AMD confirms R680 is two chips on one board

[Albuquerque]

So either Kyle is only talking about a few games, or the 3870X2 has some magic hardware voodoo, or the Catalyst drivers for the 3870's are about to get a big bump in scaling performance.

Because I don't see how any other option for the 3870X2 to do what Kyle is mentioning in that thread. Hopefully it's all drivers, so CF 3870 folks will see the same performance increase.

 

[Unknown Soldier]

Thanks. Thats great, right in the middle of Winter-een-mas.

BWHAHHAHAHAHA ..

CAD ftw.

US

 

[Unknown Soldier]

The only way the R680/R700 series can be a real success, is if the game developers actually start coding for DX10.1. That'll be a huge advantage for the R6/7 series if Nvidia don't release a DX10.1 card.

US

 

[pjbliverpool]

Yet when people politely explained the situation to you, you ignored what was presented and continued along the line of "either Xenos didn't benefit greatly from having all that extra bandwidth or that RV670 was bandwidth limited on at least a partially regular basis," which would be wrong.

Furthermore, this very topic has been discussed, at length, on this very forum. You could have bothered to read the Xenos article, searched the console forum, taken this topic to pms, or made your post as a new topic in the console forum or even this forum.

Instead, we have 4+ pages reaching the stunning conclusion that developers make the most of the hardware that is available to them.

First of all, I mentioned many times that I want to learn of any technical differences between the two which result in different bandwidth requirements. Making a partial and selective quote of what I said doesn't change that.

Second, I have read the Xenos article, several times in fact. It doesn't answer what I asked. Nor have I ever come across a post or thread that answers this specific question and if it does exist, I would be extremely unlikly to find it using the serach function given the many thousands of posts which discuss Xenos here.

Finally (and once again) your "conclusion" about what we have leared is completely wrong. It had zero to do with learning if "developers make the most of the hardware that is available to them" which is frankly a ridiculous suggestion to make given how obvious the answer is. What we have learned is that 256GB/s on Xenos is not at all comparable to 256GB/s of bandwidth on other GPU's - as many other discussions have made out that it is. Think R5xx series and RSX comparisons. And most importantly, we have learned the technical reasons why Xenos uses bandwidth differently to a more traditional GPU.

 

[Mariner]

The only way the R680/R700 series can be a real success, is if the game developers actually start coding for DX10.1. That'll be a huge advantage for the R6/7 series if Nvidia don't release a DX10.1 card.

US

Erm, the only way it can be a success? Surely if it outperforms the competition at lower prices it might be a success as well?

Not that I think this will necessarily be the case but it seems daft to say that DX10.1 will the be all and end all when we don't really know everything about the new stuff coming from either IHV.

It must be said that if NVidia don't release DX10.1 cards then I can't see many developers putting extra effort in to support it in the short term so I doubt it will be much of a factor for the next generation of GPUs.

 

[I.S.T.]

The only way the R680/R700 series can be a real success, is if the game developers actually start coding for DX10.1. That'll be a huge advantage for the R6/7 series if Nvidia don't release a DX10.1 card.

US

That would make them a success a few years down the road...

 

[Unknown Soldier]

Ye, but if DX10.1 does indeed have a performance advantage over DX10 .. and developers start coding for it, then ATI will have the advantage. But if they don't then Nvidia should still have the ball in their court(unless of course the R680/R700 beats the 9800GTX/GX2.

Of course it also depends on Nvidia's G200 or whatever it's gonna be called down the line.

US

 

[Albuquerque]

I still don't understand the fascination with DX10.1 -- it's an incremental features update over DX10.0, so I don't see how you're expecting it to be faster. If anything, it might even be slower because of the extra features.

Having "checkbox" functionality for a DX spec really isn't any way to be successful. Rather, being performant in current and and near-term games is the way to win marketshare, having a cost advantage will also certainly help. And being energy efficient may further assist with sales to large OEMs like Dell, HP and IBM.

I think your definition of "real success" is quite different from reality.

 

[Dave Baumann]

There are features in DX10.1 that can increase performance, such as the MSAA handling with deferred rendering and cubemap indexing (if an app uses large amounts of cubemaps).

 

[I.S.T.]

I still don't understand the fascination with DX10.1 -- it's an incremental features update over DX10.0, so I don't see how you're expecting it to be faster. If anything, it might even be slower because of the extra features.

Having "checkbox" functionality for a DX spec really isn't any way to be successful. Rather, being performant in current and and near-term games is the way to win marketshare, having a cost advantage will also certainly help. And being energy efficient may further assist with sales to large OEMs like Dell, HP and IBM.

I think your definition of "real success" is quite different from reality.

Indeed, although I wouldn't put it quite so... bluntly.

DX 10.1 making R680/700/RV670 far better won't happen until games that use its' features heavily/require it come out, and that will be years from now. By then, the boards will probably be too slow, as is usually the case(I imagine the GeForce 6800 doesn't run Bioshock too well...).

 

[Albuquerque]

There are features in DX10.1 that can increase performance, such as the MSAA handling with deferred rendering and cubemap indexing (if an app uses large amounts of cubemaps).

Duly noted. I'll keep an eye out for Humus or someone else to perhaps make a demo of this, as I'd like to give it a try on my CF'd 3870's

 

[Dave Baumann]

Actually, I was kinda hoping that the final version of the Ping-Pong demo would allow you to turn off cube-map indexing, however it looks like that won't make it in due to time constraints.

 

[Albuquerque]

I will say one more thing about my previous reply:

Having "checkbox" functionality for a DX spec really isn't any way to be successful. Rather, being performant in current and and near-term games is the way to win marketshare, having a cost advantage will also certainly help. And being energy efficient may further assist with sales to large OEMs like Dell, HP and IBM.

This is all true, but I was not trying to suggest that the 38x0 series of cards doesn't meet these criteria. Quite the opposite, the 38x0 series really does impress me with price, performance and power efficiency.

Now if I could just get the CF scaling like what Kyle was hinting at, I'd be in nerd-heaven.

 

[Shtal]

Actually, I was kinda hoping that the final version of the Ping-Pong demo would allow you to turn off cube-map indexing, however it looks like that won't make it in due to time constraints.

Sir! Do you have any screen shots yet!

 

[ninelven]

What we have learned is that 256GB/s on Xenos is not at all comparable to 256GB/s of bandwidth on other GPU's - as many other discussions have made out that it is.

Nor have I ever come across a post or thread that answers this specific question

1) Red Text: So either you were either blatantly lying or you just ignore responses you don't like... (I would really like to see these "many other discussions" that presumably go beyond random fanboy babble)

2) We didn't learn anything... you may have learned something... I definitely learned something, but I very much doubt it was the same "thing." Or was there someone else in this thread agreeing with you?

3) The Xenos article does answer you question if you spend more than 5 seconds thinking about it...

 

[pjbliverpool]

1) Red Text: So either you were either blatantly lying or you just ignore responses you don't like... (I would really like to see these "many other discussions" that presumably go beyond random fanboy babble)

2) We didn't learn anything... you may have learned something... I definitely learned something, but I very much doubt it was the same "thing." Or was there someone else in this thread agreeing with you?

3) The Xenos article does answer you question if you spend more than 5 seconds thinking about it...

See, that just proves that you did in fact completely miss my point. Saying I have read discussions comparing Xenos to other architectures were its superior memory bandwidth was touted as a major advantage and saying I have read posts which explain why Xenos' huge memory bandwidth actually isn't that much of an advantage are not only completely different things, they are pretty much exact opposites.

As for your "We didn't learn anything... you may have learned something...", I didn't realise people were no longer allowed to ask questions and learn at B3D. I guess you would prefer it if anyone with less knowledge than you be ridiculed for asking a question you already know the answer too, yes?

Say, if you did indeed understand what I was asking and you already knew the answer, why didn't you just reply to it in the first place? (like several others eventually did).

And no, the answer to my question - why does Xenos need so much more memory bandwidth than other architectures to produce similar results is not specifically answered in the Xenos article. Parts of the article may hint at reasons if you know what your looking for but thats far from a directed and fully articulated answer.

 

[swaaye]

Indeed, although I wouldn't put it quite so... bluntly.

DX 10.1 making R680/700/RV670 far better won't happen until games that use its' features heavily/require it come out, and that will be years from now. By then, the boards will probably be too slow, as is usually the case(I imagine the GeForce 6800 doesn't run Bioshock too well...).

Yeah and Doom3 is the poster child of R200's extra capabilties. LOL. Extra features your competitor lacks and devs ignore don't help you in the end. It also doesn't help when your performance is lower than your competitor in general, even with "beneficial" extra features, as in R200's case. Usually ends up that the next generation does those extras a lot better, too. And hardware cycles are so short in this industry...

It's nice that they fleshed out what was probably almost already there in R600, but it doesn't mean anything at all IMO for the future of the RV670 cards. What matters is where they perform now, after probably 10 months of serious driver refinement now. They aren't going to get faster.

Doubling them up on a board might be neat, if Crossfire actually works well across all games. Power usage of a dual RV670 board should only equal one 2900XT. Heh.

 

[ninelven]

I guess you would prefer it if anyone with less knowledge than you be ridiculed for asking a question you already know the answer too, yes?

I never said anything like that; I did state that this thread was not the place for it and there were certainly much better venues to pursue such a topic... and I really have a hard time believing you didn't already know the answer to the "question" you were asking anyway. Maybe I'm just too cynical...

you did in fact completely miss my point

As for this, no - I just found your point trivial.

Say, if you did indeed understand what I was asking and you already knew the answer, why didn't you just reply to it in the first place?

By the time I entered the thread, you had already been provided several satisfactory responses.

not only completely different things, they are pretty much exact opposites

Generally on this forum (and most any human interaction for that matter), when one thing is discussed, the opposite is as well.

 

[I.S.T.]

Yeah and Doom3 is the poster child of R200's extra capabilties. LOL. Extra features your competitor lacks and devs ignore don't help you in the end. It also doesn't help when your performance is lower than your competitor in general, even with "beneficial" extra features, as in R200's case. Usually ends up that the next generation does those extras a lot better, too. And hardware cycles are so short in this industry...

It's nice that they fleshed out what was probably almost already there in R600, but it doesn't mean anything at all IMO for the future of the RV670 cards. What matters is where they perform now, after probably 10 months of serious driver refinement now. They aren't going to get faster.

Doubling them up on a board might be neat, if Crossfire actually works well across all games. Power usage of a dual RV670 board should only equal one 2900XT. Heh.

Uh... that was my point. You just said what I said only with far more words and with examples

 

[swaaye]

Uh... that was my point. You just said what I said only with far more words and with examples

Yeah I know! I thought you needed a friend! LOL!
(digitalwanderer is my inspiration!!!)