Cayman is a new architecture. Barts is a tweaked Cypress.

[HurleyBird]

This post from [H]ardforums sums it up pretty well.
http://hardforum.com/showpost.php?p=1036325771&postcount=75


HardwareCanucks:
November will see the release of the Cayman XT-based HD 6970 and Cayman Pro-based HD 6950 which have all of the features seen in Barts plus enhanced rendering scalability and off-chip buffering for DX11 applications. These will be the spiritual successors to the HD 5870 and HD 5850 and should go head to head with the higher end Fermi cards.
December will see the introduction of Antilles which is meant to be the lynchpin of AMD’s renewed assault on the DX11 market. The HD 6990 will bring untold performance to the table through the use of a pair of Cayman GPU cores and additional features we can’t divulge at this time.

[H]ardOCP:
The 6900 series will have a superset of features compared o the 6800 series. This means that there will be features and architecture differences between 6900 and 6800 series. This allows AMD to take more chances on the high-end enthusiast class GPUs and architecture things different, to really step up performance that enthusiasts demand. So, just to restate, the new 6800 series will offer performance of the 5800 series, at a lower price, with lower power, and a smaller chip.

PC Perspective:
Later in the year we will see the release of future architectures that much more unique in the Cayman and Antilles product lines. We'll have to leave you with that tease for now and touch again on both of those items later.

Hexus:
These arrive armed with improvements in the two metrics discussed above, soon to be followed by a genuine performance GPU in the form of the Radeon HD 6950 and HD 6970 'Cayman' parts and, a little while later, the dual-GPU Radeon HD 6990, code-named Antilles. Phew!


And of course, this makes sense as AMD has stated that the refresh includes a "new architecture", and somehow I can't see how that could possibly be referring to what we have in Barts.

 

[badb0y]

Basically AMD had no answer at ~200$ price point that's why they released these cards first but now since these cards are out, They can smoothly transition to bring out the 69xx GPUs.

 

[AnandThenMan]

What exactly does off-chip buffering for DX11 applications mean?

 

[bunnyfubbles]

And of course, this makes sense as AMD has stated that the refresh includes a "new architecture", and somehow I can't see how that could possibly be referring to what we have in Barts.

Considering how close the 6800s perform to their 5800 counterparts, suggesting that Barts is not "new architecture" but that Cayman is gives me a lot more confidence that the 6900s will perform well.

The 6800s effectively nullify the 5850's existence when considering how close the 6850 comes in performance and how the 6870 consistently outperforms it while consuming no more power or producing no more heat/noise, and both cards are cheaper than the 5850...

 

[thilanliyan]

I can't wait for Cayman. If Cayman smashes everything else I'll get one of those....otherwise I'll just pick up 2 460s.

 

[DominionSeraph]

I can't wait for Cayman. If Cayman smashes everything else I'll get one of those....otherwise I'll just pick up 2 460s.

Don't you have a crossfire board?
And the 6850 looks to be slightly ahead of the GTX 460 in price/performance as prices stand now. ~$185 incl. shipping no rebate compared to the slower 1GB GTX 460 ~$180 with rebate.

6850 Crossfire beats SLI GTX 460 in power usage too. By ~50 watts in Crysis and ~90 watts if you want to consider Furmark.

 

[Dark Shroud]

I can't wait for Cayman. If Cayman smashes everything else I'll get one of those....otherwise I'll just pick up 2 460s.

AMD boards can't run SLI so you'll have to do a new set up.

But yes I'm waiting for Cayman as well unless the 6990 isn't priced insanely like the 5890 shot up to.

 

[thilanliyan]

AMD boards can't run SLI so you'll have to do a new set up.

But yes I'm waiting for Cayman as well unless the 6990 isn't priced insanely like the 5890 shot up to.

Don't you have a crossfire board?
And the 6850 looks to be slightly ahead of the GTX 460 in price/performance as prices stand now. ~$185 incl. shipping no rebate compared to the slower 1GB GTX 460 ~$180 with rebate.

6850 Crossfire beats SLI GTX 460 in power usage too. By ~50 watts in Crysis and ~90 watts if you want to consider Furmark.

Hehehe, they can run SLI with a hack.
And 460s OC like crazy, matching GTX470s. So I can essentially get 470sli for ~$360...not too bad.

 

[klansek]

what i really do not understand is how is it possible that these new boards, based on what is basically the same shader arch, perform so good with less TMU, same ROPs and so many shaders cut of (compared do 58xx)??? In my book AMD is doing miracles in terms of efficiency.

based on this, cayman boards should really be monster performers.

 

[betasub]

But yes I'm waiting for Cayman as well unless the 6990 isn't priced insanely like the 5890 shot up to.

5890? Is that another release I've missed? Sounds expensive.

 

[Dark Shroud]

5890? Is that another release I've missed? Sounds expensive.

That should have read 5970. I'm tired and reading the reviews instead of getting sleep.

 

[Grooveriding]

AMD boards can't run SLI so you'll have to do a new set up.

But yes I'm waiting for Cayman as well unless the 6990 isn't priced insanely like the 5890 shot up to.

Yes, I haven't looked at all the numbers and all the reviews yet. What I have seen shows crossfire on the 6870 scaling better than SLI does currently.

In BFBC2 I saw 93% scaling!

 

[Cookie Monster]

what i really do not understand is how is it possible that these new boards, based on what is basically the same shader arch, perform so good with less TMU, same ROPs and so many shaders cut of (compared do 58xx)??? In my book AMD is doing miracles in terms of efficiency.

based on this, cayman boards should really be monster performers.

It could be because the balance between different parts of the architecture wasn't as optimal in its first incantation. So AMD must have figured out the bottlenecks and its opposite to reconfigure their architecture to provide a more elegant solution.

Another interesting thing is that performance bottleneck with cypress must have lied with something other than the ALU/TMU side of the architecture. Possibly in the ROPs but this remains to be seen.

 

[n0x1ous]

what i really do not understand is how is it possible that these new boards, based on what is basically the same shader arch, perform so good with less TMU, same ROPs and so many shaders cut of (compared do 58xx)??? In my book AMD is doing miracles in terms of efficiency.

based on this, cayman boards should really be monster performers.

the clockspeed difference probably accounts for a good chunk of it.

 

[Idontcare]

Another interesting thing is that performance bottleneck with cypress must have lied with something other than the ALU/TMU side of the architecture. Possibly in the ROPs but this remains to be seen.

Yeah we always knew something was up with the Evergreen->Cypress scaling (4870 crossfire vs 5870 at same clocks for example) so it stands to reason AMD endeavored to recover that performance they left on the table somehow.

the clockspeed difference probably accounts for a good chunk of it.

That seems to be the bulk of it, and it looks like they selected clockspeeds so that they stayed within the thermal footprint of Cypress as best as possible even if it meant coming in just a skosh lower in performance.

(my perception may be wrong, but I get the impression the 6870 is like 5% slower than a 5870 while being substantially lower in price, is that a fair assessment?)

 

[GaiaHunter]

the clockspeed difference probably accounts for a good chunk of it.

Still, the 6870 should have been around 94% of the speed of a stock 5850 and it ends around 10% faster, especially with AA and at higher resolutions.

Even the 6850 seems to be around the same speed or faster if AA is applied.

http://www.computerbase.de/artikel/...ate/22/#abschnitt_performancerating_ohne_aaaf

That seems to be the bulk of it, and it looks like they selected clockspeeds so that they stayed within the thermal footprint of Cypress as best as possible even if it meant coming in just a skosh lower in performance.

I have the impression that these cards will allow some factory OC parts if the AIB use different boards and/or tweak voltages, especially the 6850.

 

[Lonyo]

the clockspeed difference probably accounts for a good chunk of it.

The HD6870 has a slight disadvantage compared to the HD5850 (clockspeed * shaders) but manages to outperform it across the board, so basically the efficiency improvements are about equal to the performance difference between the HD5850 and HD6870, so ~10% at best clock for clock.
But it's also more efficient in terms of performance relative to die size.

 

[n0x1ous]

Still, the 6870 should have been around 94% of the speed of a stock 5850 and it ends around 10% faster, especially with AA and at higher resolutions.

Even the 6850 seems to be around the same speed or faster if AA is applied.

http://www.computerbase.de/artikel/...ate/22/#abschnitt_performancerating_ohne_aaaf



I have the impression that these cards will allow some factory OC parts if the AIB use different boards and/or tweak voltages, especially the 6850.

oh absolutely, there's definitely some improved efficiency as well, but I'm just saying clock speed is a big part of how 68xx is able to do so well vs 58xx with the smaller die. If 6870 was clocked at 750 or 775 (whatever the 5850 is, I'm not sure) we wouldn't be as impressed by this launch.

But I think AMD is providing a real nice value here and did a nice job with the performance of these cards. I expect Nv would launch a 460 + or something with an official higher clock speed then 675. A 460 at 850mhz or so for maybe $229 would be much more competitive (as the reviews have shown that included the EVGA 460 FTW) with the 6870 then the 470 is.

The 470 is performance competitive with the 6870, but obviously not on other factors like power, heat, noise. A new reference 460 @ 850mhz or so would compete better overall.

In fact, for the whole Fermi range as a whole, I think their much lower clock speeds vs AMD parts are a big factor in why they didn't outperform HD 5xxx as much as maybe they could have. Certainly they were limited by the die size, heat, power etc. So I guess what I'm saying is that if they can get power, heat under control on 28nm, they should be able to do much better with the clocks as AMD has had higher clocked parts across the board for quite a few years now.

 

[GaiaHunter]

oh absolutely, there's definitely some improved efficiency as well, but I'm just saying clock speed is a big part of how 68xx is able to do so well vs 58xx with the smaller die. If 6870 was clocked at 750 or 775 (whatever the 5850 is, I'm not sure) we wouldn't be as impressed by this launch.

It is indeed.

But look at 6850. The 5850 is 725 MHz with 1440 sps while the 6850 is 775 MHz and 960 sps. And don't forget the 72 texture units of the 5850 vs 56 of the 6870.

By shader speed alone the 6850 should be 70% of the speed of 5850 and it is much closer than that.

In fact, for the whole Fermi range as a whole, I think their much lower clock speeds vs AMD parts are a big factor in why they didn't outperform HD 5xxx as much as maybe they could have. Certainly they were limited by the die size, heat, power etc. So I guess what I'm saying is that if they can get power, heat under control on 28nm, they should be able to do much better with the clocks as AMD has had higher clocked parts across the board for quite a few years now.

Don't forget that AMD and NVIDIA have a different approach: AMD clocks all the card at the same speed, while NVIDIA has a hot clock for the shaders that is twice the rest of the card clock - that means while the GTX 460 base clock is 675 MHz its cuda cores are working at 1350 MHz.

The base clock it is indeed a factor but I think it is simplistic to reduce the AMD and NVIDIA speed differences to clock speed.

 

[Bateluer]

I'm a little glad that I won't have the extra cash for a video card upgrade until the end of December anyway. Keeps my options open, though it is frustrating.

 

[bunnyfubbles]

I have the impression that these cards will allow some factory OC parts if the AIB use different boards and/or tweak voltages, especially the 6850.

Agreed. This should be a similar case of the 5770 vs. 5750. Really the 5750 performs so very close to the 5770 but the small difference of cost makes it hard to not go ahead and pony up the difference for the 5770.

But the $60 difference between the 6800s is a decent chunk of change (and the $180 price point of the 6850 is also a mental hurdle advantage of beind under the magic $200 price point) and the nearly identical memory system (same 1GB capacity and negligibly different bandwidth) doesn't provide any reservations like it does with the 768MB vs. 1GB GTX470s.

A design with a better cooler that allows for overvolting (a likely candidate would be an MSI 6850 with its Twin Frozr cooler) could make the 6850 pretty tough to compete with when considering the ease with which the 5800/5700s hit 900+MHz and assuming that Barts will follow suit.

 

[s44]

Who was suggesting we'd see 4+1 go to 2+2? That was totally wrong...

 

[thilanliyan]

^That might happen for Cayman.

 

[cusideabelincoln]

In fact, for the whole Fermi range as a whole, I think their much lower clock speeds vs AMD parts are a big factor in why they didn't outperform HD 5xxx as much as maybe they could have. Certainly they were limited by the die size, heat, power etc. So I guess what I'm saying is that if they can get power, heat under control on 28nm, they should be able to do much better with the clocks as AMD has had higher clocked parts across the board for quite a few years now.

You can't compare the clockspeeds of Radeon cards to Geforce cards. One, they are a different architecture and the clockspeeds are designed to fit them accordingly. Second, nVidia has clocks its core and shaders differently while AMD keeps them the same - and this is a subset of having two different architectures.

 

[formulav8]

Who was suggesting we'd see 4+1 go to 2+2? That was totally wrong...

There were many rumor sites saying things like that. I don't recall anyone at AMD saying it though.

But the November cards could be based on a configuration like that.



Jason