Predict: The Next Generation Console Tech

Status
Not open for further replies.
You missed the part about alot of high res textures.

This current gen has alot of repeat textures . I haven't played enough of mgs4(hated it) but from the few hours i spent with uncharted you can see this in the game.

I'd love to see a next gen oblivion that has 8 gigs of textures.... well i guess they'd use a 1 gig of memory for other stuff but still even say 6 gigs of textures would be amazing to me .

2 gigs is only a 4x increase in ram capactiy. The xbox to xbox 360 was over an 8 times jump . The ps2- ps3 was even more. I believe 16 times ( 32 vs 512mbs) I mean we will most likely have 50 gigs of storage space on the disc. Why not use it for more than video . Why only displace 2 gigs of it at once

Talking about texture variation seems to me to be a bigger problem for it's effect on increasing art production costs than it does an issue for the tech..

I wouldn't count on seeing massive increases in texture variation across the board because the cost constraints within the productin budget may restrict it, especially for smaller dev houses..

(Personally if you ask me, I really don't see texture variation as being one of the major distractions in current gen games bar a small few, Halo 3 spring to mind.. Most games have enough geometric detail variation that it hides alot of it from most players.. But I digress..)_
 
I think where texture repetition is an issue (racing car tracks and grass sidings!) procedural texturing can replace it. I really think procedural all-sorts can replace a lot of stuff, it's just not happening this gen. It's funny how many amazing technologies exist that just aren't being used!
 
The need for procedural textures is decreasing for each generation. Todays consoles have >128 times as much RAM as Playstation 1.

Cheers
 
It's not just about storage though. It's about less artist effort to create textures and an improved look through texture variation. For the console manufacturer POV, if procedural content was strong enough, they could reduce RAM and save money on each unit.
 
I was wondering though , back in the cd rom days there was a drive called true x


Yes, it was discussed awhile back in this thread. There were some issues about the particular implementation IIRC that prevented it from being feasible. You'll have to check back.
 
It's not just about storage though. It's about less artist effort to create textures and an improved look through texture variation. For the console manufacturer POV, if procedural content was strong enough, they could reduce RAM and save money on each unit.

Um, isn't procedural content mostly to save disc space and not ram?

It takes quite a bit of proccesing power to actually calculate and create textures and similar things on the fly. You still would have to fit these textures in your memory, so unless you want worse textures than you could have, these will take up just the same amount of memory.

So unless you dont have the storage space you need, i dont really see the point of procedural generated content on the fly, just takes away resources.

Procedural generation is great for creating variation, but that is usually done offline, not on the fly, while you play the game. Like the trees in oblivion, procedurally generated, but that was long before the game was pressed to a disc ;)
 
It takes quite a bit of proccesing power to actually calculate and create textures and similar things on the fly...
Which is what I'm talking about for next-gen. The RAM requirements can be reduced if assets are being created on the file with 'pixel shaders'. This gen it hasn't proven viable that way.
 
Which is what I'm talking about for next-gen. The RAM requirements can be reduced if assets are being created on the file with 'pixel shaders'. This gen it hasn't proven viable that way.

You are shifting resources there, i.e. from RAM burden to GPU burden. Hence you will need to increase your GPU silicon realestate to compensate for the additional workload. So which is cheaper? Right now it stands that it is unrealistic to do HQ procedural textures OTF. I think load time procedural textures makes a lot of sense though: dynamic, shifting some loadtime from slow optical media to compute, procedural textures can offer a bit of variety, etc. Procedurally generated content at "design time" is just an important factor: do you want an artist to touch every pixel on a massive cliff on the side of a road? Or do you want to run an application that generates it and an art spends time tweaking it? This is a great way to save time/money where it works visually. The stuff from Gamefest is pretty cool and I know MS has been looking in this area.
 
procedural content has long been used in games, from random and pseudo-random dungeons to thousands of solar systems fitting on a single low density floppy, fractal generated landscapes etc.

But, while it can be useful, it always end up looking the same! While it's OK for stuff that's meant that way (the solar systems, or placement of trees and grass), textures are the last place where I'd expect procedural content, except in tech demos, things like Unreal 1 engine's detail textures and maybe part of some terrain data.

procedural is one of the recurring buzzwords we get each gen, along with tesselation and raytracing. (though tesselation might end up be used a few years from now)
 
Which is what I'm talking about for next-gen. The RAM requirements can be reduced if assets are being created on the file with 'pixel shaders'. This gen it hasn't proven viable that way.

But this is not free!! It costs resources, with procedural generated content, if you want to do it on the fly, like you do, you use up resources. If assets are created with pixels shaders or whatever, you still need atleast the same amount of RAM to have this asset in game as you would need to have if you just had to fetch it from a disc.

If you do it on the GPU, that gpu has less resources for other things. Why are you keen on wasting up resources to generate the content on the fly, when you could have made it before hand, thus having more resources left for other things.

I dont see much point in procedural generated content on the fly, aside from simple easy stuff. For generating content? Far from the best solution.

I understand the benefits of procedural generated content created offline, as it eases the workload for artists, but generating content while playing costs way to much resources and doesn't give you any direct benefits.

Using procedurally generated assets on the fly, means you need to use more CPU\GPU (whatever you want to generate it with) and at best, the same amount of RAM as before. Its only a benefit for saving disc space. If disc space isn't relevant, then i see no benefit of say procedural textures and assets thats generated on the fly. Might as well just procedurally generate the textures\assets offline, just stream the data you need in, and have more cpu\gpu power for other things.

Procedurally generating things is useful, but i cannot see it being used in the degrees being suggested here.
 
Using procedurally generated assets on the fly, means you need to use more CPU\GPU (whatever you want to generate it with) and at best, the same amount of RAM as before.
You don't use up the same amount of RAM. Consider a rusty drain pipe with 5 textures applied for specular, rust, slime, etc. These same 5 textures are either applied to lots of pipes and you see them repeatedly, or you have a different set of textures for each pipe, maybe a mix of a few sets, with an even higher RAM footprint. If you did the thing entirely procedurally, the texture footprint would be zero. For each drainpipe pixel you'd calculate the colour via algorithm.

With geometry you also get huge savings. Rather than needing 30 different soldier models in RAM, you'd have one and adjust it per model, a la Tiger Woods character creation, to create unique models

Yes, there is a large processing demand increase, but it's just one use of the available resources. If you aren't using the advanced GPU grunt of next-next gen graphics technology for making textures more realistic and enemies more vaired, what else are you going to use it on?! A lot would come down to how much processing power is available, what optimizations are possible (you'd ideally only process each pixel drawn on screen), what other factors need to be balanced...I expect the best attainable would be a compromise of precalculated texturing, perhaps as source data for algorithms, and purely procedural stuff.

I do have a major complaint with repetitious content in games. It stands out in trees, little branches that are identical from tree to tree, hoards of cloned bad-guys, Identical walls and sand-bags, strips of grass, and all sorts. You'd need a considerable amount of RAM to have precreated assets so numerous and varied that the repeating components aren't as obvious as they are now. I don't know if that'd be economical in the future versus processing performance.
 
Didn't Barry Minor write a white paper based on Cell creating a fully procedural FPS like dungeon just done on a Cell (the 8 spe version) with "reasonble graphics?

If this is the case without any GPU load then surely this backs up what Shifty is saying regarding less resources. Especally if you can shift some of the rendering to the GPU to both use the GPU resources and free up resources on the Cell for other gaming tasks (i.e. physics, ai etc.).

Even if this not the case then surely the next generation Cell (say its the 4 ppe x 32 SPE one) could easily create a passable or good graphical game that can procedurally generate a game without having to have a super duper expensive GPU to!
 
If you aren't using the advanced GPU grunt of next-next gen graphics technology for making textures more realistic and enemies more vaired, what else are you going to use it on?!

Ha! You pulled a funny!

What, you aren't joking? Man, age is really getting to you Geezer ;) Anyhow, I started a thread a long time ago about "next gen" gripes. We can start with shadowing that have smooth edges, occlude/skew/scatter realistically, minimize artifacting, are cast by all dynamic objects and received by all dynamic objects, better AO hacks, etc. Higher quality dynamic lighting (too many games are still using pre-rendered light maps), far better control over HDR techniques, more advanced radiosity hacks, etc. We need better anti-aliasing, a LOT better texture filtering, more variety/resolution on textures, more particles that are also more interactive (water and smoke still have a FAR way to go), higher quality reflections... freaking 3D GRASS that actually behaves like, well, grass. If we can get all of that the GPU could surely strut some muscles doing some physic maybe. But looking at games like Forza and GT, just look at the general lack of most of these things. It is a bloody shame!

Spending GPU cycles, 30-60x a second, to create the same textures 30-60x a second, would surely cut into the above. I wouldn't say no textures should be done at runtime, but at game loadtime or install time makes a lot of sense. If you could generate all your textures at runtime using, say, 30% of your GPU cycles then I could see shifting more resources to the GPU and cutting some RAM. But I don't think we are near there for the texture complexity in a modern game.
 
Doesn't ProFX generate texturemaps for conventional texturing, using up RAM, rather than deal with on-the-fly shading? Regardless, those shader thumbnails show exactly what I'm talking about - rich, vaired detail, unique to every surface...if you've got the processing power to invest in it!
 
Anyhow, I started a thread a long time ago about "next gen" gripes. ...(long list of complaints. Boy, what a whiner! :p)...
I agree, all issues to be tackled, though some of which are being dealt with this gen and will show marked improvements on current, outdated technology.

Spending GPU cycles, 30-60x a second, to create the same textures 30-60x a second, would surely cut into the above. I wouldn't say no textures should be done at runtime, but at game loadtime or install time makes a lot of sense. If you could generate all your textures at runtime...
I don't see why it has to all be generated at runtime though! We only need halve the amount of textures, combining traditional texturing techniques with procedural creation, to save on some GBs of RAM. For a console designer facing the choice of, say 4 GBs versus 8 GBs, they ought to be considering the impact of procedural content, rather than saying 'we need 16x what we've got now 'coz everything's gonna be done the same old way it is now'. And if the procedural content in harmony with conventional techniques works out, then 2GBs versus 4GB may also be a valid consideration. If you can start with the same amount of assets as now and procedural mutate them into alternative assets on the fly, you won't need any more than you have now. Thus RAM size would be more a matter for improving on the current base level, 4x the RAM giving 4x the fidelity, not 4x the assets which is instead handled procedurally with processing cycles.

Edit : I'll also add that this thread is full of hypothesizing that next-gen, GPU cycles are going to be spent on calculating AI, physics, and totally non-graphics stuff. Before you take arms against my idea of using the GPU to work on texture graphics as it's taking away from the important graphic tasks it needs to do, you may want to stockpile defenses against all those who would steal your GPU cycles and waste them on CPU jobs instead ;)
 
I think where texture repetition is an issue (racing car tracks and grass sidings!) procedural texturing can replace it. I really think procedural all-sorts can replace a lot of stuff, it's just not happening this gen. It's funny how many amazing technologies exist that just aren't being used!

Yeah. And the funny thing is that procedural texturing is something Phill Harisson was talking about before the PS2 was even released as one of the technology feats that would be in the PS2 (or was it his vision?).

Yet no implementation has been seen
 
ProFX is nice and all, but we are talking about multiple orders of magnitude too slow for doing it online during rendering!
 
I don't understand why htey wouldn't put more alot of ram in . How much will 8 gigs of ram cost in 2012 and how much will it cost in 2018 when they are most likely done with the generation.

Its the same thing this gen. gdr 700 was how much in 2005 , how much is it now.
 
Not only is the amount of RAM important but also, of course, the bandwidth and also latency of that RAM.

Some consoles in the last generation (PS2, GCN) got away with less RAM but higher bandwidth, especially thanks to embedded RAM. While higher bandwidth does not make up for less RAM, it does help in other areas.

While the Xbox was usually concidered more powerful than PS2, and it was in many ways, the PS2 could do things the Xbox could never do, because of high bandwidth eDRAM.

The GCN was also able to pretty much rival Xbox performance in many (not all) ways thanks to a combination of high bandwdith embedded memory and low latency of both the embedded and external memory.

Unless extremely high bandwidth external main system memory and bus interconnect architectures are developed for next-gen consoles, I think embedded memory will play an important role, once again, in 1 or 2 of those consoles. It is odd that Sony & Microsoft reversed themselves this gen. Xbox 360 has embedded memory, PS3 does not. Last-gen, it was PS2 that had the embedded memory (as did GCN) but original Xbox didn't.

Assuming next-gen console's external main system memory bandwidth only increases by 4-8 times from this gen, providing a few hundred GB/sec of bandwidth, I hope Microsoft continues down the path of eDRAM, going from 256 GB/sec to several TB/sec, and that Sony returns to eDRAM with PS4.
 
Status
Not open for further replies.
Back
Top