Final Doom3 benches at HardOCP

The reason is, rendering shadow maps at R32F with pixel shaders, and doing filtering in the pixel shader would triply kill performance. First, floating point textures would be a hit. Next, just running the shader to calculate Z and write it would be another hit. And finally, doing filtering in the shader would be a further hit. Then there's the memory bandwidth hit too. There's no way this will run as fast as the depth buffer format.

Fourth, treating a transparent texture as a special case isn't that big of a deal. Sacrificing performance for such a small gain in situations where you can use it just isn't worth the trade off at this junction.

Perhaps what we really need is floating-point Z-buffers.

 

Well, I get the same performance in my shadow mapping test with both R32F and A8R8G8B8. So, what you say doesn't hold good. Plus, floating point surfaces don't need more bandwidth, it all depends upon the bit-depth.

That kinda contradicts what you said earlier

 

Tell me how you'd do transparent shadow mapping with D24S8. You'll need another texture to hold the transparent regions, and that means rendering to another texture. If the two aren't of the same bit-depth, you cannot use MRTs, so you end up rendering the scene twice.

Almost all the games use alpha blending and transparency. The only way to do this is to use floating point textures, nVIDIA's shadow mapping simply doesn't help.

 

First, you're not doing filtering for R32F. Compare a 4-sample PCF vs a D24S8. Secondly, with either R32F or A8R8G8B8, you're writing the same amount of data, so don't expect a bandwidth difference. The difference is between R32F + depth buffer vs D24S8 depth buffer alone.

The whole point of floating point depth buffer is solve a whole host of problems caused by lack of range in the z-buffer, plus the fact that you don't need to run a fragment shader just to write Z, you have fixed function HW doing it for you, which is presumably optimized for fastest possible z-writing/reading/and rejection.


As for the transparency problem, first, I don't see alot of transparent objects in the average scene except for water, glass, caustics, etc. A first approximation is to simply use alpha test cutoffs. But a better solution is to let artists treat large transparent surfaces as special cases, this is the approach in UE3 it seems,with glass windows, and water, etc being treated via projective lighting. Like anything else, there are tradeoffs. I don't see this problem being solved perfectly and effectively performance wise without a TBDR.

 

DC – if you are going to go down this tired road again, get your facts straight. I’ve not done any down talking of the quantity of poly’s in D3 – others have in this thread, but not me – so, don’t put words in my mouth. As for the number of players this is not speculation – multiplayer in Doom3 is limited to 4 characters.

Its generally accepted that the type of shadowing method used for D3 is not the best for outdoor scenes, however that doesn’t stop licensees altering the number of lights and plenty of other parameters, but that will change the characteristics of the engine/performance, and as has been pointed out engine performance rarely full expands to title performance dependant on what the other developer have used. I’ve not made any comments about whether or not other licensee’s will have more or less characters on screen since I don’t know – I’ve only talked about D3’s multiplayer.

What negative comments are these? You seem to be a little touchy again – It appears to happen every so often with you.

As for GT2/3 its long since been known that these would favor NVIDIA’s implementation because of their technology usage, and I’ve been saying that for a long time. I’ve also switched over to using them in the latest reviews to give some relative performance data on how on applying AA/AF in these types of usage scenarios. And, as a fact (and, as I again pointed out earlier) when I sent a bunch of questions to them some weeks back I did ask this very question as to whether they would use it ongoing – why did I ask? Perhaps because of the types of comments I’d been hearing recently from other parties? You’ll see the answers soon.

Now, the only thing I have commented on - that appears to have flicked your "upset" switch is the usage of stencilling in the future. Given what NVIDIA are evangelising to Developers, their own expectations and what they have already stated they are doing in the future, do you think significant quantities of transistors will be expended on stencil performance beyond the features we already have?

 

Don't worry DC, you're not alone in recognizing Mr. Baumann’s selective criticism, especially Nvidia related.

He may deny it, but the pattern is easily seen with but a quick glance at his posting history

Nvidia = Criticize / Downplay
ATi = Defend / Praise

He's just not called on it more due to the fact that most posters here are fanatically pro ATi.

 

You just reminded me:

Wonder which unbiased B3D member said that?

 

So, after all of that, it comes down to you not taking a joke?

You really are being touchy again DC. Now, how about apologising for putting those words in my mouth again?

 

First of all, there's no need for personal attacks like this. Second, you are at least as in love with nVidia as you say Walt is with ATI - a perfect example of the pot calling the kettle black......

 

I've noticed the pattern as well. Ahh well, still better dealings than at Rage3D for sure though... lmao.

 

Malfunction agreeing with CitizenC and DC... that's a real suprise. Fact of matter is, you guys cry wolf whenever anyone disagrees with your beloved nVidia. Even those that try to be objective. You are so biased it's impossible for you guys to even think "outside the box" of nVidia. The pattern is, if a site tries to be impartial, it's a ATI biased site. Hell, you guys even defended the FX series of cards...... and a company that lied and cheated to the point that these's no way to even begin to compare competing products anymore.

 

Nope, I give ATi the same chance as I gave nVidia. Sorry to say, nVidia has impressed me more. Many of this issues have been solved with the NV3x series, ATi has merely adopted the approach and thus formed their own.

It is pretty safe to say, with Futermark's removal of Catalyst 4.7 drivers from the approved list, ATi is in servere damage control operation.

If you learned nothing as to how ATi and nVidia have handled business last year, then it is easy to see why you cannot see what is going on now. All it means is lower prices in the end, score one nVidia, try again ATi. Maybe ATi shouldn't have sat on old tech for so long, I mean, if the consumer was the concern... it would look differently at the moment. 8)

 

Exactly how are those two things connected...?

 

Ehm, what you are talking about? http://www.futuremark.com/community/drivers/?approved

 

Ahem... http://www.driverheaven.net/showthread.php?threadid=51405&s

 

So, Ati is asking Futuremark to remove the drivers from the list. That's quite different than "with Futermark's removal of Catalyst 4.7 drivers from the approved list".

 

That's all I am gonna say anymore about this as it is going off topic. The cookie is crumbling from all sides at the moment. I have no doubt it can be fixed, though the time frame that all of these problems being fixed doesn't bode well with me.

 

The 3dmark03 problem which only effects a few card types anyway (or maybe just 9800XT not sure) is fixed in the 4.8s which are due in the next week or two so not long to wait

 

I think that my post history stands up far better than some of the others complaining. However, given the evident preferences they have its fairly easy to see how they can see dropping hints as “pumpingâ€.

I’ve never denied that I have better knowledge about various inside ATI elements than I do about NVIDA. In general however there is often more information to find out about ATI because NVIDIA’s tech is much more effectively marketed and sometimes even ATI don’t know what they are capable of. When I know about these things I’ll drop hints because some people are interested – but also because devs and others read these forums and they can be useful to them and ultimately to end users.

Take, for instance, the recent Vertex Instancing issue on ATI’s hardware – how many people knew about that until I started dropping hints? No-one, not even many of the dev’s I’d talked to. I’d known about the possibility of ATI supporting for many months, but ATI being ATI I figured they wouldn’t do anything about it – however, it wasn’t until a couple of months back that someone else started talking about MS giving instancing its own caps bit in these forums – and ATI contacting me, requesting that I should ban them as “No one from ATI shouldn’t be talking about MS dealings†(I could find no link to this poster actually being ATI though), did I get the tip off that, indeed, they were looking to support instancing. So, I then go off and do some further digging a few weeks back that I find that, yes, they do now have Beta drivers with instancing support and they are now talking to devs about it. On contacting their PR about it when I found out – they [PR] had no clue that they could do it until I brought it to their attention.

In this instance I was dropping hints about it because I knew it was a possibility some time ago, and knew it was happening a little while before anyone else because I did some investigative journalism, however I won’t always openly talk about it until they will talk about it because they are told in confidence. However, ultimately this is something that can benefit the developer and benefit the end user and I have no issues with bringing it to peoples attention for these reasons, and because it generates us hits.

This is just one such issue that I have found out about there have been numerous others.

 

SO Dave----

Why exactly do you think ATI hides these features?
Is their marketing department so bad that this is a major issue for the company-----
Nearly every other company in the world would be trumpeting better features on their products - what exactly is ati's problem with doing this?