NV30 launch *DATE*?

[Bigus Dickus]

Since laughing at a post apparently isn't allowed, I guess I'll have take the time to rip apart the thread that I found so humorous. :rolleyes:

Originally posted by gemini1313
an architecture that is much more efficient and highly optimized as compared to r300 with more features to boot.

Would you care to elaborate on what features the NV30 will boast, as compared to the R300? Are you going off of inside information, wild speculation and rumors, or just wishful thinking? :rolleyes:

Perhaps you'd also like to clarify how the NV30 is going to be "much more effecient and highly optimized?" Yes, we all hope that it has some neat trickery to further improve the effeciency, but at this point there is nothing concrete to base such opinions on. The 9700 already does color and z compression during bandwidth intensive circumstances (AA).

r300 will never be able to catch nv30 when it comes to running instructions and precision.And where did this genious come from? You know, many people seem to forget something: the long shader instruction capability is what can be done in a single pass. Perhaps they've completely forgotten that multipassing can render the same shader instructions on the R300 as on the NV30. Oh, what's that you say? But the NV30 will be faster because of single pass capability for enormously long shader routines? Having the ability to execute an instruction of a given length doesn't mean it has the ability to do it faster than multipassing. The pixel shaders in general could be slower. Also children, don't forget that the longer the shader instruction, the smaller the relative penalty for multipassing.

And, just in case you're as dumb as you sound, do you have any idea how fast a shader routine would run that was only the length an R300 could handle in single pass? About as fast as UT2003 on an integrated Savage4. Maybe. On a good day. At extremely low resolutions. Being able to single pass a 500+ instruction shader routine really means jack squat to us, since it will be completely unusable on that card, ever, for games.

And then there's the "precision." I suppose you can see a difference between 96 bit and 128 bit color? No? Oh wait, your monitor can only output 32 bit (24 RGB, IIRC). Oh, wait again, you're suggesting that the lowered precision will degrade worse on lengthy shader routines. And then we're back to that argument. Do you have any idea how many instructions would be needed on 96 bit floating point color values to cause an accuracy error that is below the threshold of any consumer output device? And do you know how quick that shader would execute? (Refer to Savage4 comparison for reference). This is completely irrelevant, because 96 fp bit color (with 128 bit texture reads, IIRC) is such a leap above what our current display devices can output (as well as what the human eye can detect), that it simply doesn't matter. 96 bit fp is better than what Pixar renders movies.

after all r300 is just a thrown together card that is barely anything more than an 8500.Barely anything more than an 8500? How about fixing a few things that were missing on the 8500, like a lack of MSAA, or the lack of trilinear filtering in combination with anisotropic filtering. How about the things it does that no other consumer card does? How about RGMS at 6X, or gamma correct FSAA, or floating point color precision, or 128 bit precision, or support for shader branching, etc. How about the fact that it's 2, 3, 4, 5 times faster than the 8500 (or any other consumer card on the planet, for that matter)?

Yes, thrown together indeed. ATi are simply magicians. They have "thrown together" a card that is as big an improvement in speed and features over the previous generation as the graphics industry has seen in many years.

any company can slap a 256bit (ripped from nvidia) memory controller on a cardAny company? Matrox tried it, and it didn't work out so well for them, did it? Has nVidia been able to just "slap on" a 256-bit memory controller? Nope, they haven't done it yet. Which raises the question: just how in hell do you figure that ATi "ripped" a 256 bit memory bus from nVidia??? Perhaps, if nVidia had done it first you might have an argument, but even in that case it's not like a radical feature that wasn't inevitable at some point. However, nVidia has never produced a 256-bit memory interface on a production card. How again did ATi steal something from nVidia which nVidia never had?

:rolleyes:

and slap some trannies on the core for dx9 pipesYes, any Joe Schmoe can just "slap on" transistors in any way they please. It doesn't even take much design effort at all. That's why there are dozens of companies competing in the consumer graphics chip market. :rolleyes: I can't help but think... if it was so easy to just "slap on" transistors to make an existing GPU DX9 capable, then where the hell is the almighty nVidia's DX9 part? lol. Any more genius statements?

however limited those pipes are.Limited? From what I can gather, they are equivalent to (what we think are...) the pipelines on the NV30 in almost every way. It's really such a shame that nVidia is having trouble producing their version of a piece of cake DX9 card with limited pipelines.

nv30 is .13u and a brand new core from the ground upYou really think the NV30 doesn't borrow any technology from previous NV chips? Really now... please be serious for a moment. The R300 is probably more of a step from the R200 than the NV30 will be from the NV25. Why? Because ATi finally added some things that were already in the NV25, such as MSAA, Tri+AF, and a good crossbar memory controller. Also, the R300 made the step to a 256-bit memory interface, and if the NV30 doesn't, it will be one more area in which the R300 made a greater improvement over the previous generation. Perhaps nV will surprise us with some exotic hybrid deferred renderer or multichip solution, but that's beyone speculation at this point.

(anyone should be able to see how the r300 is hardly revolutionary or a ground breaking product, what came out first doesn't mean jack...)And anyone (other than IQ challenged fanbois) should be able to see that, if the NV30 isn't markedly different from the R300 (like being a deferred renderer, or similar), then it will be even less revolutionary, since it is just nVidia's "me too" part.

nv30 will undoubtedly outpace r300 in every aspect. You better be right, or you will never live down the constant ragging you will receive.

[INNOVISION]

5 words:

IT'S JUST A VIDEO CARD

8 words:

SHUT YOUR PC OFF AND GO OUT SIDE

[thcdru2k]

woah..getting a little carried away are we. its not like it was a hilliarious joke. if you find post like that funny than got a warped sense of humour :cool:

You better be right, or you will never live down the constant ragging you will receive.

...your gonna rag on him for a post he made, online, on a forum. yeah..do what innovision said :)

[Bigus Dickus]

I refuse. I've been inside all day! and I'm not about to ruin the fun now!!!

Actually, you're entirely right. I'm in a bad mood because it rained all morning and I couldn't play golf. You see what sitting in front of a computer too long does? :p

[thcdru2k]

you live on west coast too..its been raining like crazy lately.

[borntosoul]

what i dont understand is color precision ,can a card render at 96/128 bit and then display it at 32 bit ?,why would nvidia and ati go for higher precision if monitors couldnt display it? how do ray traced pictures look so good without banding if we still them in 32 bit?can anyone clear this up for me

[StealthHawk]

Originally posted by borntosoul
what i dont understand is color precision ,can a card render at 96/128 bit and then display it at 32 bit ?,why would nvidia and ati go for higher precision if monitors couldnt display it? how do ray traced pictures look so good without banding if we still them in 32 bit?can anyone clear this up for me

if something is rendered internally at a precision of say, 32bits, then eventually errors start to crop up which would then be displayed on the monitor. higher precision helps to stop those errors from occurring, so that the proper image will be displayed.

hope that makes sense.

[Bigus Dickus]

Yes, it's a matter of internal precision during rendering.

It's becomes more and more important the longer the calculation being done, which is what pixel shaders are all about. They are just performing some various mathematics for each pixel, given an input color value and perhaps some other color values along the way (such as different texture layers). Using these color inputs, they apply whatever formula is required, and then output the result.

Think of this as something like balancing a checkbook. You start with a certain value, and then modify it by adding, subtracting, multiplying, or whatever with each new entry (think of each entry as another step in the shader calculation). Imagine if you were required to round off your result, at each step in the balancing, to the nearest $1. The longer your checkbook is to balance (the longer the shader program), the greater the error in the final value when all is said and done.

Perhaps, in an ideal world, the final value should have been $180,284.98 (no, that's not my checking account... I wish!) but your value, after all the rounding during each step, is $180,287. Not a huge error, but what if your checkbook is really lengthy? Perhaps your final value $180,349. That might be enough to worry about.

The "rounding to the nearest dollar" is like the 32 bit integer format we've been using (though, it's a little tricky to relate binary to decimal formats in this manner, but it's a close enough analogy to work I suppose).

A floating point color value is analogous to being able to use the nickles and dimes during each step, instead of rounding off to the nearest $1. Now, for a checkbook, you'd expect to get the perfect answer just adding and subtracting, but let's imagine you had some really bizarre bank that required calculating sine, cosine, square roots, etc. (don't ask why, just imagine). In that case, if you were forced to round each result to the nearest $0.01, then you might get a final answer something like $180,285.52.

That's pretty darn close to the "perfect" answer, but computer monitors have to round everything to the nearest $1 before display, so your value will still be off by $1 after rounding. Better than before, but not perfect. That's where higher bit depth floating point comes in. Imagine that 96 bit is like saying you only have to round to the nearest $0.0000001, and 128 bit is like rounding to the nearest $0.000000001. Just how long, and how complex would your checkbook have to be before your final result differed from the perfect value by more than $0.01? Pretty long. Perfectly analogous to pixel shader routines, except that a shader routine that long would run so slow as to be completely worthless (kind of like saying if your checkbook was that long, you'd die before you had a chance to spend your money).

Remember, displays round to the nearest $1 anyway, so there's a tremendous difference between 32 bit integer (or 40 or 48, or whatever the previous generation of graphics cards used internally) and 96 bit floating point. Only in rare cases, where say the "perfect" value was $1.49, and your calculated value was $1.50, would the final display rounding cause any errors.

And why did I write out that whole explanation? I have no idea... bored I guess. :)

[borntosoul]

lol thanx for that guys :)