I don't understand why you are imposing a binary "either or" logic tree here.
Engineering involves making tradeoffs, R&D investments, timeline for deliverables, risk management, etc.
When it comes to uncore stuff, one thing that Intel certainly has more experience with is their power-gating and turbo-clocking circuits. We can't really expect AMD to go from time-zero to being on-par with Intel all in one step when it comes to circuit layout and areal management of these features.
I also don't understand why you are so insistent on the Orochi die released at ISSCC being obfuscated...when this happened last autumn AMD was right quick with making sure everyone knew that it was. They have not said anything this time around.
Why release two differently obfuscated diemaps? Does not compute. If they wanted nothing but obfuscated diemaps to be out in the wild then they'd be releasing them left and right...or only releasing the same one but doing it over and over.
I believe the newest die shot is the real McCoy. Accepting that it is such but concluding it must be bad engineering is just needlessly silly. AMD does not have the resources to engineer products the likes of Intel, there is no need to cast negative judgement on their design engineers just because the company's bank account has less in it.
If they would need 20% more die space for that than I am with you but if they need 120 % (!) more for the same than there are only these two answers possible. And what you say is not correct. AMD was first with integrated memory controller and interconnect so if then Intel would have more problems since they did it first with Nehalem. And regarding design, the uncore is the simplest stuff on the CPU, means it consumes also the littlest space (see Intel). And if i look at the die shots from Shanghai and Deneb you can see how dense AMD can pack the design/uncore stuff. Why they should fail then with Zambezi? As I say either obfuscated or engineer failing, as their engineers already prooved that they can do it right.
They did the Module/Core so well so why should they fail so terribly with the much easier uncore stuff? And if I talk about uncore I mean not the core, not L2 nor L3 but the rest. How can this rest consume almost 50% of the die space? And even more why even waste any engineer resources on making this complex CMT if just doubling cores and invest the power in the uncore stuff would bring so much more benefit. It is just logic thinking that runs my alarm bells that there is something wrong.
And then in addition if you take into account that AMD SRAM cell size is 0.149 µm2 compared to Intel's 0.171 µm2 (both on their 32 nm process) it is even more suspicious. So smaller cores smaller SRAM and still ending with double(!) die size?
http://www.realworldtech.com/includes/images/articles/iedm08-16.png
And regarding the obfuscation. They made it once, they did obfuscate the Bulldozer die. It was so obvious that everyone detected it. So why not do it again making it a little bit more difficult to see? Or wait - no, they did not do it again, they just removed the obvious obfuscation and left the unobvious obfuscation as is. It is still the same picture. And the proportions are still the same of the previously obfuscated one.
The die size is one of the key informations that Intel wants to have as this is the factor determining production cost. So it is the most important commercial information.
And you have my answers for both ways. I do not make a binary tree of it. It is a binary tree of if this die shot is either true or false.
I'll come back when the real die size numbers are out if you just do not want to see this discrepancy (AMD is capable of making smaller cores but ends up with double the size of Intel!).
Update:
To get some additional light in that I investigated the Deneb die.
There you have for Cores/L2/L3 ~163 mm²
BD has for Cores/L2/L3 ~172 mm²
Uncore size for Deneb is 95 mm² in 45 nm
BD uncore should be similar but on 32 nm which would mean ~70 mm² (very conservative and taking into account that pads don't shrink).
Deneb already shows that AMD has a problem in the Uncore design because that 95 mm² is way too much (compared with Intel offering more functionality in less area).
Now this bad Uncore design would lead to ~240 mm² for Zambezi.
And the Deneb die also shows four HT links so very likly they will not be removed in Deneb.
If we assume that the die picture is right and we have this large 280 mm² Zambezi part that means that Uncore increased further to 108 mm².
With that comparison - let the missing 40 mm² left aside - it could be true that the ISSCC die shot is real.
This is a real surprise in my opinion. That would mean that AMD could not compete regarding performance when a little bit later the Core i7 Extreme on LGA2011 aka Sandy Bridge EN comes. Intel just doubles their core count to 8 and they have the performance crown back in 3 month.
So far I thought this would not be a problem since AMD can just issue a 8 Module single die part against this but with such a large Uncore area they cannot. AMD cannot compete because of their incredible large uncore area, it is unbelievable and some folks here talk about they do not have bad engineers. My strong advice to AMD: Get your Uncore fixed!
I now really ask for what all the effort with CMT?
If AMD would get their uncore right they could just double the core count when Intel does but this option is nullified by the uncore waste.
I am still hoping that the die picture is obfuscated but my investigation of Deneb which brougth a bad uncore die size performance is not very promising.
Let us hope that it is at least smaller than this 280 mm². With 240 mm² the situation would be better than AMD could issue at least a 6 module / 12 core part. But a 8 module / 16 core would be necessary. Yes I know they can do it by flipping two 4 modules together as they do with Interlagos. That is nice but we will not see these parts on consumer motherboards.
So we will likly see a mixed picture: In the main stream a Zambezi which will be superior to the Sandy Bridge 4 core and in the high end a Sandy Bridge EN which will be even better than Zambezi.
Normally you could say very fine because main stream market is what counts and AMD will be very good there. On the other hand you never before get such a large performance boost with Intel Extreme parts. They are not just give 10-15% percent advantage over main stream - this time they will give a whopping 100% minus application scaling.
Therefore I am looking forward to Zambezi and I am looking forward to Core i7 Extreme. AMD Zambezi could be 40% faster than SB2600 in May/June and SB EN could be 80% faster than SB2600 in October or so. And maybe at the year end we see a Zambezi 12C ... What a race in 2H/2011!
And again: AMD get your Uncore fixed!
