AMD: R9xx Speculation

No cooling on the memory either.

Its certainly a weird design, I don't expect stock cards to look like that.

 

Have you forgotten how much disgrunt that caused?

 

What are the odds that this is in reality a Cayman pro? i mean are they going to have Cayman and Barts with 256bit...? Also i read the chiphell thread, and the OP did not say anything about what model it was...

I am still hoping what charlie said about this generation being pin compatible with the 5000 series holds up...

 

Naming isn't necessarily always all that reasonable with respect to performance-levels. Marketing plays a mayor role here.

I still agree with the basic premise you formulated: An HD6870 generally losing to an HD5870 would backfire in most reviews - even if there was a bunch of differentiating new "features" to behold and DX11/tessalation perfomance was doubled.

But you could also look at it the other way round: I don't remember any HDx7xx card that ever had to add that second 6pin power adapter. An AMD high-midrange card that consumes 150W+? Doesn't bode well with a lot of non-enthusiast customers keeping a very close eye on their power bills.

At this point, I'd be surprised if Bart's performance wasn't somewhere in the range of todays 58xx family - the question really is what kind of name (and price-tag) AMD will slap on that kind of card.

 

Is't HD3870(much slower than HD2900) was first?

 

No, they are roughly equal in performance, not as you say "much slower".

 

No way, how chip with half of bandwidth could be same?

 

When the chip with twice the bandwidth didn't need that much for most workloads.

 

WTF?


http://www.computerbase.de/artikel/...670/28/#abschnitt_performancerating_qualitaet

 

I could find tons of reverse examples, but, you know, just do not want continue flame

 

Back then NVidia was really tough competitor, 2900XT was a kind of (epic) fail and in order to stay competetive, ATi designed a chip which was much much cheaper to manufacture with almost the same performance. Perhaps you missed that.

 

Just to pour some more fuel in those naming-scheme related flames:

What if AMD decided to put different chips into one (naming-wise) family?

E.g.

[HD6750 = high-clocked/renamed Juniper chips - can't think of any other way to fill the gap here]
HD6770 = lower-clocked/partly disabled Barts chips (Pro) ~ HD5850
HD6850 = selected/high-clocked Barts chips (XT) ~ HD5870
HD6870 = lower-clocked/partly disabled Cayman chips (Pro) ~ HD5870+30%
HD6950 = selected/high-clocked Cayman chips (XT) ~ HD5970
HD6970 = 2x Barts XT ~ HD5970+20%
HD6990 = 2x Cayman XT ~ HD5970+50%

It would be confusing from a purely technical point of view - but it might make for a better marketing wise naming scheme when compared to current HD5xxx performance levels ...

If Barts XT really performs ~ Cypress clock-for-clock, AMD's marketing guys surely lost some sleep pondering over how to position such a chip on the market:

- Artifically "cripple" it to fit it into a possible HD67xx series - and you'll lose a lot of profit due to not pushing your silicon as much as you could.

- Raise clocks to put it somehwere in the range of the old HD58xx series - and (a) power draw will probably give you a hard time marketing those cards under the HD67xx name or (b) performance won't be quite good enough to rectify labelling your Barts XT card as the "successor" to HD5870 ...

So maybe just try something out-of-the-ordinary to make it all fit?

 

This is what Nordic hardware had to say on it...
http://www.nordichardware.com/news/...ts-launches-in-october-as-radeon-hd-6800.html

Theres a chart at the bottom that helps situate where the new cards will fit in. Not sure if this accurate though.

 

This is not flame... I don't remember any single review showing HD3870 "much slower than HD2900"

Hardware.fr:

 

That is not entirely a bad idea. Makes a lot more sense than what is currently floating on right now. Some hell will break loose if other wise.

 

According to me lol this doesn't make any sense and is completely against everything sane. Yep, I call it insane. Some chips from Evergreen, other in inappropriate naming scheme that doesn't belong to them. I hope Dave is here and is laughing at this. Otherwise it would be really sad when turn into reality...

 

Maybe they decided to re-introduce the sideport-idea he advocated so strongly for a long time - and he had to acquiesce in a really crazy naming scheme in return

Nah, I'm just trying to make some sense of the rumours floating around recently - but I'm still pretty sure we'll ALL be surprised once NI is officially unveiled...

 

None. They aren't going to have only one crossfire connection on their highend series.

Also, isn't Barts using a 5850 PCB? Same length, just a completely different layout?

 

It wasnt Dave, it was Carrell Killebrew. Afaik Dave is in marketing(?)

As to the power draw, GTX 460 has two 6 pin PCIE connectors and it is priced smack in the midrange market at $179. I dont think a TDP of 150-160W is something that is going to worry most midrange gamers.

Also remember those are just engineering samples, the final board may require just one. HD 5850 has a TDP of 151W. A Barts chip at the rumoured RV770 die size should have a TDP closer to HD 4850 than HD 4870. Also the two 6 pin connectors may be there so that overclocking is not limited by available power

 

That right there is proof positive that the alleged Barts picture is NOT going to be a 68xx card.

As well rumors are that 68xx will be similar or slightly longer than 58xx cards. Yet another piece of info that would greatly contradict that Barts picture as being in the 68xx segment.

Barts will be 67xx and Cayman will be 68xx.

Regards,
SB