AMD: R9xx Speculation

Aren't words like "disgrace", "shady" etc, and accusing AMD of fooling people a bit rash, considering they haven't actually announced ANYTHING yet? :rolleyes:

It wouldn't be called the speculation thread if we didn't! :cool:

So does the Swedish translation for the above look like:

Är inte ord som "skamligt", "skumma" etc, och anklagar AMD av lurar folk lite utslag, tanke på att de har faktiskt inte tillkännagivit något ännu? : Roll:

Because I think we should have a rule that all baseless speculation be translated into Swedish! :)
 
http://www.chiphell.com/thread-121384-1-1.html

Cayman XT:
fnhhz8.jpg
 
Really good post about the 6000 here:

http://forums.anandtech.com/showpost.php?p=30402647&postcount=497

Quick summary:
  • Northern Islands shader is 98.5% as efficient per clock but they can fit 25% more in the same space.
  • a new eyefinity
  • UVD3 (ati's gpu video decoder)
  • new shader model
  • thicker PCB for the chip packaging
  • changed the voltage regulators buck controllers
  • vapor chamber cooler
  • 6770 supposed to have relatively low clocks ~700mhz...
  • tripling or quadding up on the tessellation power
 
Sorry, Squill but the above sentence doesn't make much sense...

:LOL:

Haha, well you speak English far better than Google can translate to Swedish! :LOL:

OT: The 6300 looks pretty interesting, will they beef up the performance substantially in light of the fact that they are already going to be offering HD 5300 level performance in Ontario?

I wonder how the rumoured transition from 5P shaders to 4P shaders would work. If 80 / 5 = 16 SIMD groups then would they be looking to increase to 24 = 96 SP, 32 = 128 SP or 40 or 160 SP?
 
I wonder how the rumoured transition from 5P shaders to 4P shaders would work. If 80 / 5 = 16 SIMD groups then would they be looking to increase to 24 = 96 SP, 32 = 128 SP or 40 or 160 SP?
I think it would be about time to increase simd width to 16 up from 8 for the low end part, no matter if that's 4P or 5P. I don't think though you're actually expected to pair the low-end part with those new IGPs anymore (not as an alternative that is), hence if it's really faster or not might not be that relevant.
 
Im guessing they have to increase to 160 SP's at a minimum, if not 240. Sandy bridge looks like its gonna match a HD 5450 and Llano is going to be at least two times as fast. I think they have to target at least Llano's level of performance

PS I still stand by my earlier speculation that AMD might build GPU's on 32nm as they already have experience making a GPU on the process due to Llano
 
Squilliam said:
The 6300 looks pretty interesting, will they beef up the performance substantially in light of the fact that they are already going to be offering HD 5300 level performance in Ontario?
An "HD53xxx" series doesn't even exist to my knowledge ;)

As new low-end chip, Caicos should actually sit at a performance level comparable to Redwood, anything below that doesn't make a whole lot of sense. With early benchmarks indicating that even Intel's Sandy Bridge (let alone AMD's Llano) will make Cedar-level GPUs run for their lives, I'd expect anything below Redwood performance to be dropped out of AMD's GPU portfolio. There doesn't seem to be a whole lot of space to even think about releasing an HD 63xx card.

Why even develop such a card when your precious dev-resources are already scattered all over the place (Desktop APUs, Notebook APUs, Desktop GPUs, Notebook GPUs) and anyone looking for a Cedar-level card in the timeframe until 2011's APUs finally arrive can just buy exactly that - a Cedar-card?

My bet given the current rumours is on Caicos trading under the name of the HD 65xx/66xx series, with Turks and Barts powering HD 67xx and HD 68xx cards respectively.

Even if Barts was a high-midrange part (67xx), Caicos still shouldn't end up below HD 64xx in AMDs well-proven naming scheme.
 
Last edited by a moderator:
So which numbers are left for Cayman and Antilles then?
Let's face it: If based on the original 32nm-high-end-design (!), Cayman either is

(a) a 400mm2+ monster chip neither economically (yields!) nor power-wise (200W+) reasonable for an (ultra-)high-end dual-GPU-card.

(b) a shrink-design currently under development for 28nm.

In both cases, its performance should probably end up in the range of a dual-gpu card (Antilles) based on 2xBarts (= HD 69xx)

40nm Cayman in Q4 2010 would probably end up slightly slower (HD 6950?); 28nm Cayman in Q2 2011 would probably end up slightly faster (HD 6990?) than Antilles (HD 6970?).

Just my personal speculation, of course. If those new chips actually don't have too much in common with the original 32nm designs, I'm probably all wrong.

Plus, like some of you already pointed out to me, there still is that Beta-driver indicating that Antilles = CaymanX2 ...

I guess given the current kind of rumours, you'll have to take sides in favour or against some of them - and I'm in favour of economical reasoning (i.e. HD 6xxx series will be heavily based on the previous 32nm designs, i.e. the "old" 32nm high-end-chip can't hit the "sweet-spot" @40nm, i.e. the "old" high-midrange chip will probably be used for a HD 6970 dual GPU card as well as the 68xx cards, i.e. any "insiliconation" of the original 32nm high-end design will, naming scheme wise, have to be placed in the 69xx territory).

Not gonna happen with a 64-bit DDR3 setup.
Why should I believe the rumoured specs of a card whose actual existence and economical feasibility I question in the first place? Seriously, if they really wanted to release another card in that price-segment using the 40nm production process, they'd be better off just taking Cedar and adding a few extra-chip goodies to rectify the 6xxx name tag ...
 
Last edited by a moderator:
Back
Top