nVidia Q&A posted

[NFS4]

http://www.hwsw.hu/oldal.php3?cikkid=361

 

[DaveB3D]

Sorry.. I have to rip into them on a few things:

Unlike some say, developers do not new software features on hardware that does not exist. NVIDIA introduced the first T&L GPU a year ago and since then game developers have been working on games with the new feature. They are about 40 games available using T&L technology. The significant performance increase comes from using the new feature, however, the GPU have higher clock rate and increase the performance of existing games as well.

40 games my ass. That is the biggest piece of bull. There are no 40 games.


Implementing NSR feature on a game is easy and does not need reworking the core of the engine, it's approximately a week's work to implement it on any game. Please find attached a list of games with T&L, this list is growing on daily basis, the direction is obvious.

Again, total bull. Does he even know what "NSR" is? He seems to think it is T&L.

Then look at that list: Diablo II? Dark Reign? Total bull. So are a bunch of other games on that list. A bunch of them don't support T&L. A load of marketing crap and actually blatent lies.

 

[Mr.IncrediblyBored]

Dave, I highly recommend you staying out of this.

I mean no disrespect, but leave the PR to the PR department.

 

[DaveB3D]

Sorry.. that stuff just pisses me off. I hope you can understand that (and I'm in one of those moods today.. very confrontational.. dunno why ).

 

[KarsinTheHutt]

<<<Dave, I highly recommend you staying out of this>>>

Why should he stay out of this? I think he does have some valid points. T&amp;L is essentially hypeware right now. nVidia cards are the best because of raw processing power, not because of T&amp;L.

 

[NFS4]

<< nVidia cards are the best because of raw processing power, not because of T&amp;L. >>


*NFS4 leaves the room as he hears Dave's footsteps approaching *

 

[Mr.IncrediblyBored]

I'm sure he has very valid points. I'm not disputing that. Its just that he works for 3dfx now and should be very careful how he speaks.

I read through the interview and not once did Alain even mention 3dfx.

 

[DaveB3D]

I disagree. If somebody benchmarks the V5 *correctly* and with performance optimizations enabled, you'll find high-resolutions side-by-side with the V5. I mean, for example, this holds perfectly true for the V5 and GTS.

 

[NFS4]

Yes I have noticed that with nVidia interviews, they never mention 3dfx.

 

[DaveB3D]

Knight,


I don't know why you and others seem to think I do, but I don't speak on behalf of 3dfx. I speak what I *personally* think. It should be taken as such.

 

[Mr.IncrediblyBored]

Whether or not you speak on the behalf of 3dfx is irrevelant.

The fact is that you still represent 3dfx and that whole FSAA debacle a week ago didn't make you or 3dfx look all that good.

 

[DaveB3D]

I don't represent 3dfx either.. I have a *job* there.. not a *life* there. The only thing I represent is myself.

 

[Mr.IncrediblyBored]

Let's face it. We can talk 'til we're both blue in the face and we won't get anywhere.

I'm going to go play some CS or TFC. Toodles.

 

[jpprod]

Some random thoughts on the subject...

One thing i find disturbing in this article, and it should be clarified. There is no NSR engine in GeForce2 GTS, nVidia just started promoting the Dot3 per-pixel lighting feature of (original) GeForce as a &quot;new shader processor&quot; with the GeForce2 GTS. Any programmer will confirm that D3D capabilities reported by the driver are *identical* on GeForce256 and GeForce2 GTS.

But Dot3 per-pixel lighting is certainly a feature which should be embraced by the devlopers a lot more, it's importance must not be downplayed no matter how nVidia tries to market it. As most new mainstream accelerators support it (cards based on new nVidia chips + all revisions of ATi Radeon cards), I see this as an every way superior alternative to currently most widely used game scenery lighting method, light mapping. With per-pixel lighting you also get free bump mapping, all in a single pass. Plus there's none of that &quot;three meter sphere of light&quot; feel you get on all current games (Quake3, Lithtech and UT engines) using light mapping.

Q&amp;A was correct about Dot3 lighting, implementing it is very simple and straightforward. And with DX8 having an abstraction layer for per-pixel lighting operations things will just keep getting simpler.

Hardware T&amp;L on the other is not a simple thing to implement properly. In order to optimize a game fully for hardware T&amp;L, developer must obey certain limitations in terms of geometry manipulation. But as I've said before, doing T&amp;L on hardware certainly limits you more than doing it on software, but so does doing polygon filling and texturing on hardware. Do we have any software polyfilling/texturing games out there any more? In a few years the situation will be same with AAA+ titles doing exclusively software T&amp;L.

 

[Dulanic]

<< I disagree. If somebody benchmarks the V5 *correctly* and with performance optimizations enabled, you'll find high-resolutions side-by-side with the V5. I mean, for example, this holds perfectly true for the V5 and GTS. >>



What do you mean by correctly? Do you mean by running a 3D benchmark program that takes advantage of V5 features? No offense but I dont play benchmarks do you? I think benchmarking a video card in a game is benchmarking correctly, because last time I played a program like 3DMark was.... well never. I could care less about benchmark programs, I want to see real world performance.

 

[DaveB3D]

I mean going into the advanced tab.. turning on refresh optimization, turning on Z optimization, turning depth precision to fastest, etc. Those types of things that will significally increase performance, but won't have any negative impact 95% of the time on the game itself.

 

[Dulanic]

I can also speed up my GeForce by playing with options too, we can say that for every card. Any card can be tweaked for your own personal preference of image quality/speed.

 

[DaveB3D]

There aren't options like these in GTS drivers. The options in the GTS drivers will sacrifice quality for performance. With the options in V5 drivers, they don't sacrifice anything. There might be a few games that it causes an issue with, so you just turn it down. However, I've yet to play a game that has an issue because of this.

 

[Dulanic]

So then why not set them right default in the drivers in the first place? Most reviewers dont have days upon days to tweak every card. Im grateful when they do, but I cant expect every reviewer to tweak your drivers. Nor do I expect them to sit there and tweak everyones.

 

[DaveB3D]

Because there are probably a couple of games it might cause an issue with and compatibility is most important to us. However, I've yet to see it be an issue myself. It doesn't take days either.. it takes 30 seconds, max. I knew about it when I was a reviewer and before I worked for 3dfx. It isn't tweaking really either, it is just setting the options. Not overly difficult. It doesn't take a brain surgeon after all. If you don't know what something does, you can right click it too and select &quot;what is this?&quot; and it will tell you.

 

[jpprod]

With the options in V5 drivers, they don't sacrifice anything.

Dave, using a 16bit Z-buffer (assuming &quot;Z optimization&quot; or &quot;depth precision&quot; setting does this) DOES sacrifice image quality over a real 32bit Z-buffer. This is not obvious to untrained eye in a vast majority of today's games, but it's very obvious in OpenGL 3D-modelling and some large outdoor areas in certain games, Deus Ex for example.

Bottom line is that all of these driver speed-up options are compromises between quality and speed (that's why they're options, not hard-coded defaults), thing that matters is what's setting's impact on image quality.

 

[DaveB3D]

jpprod,

I'm not sure what you are saying. If you are in 16-bit color, you use a 16-bit Z-buffer. If you are in 32-bit color you use a 24-bit Z and have 8-bit stencil. Turning on depth precision does not resort back to a 16-bit Z or something...

 

[jpprod]

VSA-100 doesn't support different frame/Z-buffer depths? What does the z-precision do then? Spill the beans, please

 

[NFS4]

Interesting jpprod

 

[DaveB3D]

Sorry I can't.. just leave it that it can offer a pretty substantical performance gain and typically not cause any problems (I know Q3 engine works great, MDK2 is fine, etc). If there is a problem, you can reduce it. I've yet to need to do that myself though.