You have 3 basic choices to implement this.
First is per keyframe vertex animation - for every movement of the character, store at least 15 positions per second for each of the ~8000 vertices. Oops, that's too much data; 8000 vertices * 15 keyframes * 3 32-bit floats for XYZ coordinates takes about 1.3 megabytes per seconds of animation.
Okay, then let's have our animator do the job using bones. You probably have to ad at least one bone per 2 chain links, for the ~60 links that makes an additional 30 bones for the character. An average skeleton requires 50-60 bones (with less than 5 fingers per limb), so adding the chains takes about a 50% performance hit. The bigger problem is that the animator now has to do a lot of extra work for every second of animation he creates, to make those chains rock and roll as the character turns, jumps and runs around.
So, can we do some rigid body physics instead? I'm not sure about the calculation times for Havok, but I'd expect it to take about as much processing time as an entire scene in Half-Life 2 (remember, we have 60 individual chain links colliding with at least 2 neighbors and the character's body). All for a single character walking around without any interaction with the enviroment.
Well, on second thought, let's just bake those chains into the normal map, and hope that we'll have displacement mapping soon...
Look when you compare like games to other like games the diffrence is there
You can't compare a fighting game where its 2 characters (mabye 4) and a back round at all times is in the game to a mmorpg where at anytime you can go from 1-500 characters .
So lets compare like things
This is the new mmorpg for the xenon shown yesterday . This is most likely on dev kits and is also being designed with the pc in mind .
This looks nice. The characters are pretty good polygon wise , most likely around halo 2 lvl polygons , the textures are much higher . the filtering on the textures seem much better as does the fsaa . The items on the charactes like back backs and what not look detailed and look to have depth so they not just a texture on teh body . The shadows look somewhat complex , they aren't just blobs under the character and the draw distances is pretty far with some fairly complex buildings
This is final fantasy 11 (the online one )
Looking at this things that jump out at me are the very low polygon characters , bad filtering as everythign looks grainy , no fsaa (though it could be that its not doctored like the one above ) the textures are very low polygon the shadows are just blobs , the complexity of buildings are also very low with very low quality textures and going back behind the characters things get fugly unlike the next gen mmorpg where things stay crips .
I think there are glaring diffrences between the two systems . Now i know your going to say what about this game or what about this game and I will say compare like games . Compare a fighting game to a fighting game , compare a mmorpg to a mmorpg , compare a racing game to a racing game .
I've allways said that next gen games would only look a little bit better than the last big block busters of the previous generation untill developers got deep into the systme later in life . Its how it has allways bene and will continue to be
You can't compare a fighting game where its 2 characters (mabye 4) and a back round at all times is in the game to a mmorpg where at anytime you can go from 1-500 characters .
So lets compare like things
This is the new mmorpg for the xenon shown yesterday . This is most likely on dev kits and is also being designed with the pc in mind .
This looks nice. The characters are pretty good polygon wise , most likely around halo 2 lvl polygons , the textures are much higher . the filtering on the textures seem much better as does the fsaa . The items on the charactes like back backs and what not look detailed and look to have depth so they not just a texture on teh body . The shadows look somewhat complex , they aren't just blobs under the character and the draw distances is pretty far with some fairly complex buildings
This is final fantasy 11 (the online one )
Looking at this things that jump out at me are the very low polygon characters , bad filtering as everythign looks grainy , no fsaa (though it could be that its not doctored like the one above ) the textures are very low polygon the shadows are just blobs , the complexity of buildings are also very low with very low quality textures and going back behind the characters things get fugly unlike the next gen mmorpg where things stay crips .
I think there are glaring diffrences between the two systems . Now i know your going to say what about this game or what about this game and I will say compare like games . Compare a fighting game to a fighting game , compare a mmorpg to a mmorpg , compare a racing game to a racing game .
I've allways said that next gen games would only look a little bit better than the last big block busters of the previous generation untill developers got deep into the systme later in life . Its how it has allways bene and will continue to be
BigGamer X
Newcomer
Good post. If next gen. games are still looking as blocky as today's, this will be the poorest leap in console technology ever. Hardly gets you exited about getting one. (blocky shoulders and 256MB of ram, 12x DVD aren't what I expected for next gen, it's like half way there!)london-boy said:
Geez, where were all you guys when Doom 3 came out? I thought I was alone in hating how poorly tesselated everything was.
But it was supposed to be a step forward yet everything was extremely blocky!jvd said:
WAHHH!!!
I don't think it was , add the fact that on a geforce 3 which you can get for 15$ it played extremely well at 640x480 and sitll looked better than console games shows that it looked nice.
I think half life 2 looked better , but i don't see anything wrong with doom3 . It looks leaps and bounds better than halo 2 or any fps on the ps2
I think half life 2 looked better , but i don't see anything wrong with doom3 . It looks leaps and bounds better than halo 2 or any fps on the ps2
Well displacement maps still wouldn't help your animation problems though. As far as physically animating the chain - I am pretty sure there are better and faster ways then the crazy bruteforce approach you choseLaa-Yosh said:
Well what the heck did you expect - techwise, Square games are amongst the lowest end games on PS2jvd said:
Although as far as character polygon detail goes, - most of FF11 chars are around 1500-3000polys, which is pretty standard for games this generation.Fafalada said:
But it'd at least look more 3D
As far as physically animating the chain - I am pretty sure there are better and faster ways then the crazy bruteforce approach you chose
I dunno, our other character TD guy here decided to use dynamics for a character with some similar equipment, instead of rigging it with bones. It's a relatively simple solution, once he has the proper values set, he only has to run the simulation and write our the results in a cache (as per-vertex deformation data). With a chain link, there really aren't many alternatives, as any distorsion of the rigid bodies is very easy to catch...
But I really have no other ideas about easier or faster solutions for a realtime engine
Well given that you're probably not looking for a full chain dynamics (it's just a part of the clothing) I think a set of constraints could be defined that would look pretty convincing, and definately a heck of a lot cheaper then doing full on polygonal collision of each chain link.Laa-Yosh said:
If we can't avoid individual links as dynamic objects, at least using some kind of toroid primitive instead of collision mesh should help somewhat (using primitives, I think we should be capable of several thousands moving dynamic objects on nexgen machines).
DudeMiester
Regular
I really really really want more polygons in games today. I find nothing more of an turnoff, then seeing the actual geometry behind the scene. It's as nerve grating as 640x480 with no AA as far as I'm concerned. Game and hardware develepers have truely failed when it comes to polygon counts imho. Hopefully once hardware raytracing takes over, which I'm sure it will eventually, I'll be satisfied. However, it may be a long long wait until that happens. I hate Nvidia and Ati.
I hate Nvidia and Ati.Hey that's the first time we hear about this. I don't think it ever occured to many of us that the Xenon GPU is forgoing something.Fafalada said:
EDIT
And since we're on EDRAM, any ideas about how Sony may emulate the 48Gb/s GS for backward compatibility? Ideas include
#1 The NV GPU emulates it, but not sure about any problems due to bandwidth differences.
#2 The GS is actually included in the system itself? I'm sure many devs will think of creative ways to exploit it, though it seems a bit out of place in the whole application flow.
#3 NV GPU emulates it, and it actually has a 51Gb/s local memory...
Well do you have any other console mmorpgs that we can look at ? Mmorpgs in general are allways behind the curve for the simple reason that at any point you can go from nothing happening to 500 people on screen all having a fight to the end .
Even an online fps has its limit , on 1 map in most fps you can have 32-64 people at once and you can optimize for worse case .
ShootMyMonkey
Veteran
It's nice that you talk about it as if it's just a flight of whimsy that keeps polycounts low. Yes, it takes no effort whatsoever to push 160,000 polys per frame down the PS2 pipe. It's a friggin' cakewalk. It took no effort on our part to quadruple the speed of our line-of-sight tests. I was obviously being quite lazy for a few weeks there optimizing code and ripping apart assembler code 5 times over.
I think we should all really get to work one of these days... Maybe then polycounts wouldn't suck so bad. I mean, look at these damn lazy hardware designers... oh, but the quantum mechanics is so haaaarrrd. I don't wanna put in pulldown transistors! It makes it all confusing and stuff... But it costs moonnneeeyy!! Boy and those game programmers... talk about morons! I mean, how hard could it possibly be to reduce sparse matrices to equilibrium solutions in under 4 ms? I mean... friggin hair!! Everybody knows how hair works! We see it everyday! But, noooo, they have to reinvent the wheel on every game to get some extra polys down the pipe. Cry me a river, right?
I needed that kick in the pants. Maybe come Monday, I'll actually go to work and really get serious.
I don't think anybody's bitching about their MMORPG not having 40K poly meshes just yet, but seriously. If the draw distance is low enough and the characters in view are limited to no more than maybe eight, then what the hell stops you from implementing character models between 10,000 and 40,000? No, that's not with normals maps, or PCRT, or parallax bump mapping or some exotic "virtual geometry" hack.
It's more that people are wanting to know why they don't have and possibly won't have those quality of characters in something closer to the scale of a Tomb Riader game. Geometry in RAM could be an issue, computing the shadows is an issue, simulation of dynamic objects is an issue (hair, cloth, chains, gear-work, etc.), lighting I'd assume would be an issue, collision could be an issue, etc: is there anything really stopping that scenario?
Yet another angle, I know that those dynamically generated items would be impossible on PCs at this time, but what about these upcoming consoles? Is there enough brute force (and that is what it comes down to) in any of them in anybody's professional opinion to make that possible?
We're not and nobody is asking for Nalu or even Ruby here, just for curves on something.
Later
It's more that people are wanting to know why they don't have and possibly won't have those quality of characters in something closer to the scale of a Tomb Riader game. Geometry in RAM could be an issue, computing the shadows is an issue, simulation of dynamic objects is an issue (hair, cloth, chains, gear-work, etc.), lighting I'd assume would be an issue, collision could be an issue, etc: is there anything really stopping that scenario?
Yet another angle, I know that those dynamically generated items would be impossible on PCs at this time, but what about these upcoming consoles? Is there enough brute force (and that is what it comes down to) in any of them in anybody's professional opinion to make that possible?
We're not and nobody is asking for Nalu or even Ruby here, just for curves on something.
Later
As i've said we have barely seen any games , we have no completed games and the games we are seeing are in higher res , more fsaa , higher forms of filtering , much higher res textures and many other effects going on .
IN half life 2 and doom3 seeing the game in motion and seeing a screen shot are two diffrent things . Its the way it is .
Will there be games that make polygon counts in todays games look like nothing ? Of course there will be . Just look at first gen dc and ps2 games compared to the newest xbox and ps2 games. There is no comparison in terms of effects , textures and polygon counts .
Its all in the fact that people don't realise that the hardware for next gen isn't even finished being made yet , the developers are just barely scratching the surface of this non existant hardware and the old systems have been put through the wringer to display the graphics you are currently seeing .
That's entirely understood by me. I was just clarifying the "high-poly vs virtual geometry" sentiment to "Monkey", who I posted right after. I know regardless of if or not we've seen final hardware or software that 20k, 25k, 30k, 35k, even 40kpoly meshes are possible in some games (not just fighters) a year or two out, it's a given, but I wanted to catch a group opinion on if or not some of the devs thought so as well.
It'd be interresting to see where some of them stand on the issue.
Later
It'd be interresting to see where some of them stand on the issue.
Later
