Will next-gen consoles hit the 10G pixel barrier?
- Thread starter Megadrive1988
- Start date
notAFanB said:
Don't think that is going to happen
I would figure 4x fsaa on consoles and 6-8xfsaa on the pc . Mabye only 4xfsaa on the pc .
Look at the 9700pro. When it came out I could play almost all my old games with 6xfsaa . Now with farcry and other games i'm hoping to play with 2xfsaa .
Besides if consoles want to make a leap to hdtv (which i doubt) they wont have as much fillrate to burn up .
Guden Oden:
> Also, like I said, PC chips generally render at higher resolutions than
> consoles will, so more pipes is the way to go there (polys get larger at
> higher res too, so not as much edge waste).
That's true. Next gen systems will support HD resolutions however. Not many PCs run at resolutions higher than 1920*1080.
notAFanB:
> Jvd, surely you mean 16X AA by 2006/7?
Not unless the hardware designers have found a way to significantly reduce memory requirements. At 16 samples RAM is more of an issue than fillrate is. I'm hoping for 4x AA and 1080p support... that would be nice.
> Also, like I said, PC chips generally render at higher resolutions than
> consoles will, so more pipes is the way to go there (polys get larger at
> higher res too, so not as much edge waste).
That's true. Next gen systems will support HD resolutions however. Not many PCs run at resolutions higher than 1920*1080.
notAFanB:
> Jvd, surely you mean 16X AA by 2006/7?
Not unless the hardware designers have found a way to significantly reduce memory requirements. At 16 samples RAM is more of an issue than fillrate is. I'm hoping for 4x AA and 1080p support... that would be nice.
Panajev2001a
Veteran
Guden Oden said:
Not unless I process 16 pixels from 16 polygons in parallel
.Are you saying that the NV40 will dedicate 16 pipelines to only 2 triangles at a time ?
With the geometry count the PlayStation 2 can push that is ok ( textured it renders ), but with what NV40 will be able to do it is a bit of an over-kill.
I agree with you in principle though and that is why I do not expect PlayStation 3's Visualizer ( following the patents we have seen ) to have more than 4 pixel pipelines: ideally one pipeline per Visualizer-PE ( the PE with the Pixel Engine, Image Cache, CRTC and 4 APUs would be the Visualizer-PE
).and what about 16X from Matrox's parphelias series? look the point I was trying to highlight is that come 2005/6 I'm expecting some bright use adn advance on filtering at least, instead of just rehashing what technology we got at the moment.
I'd highlight items such heirachtical-Z, early z-rejection etc... as interesting modifications to (for example so don't jump me here) old skool z-buffer on the PS2.
basically 4XMSAA+16Xaniso in 2005/6?-->'meh'
notAFanB:
> and what about 16X from Matrox's parphelias series?
It doesn't anti alias the whole screen. You're better off with 4x MSAA.
> basically 4XMSAA+16Xaniso in 2005/6?-->'meh'
Well, 640*480, no AA and trilinear (bilinear even in some cases) in 2001 was hardly impressive either compared to what PCs had to offer.
> and what about 16X from Matrox's parphelias series?
It doesn't anti alias the whole screen. You're better off with 4x MSAA.
> basically 4XMSAA+16Xaniso in 2005/6?-->'meh'
Well, 640*480, no AA and trilinear (bilinear even in some cases) in 2001 was hardly impressive either compared to what PCs had to offer.
cybamerc said:
MSAA doesn't antialias the whole screen EITHER.Matrox' 16x FAA would likely beat 4x MSAA by a noticeable margin if it worked perfectly in every situation (which it does not, unfortunately).
Panajev2001a said:Not unless I process 16 pixels from 16 polygons in parallel
Are you saying that the NV40 will dedicate 16 pipelines to only 2 triangles at a time ?
With the geometry count the PlayStation 2 can push that is ok ( textured it renders ), but with what NV40 will be able to do it is a bit of an over-kill.
Graphics chips generally process pixels in "quads", groups of 2x2 pixels. This is done for mip map LOD computation and computing partial derivatives by differencing. That's why modern graphics chips generally have a number of pipelines that is a multiple of four. You're certainly not required to do it that way, but many software renderers do as well, so it's unlikely that you're rendering 16 pixels from 16 different triangles. It's also easier to have many "quad" pipelines active when they cover different porions of the screen since you can avoid concurrency issues.
On a different note, having many z-only (or rather z+stencil only) pipelines is very useful even in lower resolutions since you can use them for multi sample AA, stencil shadows and rendering shadow maps.
Oh... Forgot to reply to this...
MSAA doesn't eat fillrate, so there wouldn't really be anything to factor in.
jvd said:
MSAA doesn't eat fillrate, so there wouldn't really be anything to factor in.
Guden Oden:
> MSAA doesn't antialias the whole screen EITHER.
Yes it does. You reuse texture samples for the sub pixels but you still oversample each pixel.
> if it worked perfectly in every situation (which it does not,
> unfortunately)
Hence why 4x MSAA is better. Especially ATI's.
jvd:
> Who really wants msaa though. I want at least 4x fsaa comparable to
> atis
MSAA is one way of doing FSAA. ATI's Smoothvision uses an MSAA algorithm. I think their Mac drivers support SSAA additionally.
> MSAA doesn't antialias the whole screen EITHER.
Yes it does. You reuse texture samples for the sub pixels but you still oversample each pixel.
> if it worked perfectly in every situation (which it does not,
> unfortunately)
Hence why 4x MSAA is better. Especially ATI's.
jvd:
> Who really wants msaa though. I want at least 4x fsaa comparable to
> atis
MSAA is one way of doing FSAA. ATI's Smoothvision uses an MSAA algorithm. I think their Mac drivers support SSAA additionally.
cybamerc said:
You *process* the whole screen yes, but you only actually antialiases polygon EDGES. So the net effect is, no. It doesn't antialias the WHOLE screen. Only edges.
no ones arguing that point it an more effective solution overall for AA than adaptive 16X AA. my question is ar the benefits of alternative rewrites/implementations simply not worth the effort to R&D?
recall that once Supersampling was actually deployed in commercial products and we've all migrated over to MSAA+aniso as an alternative.
so what, are we stuck with an (scaling) solution such as the above or what?
notAFanB:
> my question is ar the benefits of alternative rewrites/implementations
> simply not worth the effort to R&D?
Oh I'm sure ATI is already doing research on different AA algorithms but that doesn't neccessarilly mean that it has come up with a better solution.
> recall that once Supersampling was actually deployed in commercial
> products and we've all migrated over to MSAA+aniso as an alternative.
Well, super sampling still offers the best quality. If the chip is fast enough there are definitely advantages to using it over MSAA.
> so what, are we stuck with an (scaling) solution such as the above or
> what?
Maybe, maybe not. I just wouldn't hold my breath for a major breakthrough in AA algorithms. And I certainly wouldn't hold my breath for console devs sacrificing performance in favor of IQ. If we can get just 4x at 1920x1080 that would be cool. It would be a major improvement over today's consoles.
> my question is ar the benefits of alternative rewrites/implementations
> simply not worth the effort to R&D?
Oh I'm sure ATI is already doing research on different AA algorithms but that doesn't neccessarilly mean that it has come up with a better solution.
> recall that once Supersampling was actually deployed in commercial
> products and we've all migrated over to MSAA+aniso as an alternative.
Well, super sampling still offers the best quality. If the chip is fast enough there are definitely advantages to using it over MSAA.
> so what, are we stuck with an (scaling) solution such as the above or
> what?
Maybe, maybe not. I just wouldn't hold my breath for a major breakthrough in AA algorithms. And I certainly wouldn't hold my breath for console devs sacrificing performance in favor of IQ. If we can get just 4x at 1920x1080 that would be cool. It would be a major improvement over today's consoles.
I want 640x480p with 6x AA to be standard, and 1280x760p with 2x AA to be standard on next gen consoles.
Or at least AA at the low resolutions, and the option for hi res without AA.
Maybe they could have 2 versions of the system, a standard version that does 640x480, possibly without AA, and possibly at 30 fps, sold for like $200, and then a more expensive version for like $500 that could do hi res, AA(not necessarily at the same time as hi res......ah they better for $500) and 60 fps.
Or at least AA at the low resolutions, and the option for hi res without AA.
Maybe they could have 2 versions of the system, a standard version that does 640x480, possibly without AA, and possibly at 30 fps, sold for like $200, and then a more expensive version for like $500 that could do hi res, AA(not necessarily at the same time as hi res......ah they better for $500) and 60 fps.
Similar threads
