On NV30 class chips, 5200-5600 and the rare 5800 which were flooded into the market at very low costs benefit lots using integer precision/FP16. Only need to look back 9 months ago to see what integer precision looks like, and the perfomance hit associated with these chips running in full precision...since the 5200/5600 make up a majority of Nvidias DX9 line-up and not 5900's or 6800's....or don't these cards count. http://www.hardware.fr/articles/468/page4.html http://discuss.microsoft.com/SCRIPT...L=directxdev&D=1&F=&S=&P=7149 If we are talking IQ overall, how about developers that are partnering up release patches that intentionally cause poor visuals, lower IQ without my consent, the person that paid for the $400 dollar graphic card and $50 dollar game...for graph wars. http://www.nvnews.net/vbulletin/showthread.php?t=26855&page=1&pp=15 After all the company that made texture stage, brilinear and shader replacement a house hold name needs no defending.
Sigh if I must. http://www.nvnews.net/vbulletin/showthread.php?t=30854&highlight=NV3x (I apologise for linking nvnews content here beyond3d. I would normally not do such a thing) I'm not claiming that it does or it doesnt. I am saying that its clearly not a valid comparison to turn one off and leave another on, Texture stage opts affect performance for both ATI and Nvidia. These are not relevent to brilinear or trilinear. Nor does turning one off and leaving another on do anything to help the disparity between FP16 and FP24. Obviously. But only if you have ATIS AF set to Aplication Preference. if its set to control panel. Texture stage optimisations occur no matter what on stages 1-7. Regardless of what the app is asking for if you set the AF to 16 and force it through the control panel. Depends the engine. Since we are discussing Control Panel AF, Requested AF from the developer becomes a moot issue. Since Control Panel AF over rides any requests developers make for AF. Read above please. I dont see why your making a deal out of the Doom 3 benches. Obviously if you dont think 10 FPS is a big deal, Then why would 4 FPS be a big deal?
Chris talking to you is obviously pointless. You feel that optimizations that nvidia do and don't disable are fine. But things ati does that it doesn't disable aren't fine. Its okay. I've read enough of your posts to see this trend. Simple fact is if its bad for one company to reduce options and do less work and not allow us to change that . THen its bad for the other one to do the same thing. If ati's texture optimizations are bad then so are nvidia's interger and fp16 optimizations . If you care to admit it or not.
Yeah. Lotsa hot air. :? I really, really wish this round brought more to the table from a technological standpoint, you can have only so much intelligent, technical discussion about SM3.0 without sinking to the level of "it's great! - no, it's useless!" type of debates. Optimizations and game devrel policy got kinda boring long ago. Right now, IMO the most interesting question has to be what the mainstream / budget cards will be like (but that's kinda OT, now ain't it).
JVD, you are of course entitled to your opinion. But you have evaded every issue I have brought up and put words into my mouth. As I have been saying since this discussion was brought up. Disabling one optimisations in some vain attempt at countering another is that isnt even relevent (Since when did texture filtering level the playing field for Shader precision) Disabling texture stage optimisations. Which are irrelevent to issue of FP16 and FP24, in some attempt to level a playing field and make benches more "accurate" is completely folly. This has been my argument since the beginning. If you want to try and level the playing field. There are other methods to use, Take those shaders (As tom from tomtti systems has done) and make them in a more comparable mode. if you feel thats fair) But it leads us back to the same point I made from the beginning. It's folly to disable a filtering optimisations as a justification to level the playing field when comparing FP16 to FP24 performance variables. Which brings us back to what you originally said.
Guys, please try and respect each other....... Also, no real numbers - percentages, yes - were given. So whatever optimzations were set where ever really doesn't matter. You guys are argueing over semantics..... Chris & JVD, both of you need to remember that there is absolutely no way to do an "apples to apples" comparison of these 2 architectures. When the new Cat's are availible and when the 1.2 patch is re-issued, I'm sure there will be benchmarks all over the place,. Then you guys can argue these points with some knowledge..... Right now, it's kind of like pissing into a strong wind..... I like pie too.......
I don't think you could ever rule out PP. :? (not refering to PP hints) What's there to stop the hardware from forcing less precision at a low level?
Even though as a developer I don't like the extra work of 16/32 versus single precision, the effects of fp16 are easily exaggerated as has been seen in this and other threads. As long as the inputs (vertex and texture colors) and outputs are 8-bit fixed point, as is very much the case still today, you are hard pressed to have complex enough shader (without trying obviously) that doesn't do: a) dependent texture lookups b) calculate using world scale coordinates c) log/exp-type operations to have any difference between fp16 and 32. Those paths that might have problems can be quite easily identified and made use fp32 if necessary. Normal dot3 bumpmapping with tangent space vectors is perfectly fine with fp16 for example. If you are adding 1+1, you don't need to do that in doubles or calculate 10 digits of PI if you wan't to know if it still is 3.14 and something.... Still, obviously I'd like to be able to use fp128 all the time without thinking, but that time isn't quite here yet.
chris u say its not fair to turn off nvidias texture stage and leave atis on. is it fair to turn on nvidias filterng ops when its noticeably worse than atis? is it fair that ati with its ops on looks as good as nvidia with its ops off and yet nvidia is still being compared with its opts on?
MMmmm Warm Router/Switch for kitty to cuddle up to. Must be all the Pr0n passing through. BTW, what about the NEW Cats, err.. Catalysts? Anyone got the 4.8s yet (anyone who can share them that is)? Interesting to see what the benifits are (lotsa rumours out there, especially regarding OGL-D3).
I haven't caught any of them yet. I've been feeding about a 6 kitten family that lives under my front porch and cars for a couple of weeks now, but they're still too skittish to catch. I'm thinking of just leaving the garage door open about a half foot and hoping they'll sneak in...then nasty slam the door and MAKE them love me! (Oh, the kitty on the router is Chance. His twin sister Buttercup is lying on my mousepad right now annoying the bejesus out of me shedding all over! :? ) As for the new Catalysts? I honestly don't think even ATi knows right at this point what is up, I think that the FarCry patch pull threw them for a loop too and I think they all just said "the hell with it until Monday!" and went home for the weekend. I don't expect any news on anything until Monday, and I'm pretty damned sure I'm right. (And I'm only telling you that since you were nice about me kitty. )
THat is exactly what I am saying. But he is saying these numbers don't count since ati has optimizations on. What he doesn't want to admit is that nvidia is forcing lower percisions (even less than fp 16) which is cheating as pp does not mean interger. He brushes it off as okay since there is no image loss. But ati's optimizations that don't affect image quality are according to him invalid . Its typical bs .
Oh JVD, We're discussing texture stage optimisations which are irrelevent to trilinear optimisations. You still have yet to distinguish the difference between the two, Both ATI and Nvidia have been doing texture stage optimisations for over a year now, which are irrelevent to the new trilinear optimisations discovered. You seem to be under the impression these texture optimisations are same. And the fact both ATI and Nvidia employ bilinear on texture stages 1-7, Makes them almost completely comparable. Making the entire issue of turning them off and one, completely moot. I have been very polite with you, And each time I try to respond to you. You throw out a personal attack on me and question my credibility, When I have gone out of my way to at least show you what I am talking about. In previous threads I have pointed out what texture stage optimisations do as well.
I know exactly what the diffrences is between the two kinds of optimizations. Fact is if the program asks for full trilinear than you will get full trilniear from the video card. If it does not ask then u will not get it . Nvidia allways foces bilenear on the stages after the first if you put the opitmization on So how do you compare that. Ati is using smart optimizations that will turn themselves off when the developer requests it or will turn it off when it results in image quality reduction. Nvidia does not do this . With its texture stage optimizations they are either off or on and its up to an uneducated (in these matters ) to judge when they should leave them on or off. This is not better than the programer choosing when it should be on or off like they can with ati . With trilenar filtering. Nvidia is either tirlenar quality or less than trilenear quality . With ati its allways trilinear quality as if it would affect image quality it would shut it self off . Yet you are claiming that nvidia's methods should be turned on when they do not result in the same quality as ati's You further go on to say that hey . fp16 and less is no problem and hey it just gains them 3-4fps . So who cares if they allways do that and ati never does. We should give nvidia as much advantage as we can by putting on all the optimizations on even when they will most likely reduce image quality compared to ati's versions and go a step further as to allow shader replacements , fp 16 or less in games. Because it provides no image quality. YOu claim this is the only way to get a fair comparision. I call it bullshit. Its the only way to make nvidia look better. THe fact is ati has optimizations that are "smart" and will turn themselves off. Nvidia does not have this . They leave it up to us to choose. Driverheaven choose to turn off the optimizations and or leave them at defaults . Though since nvidia does not allow us to force full percision on at all times it can never be a fair comparision with ati that does force full percision all the time. As I said its pointless talking to you. You obviously want nvidia to be painted in the best possible light . That is fine. ITs what I expect you being a memeber of nvnews . I wont fault u for it as if we were talking about powervr i would try and paint it in the best possible light aswell.