[Jawed]

Originally Posted by Pete

While I know this is reality, I'd just to be clear on this technicality: FP blending (and thus "HDR" as it's known/implemented in current games) isn't a required part of the SM3 spec, right?

Correct.

Jawed

[Karma Police]

Let me ask a noob question: How can an internal 512-bit bus improve performance? Also, what is a "ring-bus"?

[Kombatant]

Originally Posted by trinibwoy

Adaptive AA? What could that possibly mean?

Nobody pays attention to signatures anymore

[Dave Baumann]

I think we may have already seen a form of "Adaptive AA" not so long ago...

[Chisholm]

Originally Posted by Dave Baumann

I think we may have already seen a form of "Adaptive AA" not so long ago...

Mmm... Does that mean some form of Transparency AA for the R520?

[Chalnoth]

Originally Posted by Karma Police

Let me ask a noob question: How can an internal 512-bit bus improve performance? Also, what is a "ring-bus"?

Well, I have no idea what the "ring-bus" moniker means, but an internal 512-bit bus is nonsense: I'm sure that all video cards that have had 256-bit DDR memory interfaces have had 512-bit internal busses to their memory controllers. If they didn't have 512-bit internal busses, they'd either never get to use most of the bandwidth (which would mean we would never have seen a benefit from the 256-bit busses), or the memory controllers would be running at typically more than twice the clockspeed of the rest of the core (which seems rather silly).

[Tim]

Originally Posted by Pete

Edit: Dave, did ATI predict some quads would fail to run at the desired speeds, or that they'd be otherwise (perhaps more traditionally) defective? This seems to be new to GPUs, for a company to bet on partial-core failures yet whole-core success.

I think you need to fine tune you sarcasm detector. He was pointing out that the R520 was 16 pipes from the get go, there are no pipes disabled on 16 pipe r520s.

[Ali]

Why are the new cards called X1800? Whouldnt it make more sense if they are XI800, as in roman numeral 11 (XI) then 800?

We had 7500, 8500, 9700, X800. Gives us a 7,8,9,roman 10 as the first character. 18 doesn't make any sense to me.

Also, the Parhelia was the first chip with the 512bit bus. If everybody else had it then they would have made a big deal of it way back then. Although it does seem odd that the internal bus would have a smaller apparent bandwidth than the external memory bus.

Ali

[Geo]

Originally Posted by Jawed

Well, I'm pleased as punch to hear about "ultra-threading" and the ring bus

Adaptive AA, that sounds groovy too

Jawed

YEEEESSSSS!! IT GOES ROUND IN CIRCLES! Ahriight!

Edit: Sorry. . .but when you've been stalking a hint for nearly a year, you can get a little excited about these things. . .

Edit II: And for those who like to doubt that there is signal in the noise around here: http://www.beyond3d.com/forum/showth...ght=token+ring

[Geo]

Anybody notice this article also says the exact opposite of the previous leak on where the 512mb cards will be? This one says the CrossFire versions will be the only 512mb cards. Which still doesn't make sense. C'mon, Wavey, can't we get a little clarification on that?

Also, now that we've confirmed the ring bus (oh, happy day!), would it be fair to say you just don't do that unless you've done something else interesting that it is intended to support?

[Arun]

Originally Posted by geo

Also, now that we've confirmed the ring bus (oh, happy day!), would it be fair to say you just don't do that unless you've done something else interesting that it is intended to support?

Well, there is the VS using the PS' sampling units at least... Maybe more, but I can't see what personally. This also has the nice indirect advantage of giving native bilinear/trilinear/AF/etc. support for the VS' texturing operations; the G70, on the other hand, only supports point sampling in the VS, and (bi)linear has to be done manually at a huge performance cost.

Uttar
P.S.: Regarding the 512MB boards, I would assume that to be AIB-dependant.
Regarding adaptive AA, this could mean a huge number of things; Transparency AA comes to mind as Dave says, but 3DLabs and Falanx both also boast adaptive AA-related technologies that have nothing to do with that. And considering the FP16 AA support and the video memory cost of that, I wouldn't be surprised if it was more similar to the 3DLabs "adaptive AA" algorithm.

[tEd]

Originally Posted by Dave Baumann

I think we may have already seen a form of "Adaptive AA" not so long ago...

The only one i would put in this categorie is 3dlabs super scene AA.

[Arun]

Originally Posted by tEd

The only one i would put in this categorie is 3dlabs super scene AA.

NVIDIA's Transparency (SS) AA would also be adaptive, as determines whether it should be using SSAA instead of MSAA on a per-pixel basis.

Uttar
P.S.: I'm curious how that works in the traditional "quad" scheme actually; hmm... Since there could suddenly be more than 4 pixels that would theorically have to go to a quad engine.

[tEd]

Originally Posted by Uttar

NVIDIA's Transparency (SS) AA would also be adaptive, as determines whether it should be using SSAA instead of MSAA on a per-pixel basis.

Uttar
P.S.: I'm curious how that works in the traditional "quad" scheme actually; hmm... Since there could suddenly be more than 4 pixels that would theorically have to go to a quad engine.

then you could add temporal AA also as adaptive to the framerate

[Chisholm]

Originally Posted by tEd

then you could add temporal AA also as adaptive to the framerate

The problem is that Temporal AA is hardly a new feature. I don't really see ATI advertising it.

[Joe DeFuria]

So, no mention of kaleidoscope with any of these leaks? What happened to that...or is this Avivo thing the official marketing name for kaleidoscope...whatever that is.

[no-X]

Originally Posted by kyleb

Hopefully what I have been wanting for years; you set the minimum desired framerate and the level of AA adjusts itself to meet that goal.

Wasn't this called dynamic AA? Adaptive AA was something different (test and determination of specific AA mode for each pixel... or something like that)...

[Rys]

Originally Posted by Joe DeFuria

So, no mention of kaleidoscope with any of these leaks? What happened to that...or is this Avivo thing the official marketing name for kaleidoscope...whatever that is.

Yeah, that's pretty much spot on. It also had another name in between Kaleidoscope and AVIVO, if I remember rightly. I'll dig out the back story to that, it's worth a little writeup.

[Mark]

I don't think "Adaptive" AA is the best name for it.

[Kombatant]

Originally Posted by Ratchet

I don't think "Adaptive" AA is the best name for it.

Actually, if you think about it, that's exactly what it does. So I disagree.

[Mark]

Originally Posted by Kombatant

Actually, if you think about it, that's exactly what it does. So I disagree.

If I told you water was wet you'd disagree.

[digitalwanderer]

Water is only wet at certain temperatures....

So Kalidescope/AVIVO is just some video playback stuff? BORING!

Give me more features, more pipelines, more precision, and a bigger cooler....

[Dave Baumann]

Adaptive means to change dependant on the circumstances.

[Mark]

Originally Posted by digitalwanderer

Water is only wet at certain temperatures....

Yes but then it's known as ice or vapour...

Originally Posted by digitalwanderer

Give me more features, more pipelines, more precision, and a bigger cooler....

I think you'll get all those, though to varying degrees.

[Kombatant]

Originally Posted by Ratchet

If I told you water was wet you'd disagree.

Actually water is never wet; water's "influence" on other materials make them "wet"

But, anyway...

Originally Posted by Dave Baumann

Adaptive means to change dependant on the circumstances.

What he said.