Is the Xenos a shader monster yes or no?

I'll just pop in a few notes here.

You almost never get near full utilization in both the vertex shader and pixel shader simultaneously. When you do, it's fleeting and not sustained. It doesn't really matter how you tailor the geometry in a game, either. Therefore adding both vertex and pixel shading power together is a useless metric to compare against Xenos.

Note that if achieving the peak MADD rate, RSX and G70/G71 is incapable of any texturing. If you had 3 MADDs for every texturing op, then they will perform 36 MADDS and 12 texture lookups per clock.

I don't think I've ever said the mini-ALU is "worth 30%" of a full one. I do recall saying a GF7 pipe is about 30% faster than a R300 pipe. Not sure how much info you can get from that.

To the original poster, yes, Xenos is a shading monster. Just wait a year or so for the results. Even on the PC, the graphics are nowhere near as good as they should be given the power of the top end cards.

 

The reason the brickwork and tiles were so impressive in Toy Shop was that this abberation didn't occur. That's the whole point of POM. The good news is that Xenos should do POM every bit as well as the X1K series does.

 

heres a couple shots of my first experiments with parrallax a coouple of years ago (hence the crap shadowing)
http://s24.photobucket.com/albums/c26/zzeek/?action=view&current=parrallax1.jpg
http://s24.photobucket.com/albums/c26/zzeek/?action=view&current=parrallax2.jpg
(note this is programmeur art)

i cant see the attraction of the parallax in kameo et al, its an extremely simple technique which quickly becomes tiresome. in short parallax sucks and is little better than plain bumpmapping. there are far better alternatives out there now (though there are more computionally expensive)

 

Rare just really overused the technique.
Parallax mapping should only be used with small bumps, where there is little possibility of seeing the silhouette of the object.
Never as a substitute for displacement mapping.

There are however, as mentioned, new relief mapping techniques out there that improve on the problems with parallax mapping.
Actually I'd be interested to know if anyone thinks they could be a substitute for higher polygon models.
Ie. a whole new rendering paradigm.
Maybe it would be more efficient to have really simple geometry consisting of quads, boxes and spheres with one of the various kinds of relief mapping applied?

 

So if a machine can't do displacement mapping effectively within the game, they shouldn't have any large bumpy surfaces in the game?

Personally I'm happy to accept a mid-way solution rather than no solution at all. Bumpy surfaces without bumpy edges is better than only flat surfaces. If bumps are to be kept small to minimize the straight-edge fault, you're limiting art direction considerably. Castles will be made of pebbles instead of rocks!

We're not in a position to have perfect graphics yet. There will be compromises.

 

Actually on a large flat castle wall I think it should look acceptable. It’s with small objects where you can see the edges all the time, that I think it breaks the illusion completely.
In that case the artist will have to make do with polygons as best he/she can.
After all, low frequency displacement mapping shouldn't be too difficult with todays flexible vertexshaders and the SPEs on Cell.

 

There are parts in Oblivion which look just as effective. And he's right, the effect isn't as convincing in game, for a start thats a digital photo of the game running on a TV screen and games always look more realistic under those circumstances.

EDIT: in one level of UT2004, they actually achieve a very similar effect on walls and floors using geometry.

 

You seemed to like it quite a bit back in December:

http://forum.pcvsconsole.com/viewthread.php?tid=17749&page=2

 

Yes, and im still extremely impressed with them now. But Oblvion wasn't out in December which was my comparison point for this level of parralax mapping on the PC.

Im not saying it isn't great, im just saying its not unprecidented on the PC.

 

Well I've clocked over 80 hours on oblivion, and you have to cherry pick big-time to get parralax mapping on par with what Kameo has throughout the entire game. The only nice example of this I can think of in oblivion is the walls of caves, and even those are a step below Kameo imo.

 

Admittedly, Obvlivion is a much bigger game than Kameo, and also it was never designed for one single console, but for PC, and we know how things work in that field...

 

I totally agree, it's not a fair comparison...yet it was made!

 

For ALU FLOP's your totals are way off (hint: at the same clocks X1900 has 50% more than 7900!).

 

'Efficiency', or utilisation, will be a function of developer effort. Certain architectures might make it easier for a dev to utilise a chip well, of course, but there's no one number that could sum up the potential utilisation of any GPU, in different pairs of hands.

 

On top of that, who's word do we have that Xenos is '95% efficient'? ATi and Microsoft's? Excuse me if I take that with a grain of salt.

 

I see your using the free FP normalise in the G7x calculations. I decided not to use it as its not really comparable to the rest of the ops being more specialised. In the same way I havn't included the free branching operations for the ATI hardware.

We found out a while ago that its actually 216 for Xenos since the scalar opp is only one FP operation as opposed 2 as was initially thought. Thats makes the total flops per ALU, per cycle at 9 rather than 10.

Thats just marketing speak. ATI also mentioned something about its pixel shaders in R5xx being 100% efficient. Its likely vertex shaders are often underused but I would hope that they are overkill enough in a modern GPU to allow the pixel shaders to remain pretty close to full utilisation most of the time.

 

well i dont know a shit about GPUs, but reading a few posts from different B3D usres, i got impression that USA allows you to reach this 95% efficency, and it's not some "full of hope" promise nor marketing BS

 

Presumably though, GPU efficiency is greater than simply the efficiency of its pipelines.

 

Please provide evidence for that. I take all these efficiency claims with a grain of salt.