Next Gen bottlenecks (Volition interview)

I just pointed out if these devs have data throughput problems then they're doing something wrong. I'll point anyone curious as to my reasoning in the direction of warhawk, a game where the devs claims a seamless, 'levelless' world where you can walk, drive, fly around freely. So obviously it IS doable.
Of course it's doable. By sacrificing something else. I can give you an example of how we got away with streaming of really enormous outdoor levels on various games on the PS2 at my last job -- by having zero variance in textures (other than lightmaps, anyway). Every stream block had the same textures, so all we really had to load was one new lightmap and geometry (which itself was made up of instances of geometry blocks used elsewhere).

You want it? Fine -- just be prepared for monotony or Lego-land or low visibility or some combination of them all. Warhawk is no exception to this rule.

The problem is that won't please many people. Everybody seems to think "oh, technology has advanced, we can get games that try to be everything for everyone", and I guarantee you that's a recipe for mediocrity if you bother to shoot for that. Not just in games, but in all products and services that can ever exist.

Consoles have always been bitches to program, I don't know of a single piece of hardware where devs claimed it didn't have at least some sort of PITA quirk... It's always something!
What makes you think it's necessarily solely about programming? What makes you think it's necessarily a problem of internal dynamics of the platform or the studio? What makes you think that there are no repercussions associated with breaking from what's natively supported? What makes you think that cheating here and there or trying something unique has no costs that can be deemed unacceptable? What makes you think that it all really comes down to one problem and/or one solution?

I mean, it's relatively easy to use your own software codec for textures, but that means either uncompressed textures in memory (which means less content can be actively in use at a time) or recompressing into some DXT format on load, which induces extraneous losses in quality because you've gone through a doubly-lossy process. And in practice, that may not be acceptable in the least.

You seem to be belaboring under the infinitely misguided notion that there's an obvious solution, or even a solution at all, for anything. What's worse is that you're just one of a sea of people who seem to believe that everything simply just works because someone else said it works for them.
 
the future is no more slow optical discs, including the future HVD format(s).


the future is also doing away with texture-mapping altogether.
 
the future is no more slow optical discs, including the future HVD format(s).


the future is also doing away with texture-mapping altogether.


A replacing it with what? cartridges? very dense very fast HDDs? Holographic memory?

What happens in place of texture maps?
 
solid-state media. like what we see with the Nintendo DS, only faster and higher capacity, even if they don't have as much capacity as Blu-ray. we don't need more capacity IMO, we need vastly higher transfer speeds and higher seek times. something that optical media probably cannot deliver. even if you goto 8x Blu-ray or whatever.

as for texture-mapping, it's a "cheat" to provide the illusion of more detail. we need more polygons instead. textures will be around for awhile, (the Xbox720, PS4 gen will still have textures) but they will ultimately go away, maybe in 15 years, I dunno.

something like REYES, that Panajev was excited about, would do away with textures, or at least vastly reduce the reliance on textures.


texture-mapping is technology from the 1980s from what I understand. it moved into consumer applications in the 1990s. but it's an outdated way of providing more detail.

even though textures won't completely go away for a long time, they'll have to change if graphics are to improve. I don't have the answers. maybe procedural textures are the mid-term solution.
 
Last edited by a moderator:
Procedural engines seem to be the wave of the future.

http://en.wikipedia.org/wiki/.kkrieger

96kb Procedurally Generated FPS demo.

Looks like there's bumpmapping, dynamic lighting when you fire, and a bit of a bloom effects on a resolution of 1024x768 all running at 60-80fps on just the CPU.

There's also sound in this 96kb demo.

Here's some screens:

http://img161.imageshack.us/img161/9050/kk1qx5.jpg
http://img461.imageshack.us/img461/7572/kk2vq7.jpg
http://img134.imageshack.us/img134/3584/kk3ov9.jpg
http://img70.imageshack.us/img70/2865/kk4xy3.jpg
http://img207.imageshack.us/img207/5753/kk5ln9.jpg
http://img488.imageshack.us/img488/1193/kk6nw7.jpg

Add'd system requirements :

- A 1.5GHz Pentium3/Athlon or faster.
- 512MB of RAM (or more)
- A Geforce4Ti (or higher) or ATI Radeon8500 (or higher) graphics card
supporting pixel shaders 1.3, preferably with 128MB or more of VRAM.
- Some kind of sound hardware
- DirectX 9.0b

Thats a pretty decent system to run those basic graphics.

Also the game is slow, boring and perhaps has the most repetative level and monster design in a game.

And what you meen on the CPU? is that a typo? Because the game needs a GPU.
 
Last edited by a moderator:
Megadrive, Pixar PRman (the Renderman as far as I am concerned) spends on average, in their movies, a LOT of time on textures' I/O :). They still use textures AFAIK, but ask Laa-Yosh for more details... he is the expert :).
 
something like REYES, that Panajev was excited about, would do away with textures, or at least vastly reduce the reliance on textures.
I don't believe REYES renderers rely on textures less than other methods. They even use textures for things games currently don't like displacement maps. If we truly store all detail in polygons then we definitely won't be using solid state media anytime soon as the memory requirements will be outrageous.
 
Well, there's the room to breathe and the further programmability insofar as supporting procedural texturing, as well as the fact that when sampling micropolygons, you only really need constant color over the face.

Put those two together, and it *sounds* like REYES renderers are less reliant on texture images. How it sounds and how things really are is rather different, no?

In practice, I don't think you can ever really get rid of textures simply because the idea of being able to "paint" color information onto a surface is a very straightforward thing for artists, and even when you get into geometric detail where that's no longer necessary, it's easier for them to think in terms of a 2d image rather than painting in fine detail on a 3d surface.
 
btw, that tech demo is pretty old....wave of the future? I don't think so... It's certainly not worth the loading trade-off! But hey... I guess we can all go back to floppy disks (5.25" style of course ;) )

ch'ure crazy mang. not worth the loading trade off? how about coupled with an HDD? how about no download times? no installation times? how about no need for physical media? this is of course only demo but if you really think procedural is a dead-end, talk to... every demo programmer who has gone on to become a rather well-paid graphics guru? i can't wait till spore gets here and shuts ppl up about this subject! XD

in all seriousness, for things involving human models, especially, procedural seems a long way off. plus there's lots of natural phenomenon that might be better done with photo-realism for awhile. but... it would be SO WORTH it if these bloated teams could fire half their "artists" (Blizzard excepted) and maybe hire more creative people to develop some solid gameplay and think about how to adapt the settings of their games to these more abstract technologies. stop wasting my fucking time with your *%@^%!@!ing shooters on rails and photorealism!!!

Edit: okok, so, i just replayed the produkkt for the first time in a couple years (woo a couple of years of hardware development has yielded some bugses!), and it's not replacing hand-drawn art anytime soon. ;D but i foresee a long period of co-dependence, a period that has already begun. for the record, on my overclocked a64 3200 and unlocked x1800gto, it takes FAR FAR longer (i'm thinking triple) the time to load a level of FEAR than to generate the produkkt. FEAR of course looks nicer, but it wasn't actually any more fun. Monolith, if you're reading this, bring back the fun times!!!!
 
Last edited by a moderator:
ShootMyMonkey said:
because the idea of being able to "paint" color information onto a surface is a very straightforward thing for artists
And I don't see how that precludes you from using storage types that aren't 2d arrays - it's just a tool's user interface, the data could be stored inside purple muumuu's for all the artists care.
 
ch'ure crazy mang. not worth the loading trade off? how about coupled with an HDD? how about no download times? no installation times? how about no need for physical media?

Edit: okok, so, i just replayed the produkkt for the first time in a couple years (woo a couple of years of hardware development has yielded some bugses!), and it's not replacing hand-drawn art anytime soon. ;D but i foresee a long period of co-dependence, a period that has already begun. for the record, on my overclocked a64 3200 and unlocked x1800gto, it takes FAR FAR longer (i'm thinking triple) the time to load a level of FEAR than to generate the produkkt. FEAR of course looks nicer, but it wasn't actually any more fun. Monolith, if you're reading this, bring back the fun times!!!!

whoa whoa whoa.. I was referring to complete dependence on procedural media. I was only commenting about this demo, and it's not worth the load time to me, not for what I saw.

There are things that procedural generation is handy for like tiled textures or lots of repetition. But an entire game? Cut-scenes, voice acting, multiplayer...On the scale of a first party game...not for a long time IMHO.
 
Last edited by a moderator:
as for texture-mapping, it's a "cheat" to provide the illusion of more detail. we need more polygons instead. textures will be around for awhile, (the Xbox720, PS4 gen will still have textures) but they will ultimately go away, maybe in 15 years, I dunno.

something like REYES, that Panajev was excited about, would do away with textures, or at least vastly reduce the reliance on textures.

texture-mapping is technology from the 1980s from what I understand. it moved into consumer applications in the 1990s. but it's an outdated way of providing more detail.

even though textures won't completely go away for a long time, they'll have to change if graphics are to improve. I don't have the answers. maybe procedural textures are the mid-term solution.

The entire movie VFX industry is still relying on textures and there's no sign of droppping the technology. One of Pixar Renderman's greatest strengths is its texture caching.

Replacing textures with more polygons is not practical either. You'd store color and other information per vertex, right? So why would that be better than using a low-res mesh, subdivide it at rendertime and displace it with a texture, then use more textures for other attributes as well? This approach has a lot of advantages, you know, like easy LOD from texture MIP-mapping and easier memory management... and it's more artist friendly, and you can easily use photos and reuse textures and so on...

I mean, what's the problem with using textures in general? They work pretty well.
 
In practice, I don't think you can ever really get rid of textures simply because the idea of being able to "paint" color information onto a surface is a very straightforward thing for artists, and even when you get into geometric detail where that's no longer necessary, it's easier for them to think in terms of a 2d image rather than painting in fine detail on a 3d surface.

Actually, no - with programs like Zbrush and Mudbox, it's pretty easy to use a brush-based interface to create detailed geometry. But then this geometry is converted into a displacement map because it's easier to manage the data that way.

http://www.mudbox3d.com/gallery/index.html

Nevertheless, there actually is a desire to be able to somehow keep that detail accessible, so that one could animate every little wrinkle and fold that'd otherwise be static in a displacement map. We may see some developments in this area, but it will take many more years and a lot more computing power and memory...
 
And I don't see how that precludes you from using storage types that aren't 2d arrays - it's just a tool's user interface, the data could be stored inside purple muumuu's for all the artists care.
There's no reason to do otherwise. What motivation is there to reinvent that wheel?

Actually, no - with programs like Zbrush and Mudbox, it's pretty easy to use a brush-based interface to create detailed geometry. But then this geometry is converted into a displacement map because it's easier to manage the data that way.
That, and so it can be useful for other tool pipelines such as games.

Sure, I've seen ZBrush a million times, and the results are great and all, but I don't see it replacing everything. It makes perfect sense for modelling, but I don't see the same principle working for textures or even images which have some alternate meaning beyond just color.
 
ShootMyMonkey said:
There's no reason to do otherwise. What motivation is there to reinvent that wheel?
None if you stick with existing rendering pipeline and hardware of course. But this was in reference to people talking about Reyeses and Raytracers and what not - where reinventing the (realtime)wheel is the crux of the whole debate.
 
You want it? Fine -- just be prepared for monotony or Lego-land or low visibility or some combination of them all. Warhawk is no exception to this rule.
Nobody said there wouldn't be limits. Game creation always involve compromises of some sort... Anyway, for someone who have seen some warhawk screenies, I'd say personally that things do look pretty complex and varied. True, there's not a tremendous amount of detail once you LAND and get up-close to buildings, but who can say if it's streaming or just memory space that is the limit. After all, you can se A LOT of the level when you're out there flying.

Perhaps they cheat by interspacing the complicated stuff with zones of simple stuff like forests or plains, so that they have time to load up the next city or mountain canyon or whatever landmark before the player reaches it. Then again, elegant cheating has always been a hallmark of good game design. In half-life 2, if you're some distance from a window, it's merely an opaque surface. You have to get pretty close before it turns transparent. FarCry uses cardboard cutouts for any tree or foliage that's more than 100-ish meters away, WoW won't draw any fences or signposts etc that's some distance off and so on... Everybody cheats.

What makes you think it's necessarily solely about programming? What makes you think it's necessarily a problem of internal dynamics of the platform or the studio? What makes you think that there are no repercussions associated with breaking from what's natively supported? What makes you think that cheating here and there or trying something unique has no costs that can be deemed unacceptable? What makes you think that it all really comes down to one problem and/or one solution?
I think you assume too much. I'd suggest you stop extrapolating wildly from my replies, you did the same in your previous post. Maybe it works in coding, but it doesn't work with my posts. :D Stick to what I actually write, k?

You seem to be belaboring under the infinitely misguided notion that there's an obvious solution, or even a solution at all, for anything.
I get this thrown at me because I said Warhawk's a launch game that does seamless world streaming on PS3? :oops: I thought stating the obvious would be met with slightly less hostility. :-?

What's worse is that you're just one of a sea of people who seem to believe that everything simply just works because someone else said it works for them.
Uh, well, thanks for having another cliché thrown at me, but seriously... Chill.

Then again, if we have one dev team that does streaming in a launch game, and another that says they need 4 gigs of RAM when the machine's got 256MB x2, well, which of the two appears to have the more efficient solution?

And don't go off on a ranting tangent and assume all sorts of things about me again, thanks. I grow tired of stuff like that pretty quick.
 
True, there's not a tremendous amount of detail once you LAND and get up-close to buildings, but who can say if it's streaming or just memory space that is the limit. After all, you can se A LOT of the level when you're out there flying.
I'd wager it's a whole lot of both. Lower detail also frees up the option for more varied stuff, which in turn affects how much you can stream. And it's quite possible to get detail without relying on high res (and instead relying on high *effective* resolution). And seeing a lot of the level simply implies that they aren't very render-limited. From what I've seen, there seems to be a lot of open space populated by bots more so than "world stuff."

Then again, elegant cheating has always been a hallmark of good game design.
Hm. "Elegant cheating"... never quite heard it put that way.

I think you assume too much. I'd suggest you stop extrapolating wildly from my replies, you did the same in your previous post.
Actually, in case it wasn't clear, I was also pre-empting further replies that just about anybody would have otherwise said in response (e.g. the compression part).

I get this thrown at me because I said Warhawk's a launch game that does seamless world streaming on PS3? :oops: I thought stating the obvious would be met with slightly less hostility. :-?
Obvious what? That Warhawk streams their world? Sunshine, lollipops, and rainbows... That proves diddly-squat, contrary to what you might believe. Warhawk's solution isn't proof that streaming invariably works for everybody.

And don't give me any garbage about you not having made any blanket statements... It's not as if it's totally obvious from phraseology like "Warhawk... 'nuff said" that you didn't mean to imply that Warhawk's streaming is some sort of catch-all proof that streaming from disc/disk is perfectly feasible.

Then again, if we have one dev team that does streaming in a launch game, and another that says they need 4 gigs of RAM when the machine's got 256MB x2, well, which of the two appears to have the more efficient solution?
Don't really know. What does one game demand of its content and design goals as opposed to the other? What does the public at large demand of the genres of these two titles? What are the key marketable features of these titles, and what does that imply for content? What's safe to strip away from one title as opposed to the other?

As far as I'm concerned, efficiency is something that is measured within constraints... not something that occurs because of the lack thereof.

And don't go off on a ranting tangent and assume all sorts of things about me again, thanks. I grow tired of stuff like that pretty quick.
I advise you to watch how you word things, then.
 
Of course it's doable. By sacrificing something else. I can give you an example of how we got away with streaming of really enormous outdoor levels on various games on the PS2 at my last job -- by having zero variance in textures (other than lightmaps, anyway). Every stream block had the same textures, so all we really had to load was one new lightmap and geometry (which itself was made up of instances of geometry blocks used elsewhere).
You were just held up by all the futuristic Nazis you had to generate... :p ;)
 
You were just held up by all the futuristic Nazis you had to generate... :p ;)
Different job. My last one involved the owner whose entire character was imbued with Anti-Microsoft dogma and the publisher who likes to play games of brinksmanship at E3 and also not tell you what platforms your game is supposed to be on.

Jabjabs said:
It's official Shootmymonkey will forever be associated with Nazis and chrome guns, not the worst reputation you could get though. :LOL:
You're forgetting, the guns all have to fire from the same point! And the glowing functional spies, don't forget how important it is that spies glow!
 
Back
Top