R420 Hints from people under NDA!

Commenting on speculations on un-available products from drooling gaming fanatics is part of a " sound business strategy"? I see your "please" and raise you one.

The 420 may not indeed support PS 3.0, but it's extremely specious reasoning to presume that it's been anywhere close to "confirmed" by ATI largely because they refuse to comment on the ridiculous array of rumours that perpetuate any geek-product on the net just before its introduction. This is especially bizarre reasoning when your competitors equivalent product isn't even available yet either - what has ATI to gain from commenting on rumours when the actual release of their product is but a few weeks away?

Developers know the feature set, as do the important OEM's already. I seriously doubt that ATI is concerned about the 10 Nvidia fanboys who have pre-ordered 6800 Ultra's based on the speculation that the 420 may not have PS 3.0. There's simply no reason for ATI to dispell of confirm any rumours about the 420 at this point, if anything it would setup a bad precedent that forces ATI to comment on future rumours.

 

I gaurantee you the R420 does not support PS 3.0. Otherwise ATI wouldn't be downplaying branching (some branching support is requird for PS 3.0), and they wouldn't have been talking about the PS 2.0b profile at GDC (a profile which no current hardware supports).

I just don't see any possibility for PS 3.0 support given the above.

 

Well since its not that much of a leap forward from the ps2.0, I'd say it might be for the 9800 series with their f-buffer etc?
But then, I guess they'd have been mentioning publicly that they'd gotten M$ to make a profile for the 9800 series...

 

ah the wonders of cut/paste

thoroughly cruel and yet extraordinarily entertaining

-edit-

in reference to HB's quote's from the vault

 

Then I guess they don't feel they should release some information.

Maybe because when ATI say they don't comment on unannounced products, they actually mean it, as opposed to Nvidia who say it and then leak rumours all over the internet?

 

I'd say that's enough confirmation of no PS3.0 support in R420...

 

There are countless other posts trying to decipher this 'note'. Questions such as what does it mean (flow control or ps3.0) and main extra feature, and shows up with decent performance are all open to subjective perception of the context. Also this could have been a pure Black Intelligence stunt as well.

The Bottom line is we will know in a week won't we and all of this bickering over it is pointless. ATI could be keeping quiet for three reasons.

1)They don't have it and want to downplay PS3.0 not up play it.

2)They have it and don't see any advantage of releasing information when they believe that they will hit the market first and announcing it as a suprise would burst the NVIDIA Bundle.

3)They have outstanding shader performance without it and will best NVIDIA Benchmarks even without branching or longer instructions and the Difference now is negligible.

Or it could be anything until we see the cards perform side by side I don't know what to tell you.

 

Or maybe NV and ATI will both support it..and when it is found out the neither card can run it very fast, if at all, they can at least say "we told ya so" Instead of do what NVidia did last year when it was shown that their famed FP32 was unplayable and say " Well Flow Control (FP32) is really not all it is hyped up to be in PS3.0..As you can get the same result in SM2.0(FP16) So if it is good enough for SM2.0 it is good enough for us.

(This is in reference to the comment made BY NV about how a popular company who makes graphics for the big screen and did movies like the toy story used only FP16..and if it was good enough for them it was good enough for us, and FP32 was really not needed nor important.)

Of course now that they can do FP32..they are back to saying FP-32 is the latest and greates..They change their tune to fit their needs..

 

My bolding. Given what has been speculated about some of the SM3.0 features being in R420, it *could* mean that branching is one of the few missing things. Given the performance hit that NV40 supposedly takes from branching and Nvidia's recommendation to avoid it, lack of branching may not be a big deal if ATI offsets it with higher speed (at least for this generation).

Bear in mind that there was so much that was optional in SM2.0 that you could miss out a lot of functionality and still say you are SM2.0 compliant. However, in SM3.0, all functionality is manditory - if you miss out one part of SM3.0, you cannot call yourself SM3.0 compliant at all, regardless if you support all the most useful features.

I think we'll have to wait to see, but I suspect ATI is not going to simply stick at SM2.0 . They might add enough extra SM3.0 functionality to be useful without going all the way to SM 3.0 (which I don't think they could do without FP32 anyway).

 

I was under the impression that flow control aka branching is one of the chief elmements of PS3.0, and will most certainly be recognized as "PS3.0 itself" by the less informed ( = many ). Thus, no flow control would mean no PS3.0, right?

As for Black Intelligence... yeah, possible, but not likely. I can understand that ATi would rather not bother about smashing the "no PS3.0" rumours, but I have a hard time believing that they would actively spread them.

 

The strange thing is, the only real problem I remember having with my Xpert 128 was corrupt textures in MBTR. That's it. This on a Cel266@400 on Win98, not exactly the foundations of a stable system. So it's really a toss-up, IMO, though it does appear that anecdotal BBS evidence puts nV ahead of ATi in that time-frame WRT driver quality.

BTW, how can drivers be "junk" in an office machine? Were you using it for anything more than typical office work (read: Word/Excel/other 2D work)?

 

Either way, without flow control, you don't have PS 3.0 support.

 

It seems to me that most of the cool effects will be achieved with VS3.0, so if the R420 has kept the Vector processing side of the original R400 design, we'll see most of what Shader Model 3.0 has to offer.

The problem is ATI wont be able to put it all over their boxes like nVidia will undoubtedly do.(E.G To claim Shader Model 3.0 compliancy,you must support everything.)

Of course that won't matter so much to people like us who want to know every last detail about these GPU's, but to JOE BLOGGS who walks in to Harvey Norman(equivalent of Walmart I Suppose) the 6800 will appear to be better than X800 because it has a higher number on the box e.g SM3 v SM2.b.

EDIT: Spelling

 

You're right, they're bound by the terms of their employment--something they might want to keep. ATI clearly doesn't want the cat out of the bag, so to say that ATI could let it is out is meaningless.

 

I would appreciate it if people would try not to keep pointing this out. I don't think it holds a whole lot of water as an argument at the moment.

 

Was this quote lifted from some "leaked" document that nobody can verify actually originated with ATi?

I ask because the statement itself seems contradictory and illogical: if "flow control" is something ATi thinks "hurts badly" in regard to performance, to the extent that ATi would consciously choose not to implement support for it in R4x0 hardware, it doesn't follow that it would therefore even be something ATi would want in R5x0, does it? Also, if we assume that the phrase doesn't mean "hurt performance badly," but means "the performance of flow control is so good it will hurt ATi badly," then on what basis might Huddy expect to successfully "steer people away" from it? That would be like asking people to take nVidia's word that ps2.0 stinks on R3x0, which is exactly what nVidia tried unsucessfully to do for most of last year. It would be futile. Also, since Huddy himself would know practically nothing about nV40's ps3.0 "flow control" performance, he could not possibly know that it could "hurt" ATi at all...

But the comment about R5x0 "showing up with decent performance" seems only to buttress the fraudulent nature of the quote, as this implies Huddy doesn't think R4x0 will have "decent performance." Yet, ostensibly, Huddy's reason for ATi not to support "flow control" in R4x0 is solely to improve performance. So I would conclude that the statement itself is falsely attributed to Huddy and/or ATi, and that instead its source is someone desiring to cast dispersion on R4x0 (but not doing it very coherently...)

 

In other words, we might not know. . .but Richard Huddy does, and he's the one who wrote that ATI doesn't have branching.

 

I'll verify. I got my copy of this presentation from ATI.

It's simple Walt. ATI most likely didn't implement branching in the R420 because of schedule constraints. Huddy's trying to say "look, we didn't have enough time to implement branching with proper performance, and neither did NVidia, but let's suggest the R500 will have it good, and Nvidia's NV40 branching performance sucks." He's trying to discourage people from using a feature ATI doesn't have until the R500. He's saying to developers "Stick with us. Do not be blinded by features like dynamic branching. Atleast until we have them." Want simpler? ATI suggests NVidia's dynamic branching is too slow to use, wait for better ATI chip.

I take issue with this, because dynamic branching is a performance hit, but 95% of the time you don't need it (most branches in shaders are small, and predicates work fine) But when you absolutely do need it, it's a big performance win, and it sucks not to have it. Utilizing dynamic branches is about knowing when to use them.

 

For me, the better evidence is that PS2.0b doesn't support branching, gradients, predicates, nor arbitrary swizzle. You'd think that if the R420 supported something between PS2.0 and PS3.0, then they'd support a compiler profile that exposes all those features.

 

That document came right from ATI's website, a powerpoint that they didn't realize the comments section had been left in because they did not show in regular web mode. But save the powerpoint to disk, open in Powerpoint, and there it was with a whole bunch more. After the thread here the presentation got pulled down from ATI's site in a hurry.

If I recall correctly, the thread here got pulled too out of courtesy --also an indication it was the real deal.