Ok let's make this clear as you don't seem to be listening: ARM outright said not to expect any changes that in any way significantly change the microarchitecture as they are not in the scope of the license.
I guess this all goes back to what someone would consider significant. For instance I would personally consider the following a fairly significant change to the architecture:
But if you don't consider changes like that significant, then I guess I can understand why you are make the claims that are making (well the new claims at least, since you at least seem to have abandoned your original claim that Kryo 280 was A73/A53 straight up).
Again I'm not saying that Kryo 280 is as different from A73, as say the old Kryo cores or even the A72 or A17 cores.
They said this, they make the microarchitecture. I was there in the room in their offices asking them about it on TechDay. Also on a separate occasion a major licensee disclosed that the changes don't exceed 5% from the original designs and that it's all about the marketing power.
Of course ARM is the one implementing the changes, and thus the ones making the microarchitecture, but again that doesn't necessarily preclude them from making something that differs a fair deal from the A73 (all though as mentioned above, what "a fair deal" encludes would then come down to what one considers significant).
Do you have a link to the 5% link, or was that something told to you directly? And did it refer to their own specific design or was it a more general claim about the limits of what was possible if the license was pushed to it's limit?
@antihelten: I am not sure what answer you are looking for. We have rather clear statement from ARM itself, yet you dismiss it as "rather superficial examples," and I do not know what Andrei (or anyone else for that matter) can do to satisfy your demand.
As far as I know the only example we have is the instruction window change I linked above, and I would consider that example to be fairly superficial in the amount of details offered, but by all means if you have some links to a more in depth list of the various changes that are possible with the built on ARM license then feel free to link it.
There are other circumstantial evidences: that Qualcomm is rather coy about the details of the S835's design; that Geekbench subtest scores show remarkable similarity with A73's (not related to throttling affecting the scores); that ARM's "new" licensing scheme debuted coincidentally when Qualcomm announced S835; that Qualcomm is taking advantage of that license instead of shrinking S820 which was semi-successful; that what few details Qualcomm disclosed about S835 are identical to what we know about A72/A72, such as L1/L2 cache arrangement.
Again variation of up to 10-20% is not what I would consider "remarkable similarity". Also no one is disputing that Kryo is based on the new licensing scheme, so there isn't really anything coincidental here. Also we're talking about A73/A53 here, not A72/A72, so not sure why that would be relevant?
http://www.anandtech.com/show/10948/qualcomm-snapdragon-835-kryo-280-adreno-540/2
There will be time enough more details and die shots of S835 will come out. But it is unlikely Qualcomm will admit that S835 is utilizing stock A72/A73 for obvious reasons that are not tech-related. Consumers do not have a fond memory of Qualcomm's last attempt at utilizing ARM cores (S810), and there will be renewed questions on Qualcomm's competence among tech press and the investors in relation to Qualcomm dropping Kryo design.
Of course Qualcomm will not admit to using stock A73 cores, since that would almost certainly be a lie. Even Andrei admitted that
Kryo 280 is most likely modified in some manner (however small and insignificant), even though his
original claim was just A73/A53.
Besides which, depending on how you define "custom," Qualcomm might even have a enough wiggle room to justify calling Kryo 280 a custom design. In that case we are not going to have a consensus because some might consider Qualcomm's implementation "superficial " and some might consider it a genius chip design that no one else did.
To tell the truth I am not sure why all this is important. What matters is S835's performance, its performance/watt, and what ticks it achieve that performance and efficiency. Whether it is called "all-new custom design," or "the best implementation of ARM's big cores to date" seems a rather trivial concern.
There's really only one way to define custom, and that is "not stock", obviously a core can be more or less customized and Kryo 280 could certainly be only superficially different from A73, but that would still be a custom version however marginally.
I'm not saying that Kryo 280 a genius chip design, nor is anyone else as far as I can tell, nor is anyone claiming that it is an "all-new custom design", so those are really just strawmen.
The issue is that Andrei originally claimed that Kryo 280 was just A73/A53, he then later admitted that there was most likely some changes but claimed that they were insignificant, using a singular geekbench score to justify this claim, even though the subscores from this test could differ by up to 10-20% from the A73 in Kirin 960. He may certainly still be correct (as I said geekbench is not the most reliable thing in the world), but I just don't really think the evidence provided is particularly strong at this point.