AMD: R9xx Speculation

I guess you are right after all .

Anyway , we shouldn't argue much into these leaks ,they are dodgy .. we should save the words for something solid .

 

mm.. that leak doesn't sound right. 960 sp of Bart XT can beat 1440sp of Cypress PRO?

 

http://semiaccurate.com/forums/showthread.php?p=71796#post71796

 

Dave's hi(n)ts are sorely missed these days...

 

Apparently AMD has redone the shaders and Bart's 4D ALU is more effecient than Cypress' 5D ALU.. or is that a new Bart 5D ALU is better than Cypress' current 5D ALU.

Something like that.

http://forum.beyond3d.com/showthread.php?t=55313&page=76

 

Is the core clock same?

 

Well, the drivers had both Antilles Pro and XT, but then again, IIRC there was 2 entries for Hemlocks too with only 1 card on the markets in the end

 

Remember those 960SPs perform @ 98.5% of the 5-wide Evergreen ALUs, so 960 NI SPs @ 850MHz run something like 1385 Evergreen SPs @ 725MHz. Assuming increased texture/pixel fill rates and bandwidth due to the increased clock rates should make up for the minor math rate deficit.

 

Ok, we got all the bases covered now, (Barts Pro) from barely faster than a 5770 to almost as fast as 5850. We are prepared for everything.

I wonder about the positioning of the 5XXX series cards, price-wise and how they are going to market them. If they get a price cut, 58XX could be for example marketed as best DP FLOPs/dollar card. Or something.

 

http://www.digitimes.com/news/a20100927PD228.html

Is it Northern or Southern islands on 40nm ??

 

interestingly, thats the same mistake BSN made, is it not?

 

That whole article is wrong, just like nordichardware has hardly ever been right, or Copper for that matter.

 

That could leed to something like...

- Old 57X0 renamed to 67X0 (OBRs story)
- Barts Pro to fill the cap between 5770 and 5850 via 6850
- Barts XT nearly same as 5870 via 6870
- Cayman better than 5870, nearly 5970 via 69X0
- Antilles better than 5970 via 6990

Thats just too crazy I hope it's all fake in NordicHardware article.

 

"Fake talk aside" I believe AMD should back off in regard to die size to HD4xxx standard.
At this point I don't think absolute perfs should be their only concern as they still would have no problem to be competitive with Nvidia ( a hypothetical HD5750x2 would be more than enough to beat Nvidia most compelling product the GF460).
Profitability is critical to AMD, I hope they didn't give up on small chips and a regular update (especially when Nvidia has nothing really compelling under 200$).

 

Agreed,

AMD suddenly stepping away from their sweet spot strategy would be very awkward. But I don't think that people would be satisfied if they got ~5850 performance from a Juniper sized die because "the name is wrong!!!11."

Even if the 5770 remained and wouldn't be re-branded but the 68x0 and 69x0 would slot in above it, people would still complain.

 

Someone seemed to hint that the HD5xxx would stay in the market. I'd take a guess and suggest it's the new 5770 and 5830's that would stay(or at least that's what I read into it. However I won't guess on the new price range.

 

There is no speculation about a sub-200mm² 128bit part, so we can expect, that until they release anything like that, Juniper will be with us

 

Turks... 1H '11.

Oh, if only he had the most recent info.
So either Charlie doesn't know everything or more than likely Charlie just isn't telling him everything.

 

Of course we will complain. What do you want? Just to sit, "enjoy" the second life of some chips which are morally old and need to be replaced by something new. 5770 is nothing interesting and performance wise it sucks. Especially in DX11.

 

If the performance predictions of nordic are anywhere close to accurate AMD's marking department is going to have a fit given they now have to explain why the new generation parts of the same name are going to have little to no performance gain on the old.