AMD: R9xx Speculation

Considering this, Juniper should have been marketed as HD5800.

On top of these improvements, it offered full DX11 capability over HD4870 and power draw was down by ~50%.

Is it so hard to find a good example that you show the perfect opposite?


And I think I already said "economics" wasn't in favor of such a move... The goal is not to pile many references in a $10 range with the most expensive one $30 more expensive than the cheapest one at the production level, leading to a situation where you force the client to reduce your margin.

To counter that, the opposite move is preferable: $100-150 75-watt "6", $200-250 150-watt "7", $300-400 225-watt "8" and a $600 300-watt "9" using twice the "7" part has nothing shocking. This leaves "5" and "4" as entry/value and 3 digits for the integrated stuff.


Of course... nothing is known yet (not even are we certain the naming scheme will persist), it's stupid to state Barts can't be called "8" on its own merit and already tag it as a fail, but if it's indeed "8" and the mainstream part, it's going the wrong way, however you look at it.

 

Did anyone of you who are scared that someone might be confused about the 6870, that they might be even more confused about the name change from ATI to AMD?

They might wonder if they should buy a Nvidia GTX 480, an ATI 5870 or an AMD 6870 when standing there in the store...

 

Both are branded as Radeon HD though...

Edit- If Barts really is 960/800 and 2 RPEs... that puts Cayman at 1440SPs again if it is 3RPEs. (I know those both are bigs IFs)
Which would line up with the early rumor that it has less shaders than Cypress but Charlie seems adament that it does indeed have more than Cypress.

Oh silly season, you constantly mind-f' me... but yet I love you so much.

 

Ay, but their main brand has changed. In addition, their GPU lineup is about to change as well, with adding the APU's. If there were any time to change numbering, now would be it when they change their main brand name.

The 6000 series are not ATI cards. They are AMD cards and obviously get their own numbering. It might confuse some, but in time they will figure it out.

(Going from ATI to AMD should ring a bell for some though that more then the main brand has changed).

 

Yep, it will be one of those moments: If you want better DX10 and older DXs performance- get an 5800, If you want better DX11 performance- then get an 6800 series. But I deeply doubt that 6800 series performance will be enough for more than a slide show in a proper DX11 application.

 

I am totally confused now.
How would these 960 SP's (Barts XT) or 1440 SP's (Cayman) be organized, if they are really 4D?

2 (3) RPE's a 120 4D SP's ?

SIMDs with 16 4D SP's would not be possible then.
Except, that we will see again 5D SP's
Then it could be organized like that.

2 (3) RPE's x 6 SIMD's x 16 5D SP's

Edit: In this case, I cannot see how these 960 Barts SP's can be as powerful as 1440 Cypress SP's... or 1440 Cayman SP's compared to 1600 Cypress SP's

 

Probably the numbers are wrong. Otherwise, where would this performance increase come from (I mean if the architecture is the same)? And the initial rumour was for 1536 SPs in Cayman, not 1440 SPs.

 

Or 1920 for Cayman which would correspond to 480 4D SP's, when this one slide with 320 4D SP's for Barts is true...

 

I think that AMD should just give Juniper a new codename. How would we even be able to tell? NV should have done the same with G92. Just call the thing "revised and optimized" and all is forgiven.

What would be a nice island name for it??

 

Hmm yes indeed. Wasn't aware drivers made such a difference with vantage, and based conclusions on earlier score from HD5870/HD5850 (which put the 6870 closer to 5870 rather than 5850). So I guess this also means there won't be much of perf/W improvement. Still, if the chip isn't too big it should be ok - basically would be a direct GF104 competitor but with smaller die (that is the 6850 could compete with GTX460, the 6870 with the yet-to-be-released-haven't-heard-anything-lately full GF104 part).

 

Define normal

 

Turks?

(j/k…)

 

If Barts has really only 960 SPs (Juniper + 20%) then the other related rumor, that Cayman is a Cypress + 20% (=1920SPs) could to be correct too.

Therefore, if that is true then AMD has not changed the strategy and therefore Barts = HD68x0 is imho the wrong move.

 

Those numbers are weird. 6850 seems on par with 5850, yet 6870 falls way short of 5870 (i.e. reaches ~ GTX 470 performance).

In other words: Performance difference between Barts Pro and Barts XT seems way too small to make any sense : About 15% difference in such a semi-synthetic benchmark will probably translate to ~10% difference in real-world gaming performance - it's either a fake or the first Barts Pro OC cards will practically obliterate the market for Barts XT

Seriously, ANY of the possible specs leaked so far indicated that Barts XT will be up to 40% more powerful than Barts Pro on paper ...

 

Who ever is in charge of the pre launch marketing campaigne is either doing a hell of a job in not over hyping the new arch.....OR ...hes doing a terrible job in promoting it LOL

 

Why?.. GTX470 performance at a prospective 1/3rd of the die size? what is there NOT to like?

 

Maybe all of these leaks were wrong?

Also, on paper 5770 has 36% more raw processing power than 5750. Real performance difference is 20% at best.

Last but not least, look at last generation, the highest default OC for 5850 was 765 MHz. Yes, if you manually overclock you can reach close to 5870 performance levels, but that didn't stop people from buying 5870s, despite a ~40% higher price.

 

Barts is 1/3 of GF100, i.e. about 175mm²

 

550mm² /3 = 183mm²

Maybe neliz used the "other" die area estimate of the GF100 for his comparison?