New dynamic branching demo

Re: SM3

you sounds like politician parsing every sentence.

 

Re: It is, what it is.

The go and do it, and stop wasting people time! You say you want to write shader 3.0 code and you appear equally P.O.'ed at both companies so go and buy a 6800 and write shader 3.0 code, and stop wasting peoples time here. If you are a code developer then there shouldn't be anything that stops that code operating on other SM3.0 boards when they are available!

 

Re: SM3

For 3D RTR in general, it would seem the answer is yes.

Yes again.

Wait, are we talking about the present or the future? Because you just leaped from two general questions into a rather contextual one. At present, FC v1.2 SM3 benchmark numbers don't show SM3.0 as useful in the sense of offering more performance for a gamer's money.

Please don't mix current with future usefulness to prove a shaky point. Yes, judging by game dev interviews, SM3.0 is a very desirable feature. But the bottom line for a gamer is usually price/performance, and in that sense, SM3.0 doesn't seem a must-have feature for now. Extrapolating from the 9700P, I also don't think it'll be a major factor in the 6800U's and X800XTPE's reigns as the kings of the hill, either, but that's not a certainty.

 

Re: SM3

Is it useful *now*? Will it be useful for the life of NV40?


Why bother implementing SM2.0 or SM3.0 when everyone will be implementing SM4.0 in the future?

SM3.0 is useful, but like all new features in graphics cards, it will only be worth having when (a) developers start using it, and (b) when it's fast enough to play games with. Unless you are a developer (and thus in the small minorty of the market).

 

useful?

So it's useful in future but not useful in present? so for another
year until R500 comes out it will not useful?

FC 1.2 currenly implement only small set of SM3.0 features.
next patch 1.3 probably contains more substantial changes as
well as possible HDR and others. Also don't forget more
optimization NVDA is capable on SM3.0 compilers.

In fall, we should see NV48 which probably will turbo boost SM3.0.

So you are saying don't bother SM3.0? sitck with SM2.0! not until
ATI says so?

 

Re: SM3

why bother to implement SM4.0? why don't you wait for SM5.0?
wait why bother? why don't wait for SM99?

hmm. so SM3.0 is not useful because nobody is coding but wait
Farcry is doing that. no wait they are not real developer because
they have foresight to develop next generation. real coders
use SM2.0!

 

Re: It is, what it is.

That's not an object instancing demo. It's a collective AI demo. There will still be a performance hit from making multiple draw calls, but it may be reduced by using reasonably-complex vertex programs and relatively high-detail models.

 

Re: SM3

And Farcry is not showing any difference in speed or visuals using SM3.0 on NV40 compared to SM2.0 on R420. Where is the SM3.0 advantage for the customer here?

The question is, do you want to pay for a checkbox feature that doesn't help the game? Do you want to base your buying decisions on features that are unused or give no benefit?

 

Re: It is, what it is.

But it serves the same purpose. It allows one to have "an insane number of dudes" on the screen at once without things becoming unbearbly slow. It seems like it's possible to do that without SM 3.0.

Yes, it's not the same means, but it's the same end. That's all that really matters in PC Graphics.

 

Are those some of those zen questions I've heard about, ZT?

Look, we've gone over this already. Of course SM3.0 is useful, but no one's going to ditch SM2.0 and its much larger user base just because SM3.0 is more convenient/capable--at least, not yet. Devs aren't going to dump the NV3x and R3x0 generation in favor of NV4x just because it's easier to program for SM3.0 cards, just like even very vocally critical devs like Valve didn't dump the NV3x in favor of R3x0.

We can bicker all we want, but we'll only know for sure this fall whether SM2.0 has put X800 owners at a disadvantage to 6800s in current and upcoming games. And don't forget that the market may force the X800s to lower price points if the 6800s turn out to be faster or more desirable to the majority of buyers due to SM3.0. I'm skeptical, you're not. Only time will tell.

 

Re: SM3

To be fair, the 6800s do benefit noticably from SM3.0 in some maps--at least before AA+AF are applied.

 

Re: SM3

Are those the Nvidia supplied demos using the Nvidia supplied drivers by any chance?

Yet the competition's SM2.0 card manages to do the same with less transistors, less heat and less power. When you include other visual enhancements like AA/AF, the competition's SM2.0 provides better speed with the same visuals. Again, how does this show SM3.0 is an advantage to the customer choosing between slower SM3.0 hardware, or faster SM2.0 hardware?

 

Re: It is, what it is.

No, it doesn't. Object instancing serves to reduce the small batch overhead of Direct3D. ATI's demo does nothing to combat this problem.

 

Re: It is, what it is.

But is it really a problem? That needs a solution? It seems like it runs fast enough without object instancing.

 

"And Farcry is not showing any difference in speed or visuals using SM3.0 on NV40 compared to SM2.0 on R420. Where is the SM3.0 advantage for the customer here? "

Just out of curiosity as people have seem to become rather single-mindely focused on dx9 for farcry, but does this particular game even show a difference between shader2.0 and shader1.1 thats worth caring about?

 

Yes.

 

Re: It is, what it is.

Add more 3x more detail to those models and watch those framerates fall.

 

Re: It is, what it is.

Oh no, Dave's gone supersaiyan!

sorry, I just was catching up on this thread, and this caught my eye. it had to be commented on.

 

Re: SM3

I guess we have too much time on our hand. here goes..

SM3.0 usage in game is in infancy. let's not jump to conclusion yet.

here's another far cry 1.2 update review:

http://www.hothardware.com/viewarticle.cfm?articleid=550&cid=2

more interesting is this one:

http://www.nordichardware.se/Artiklar/?skrivelse=252&page=3

current games are not really optimized for SM3.0 once they do
we should see much more improvement and will be ready for NV48.

And when you add HDR and FP16 blending....

 

Re: It is, what it is.

You'd need a lot more tests to determine that. For example:

1. How much CPU load is there on all the draw calls? Could this be reduced with object instancing, therefore increasing overall performance in a real game that attempts to draw that many objects?

2. These are relatively complex objects. There are other, simpler objects for which object instancing is much better-suited, such as grass or trees.

3. We have no way of determining whether the performance in ATI's demo is good or not, given that we have nothing to test it against.