Replies to post #129707 on Intel Corp (INTC)

[chipguy]

Why did Bay Trail come in at 102mm^2 even with:

* A better/denser process (we all agree Intel's 22nm process should be denser than anything on TSMC 28nm, all things equal)

Which 22 nm Intel process is that?

What is the respective yield of the Bay Trail and Snap Dragon?

You do know that design for high yield and good manufacturing
margins results in lower density chips?

Perhaps you can leave chip design to engineers and comment on
things you actually know something about.

[wbmw]

Why did Bay Trail come in at 102mm^2 even with:

* A better/denser process (we all agree Intel's 22nm process should be denser than anything on TSMC 28nm, all things equal)
* No connectivity/comms
* Inferior GPU performance to its contemporaries
* No integrated sensor hub

Note that Snapdragon 800, with similar CPU performance, better GPU performance, integrated connectivity/comms, integrated sensor hub, etc. all worked out to 118mm^2.

Your incredibly insightful analysis has determined that Qualcomm ended up with better graphics performance and integrated comms, and all it took them was 16mm2 MORE DIE SIZE. Wow <yawn>.

You really don't have a metric or bar to set in terms of what Intel *should have* achieved with this part. Nevermind that Qualcomm started out several years ahead of Intel in filling their technology pipeline with mobile focused SOCs, and at worst, Bay Trail is maybe half a generation behind in capabilities. This looks to me like Intel is moving faster than Qualcomm.

- H1'2012: Medfield vs. Snapdragon S4 (2+ generations behind)
- H1'2013: Clovertrail+ vs. Snapdragon 600 (1.5 generations behind)
- H2'2013: Bay Trail vs. Snapdragon 800 (0.5 generations behind)

Not bad, for fewer than 2 years worth of product introductions. Hardly what I'd call incompetent - and in fact, I wouldn't expect a design team on this Earth to do much better, or cover so much ground more quickly, unless the case was against a competitor standing still. While Intel made some great strides in the past, their competitor AMD simultaneously made multiple missteps, but Qualcomm has continued to execute superbly. There's only so much even the world's best design team can do, given the state of the technology pipeline vs. a competitor with years more of a head start.

Maybe the problem here is that you've set your expectations impossibly high, potentially due to a naive understanding of what's possible to achieve at an engineering level. It's not like people haven't been trying to point that out to you for the last couple dozen posts, which you have largely ignored in support of your pre-supposed thesis.

Also, even with a process lead, the Silvermont cores clocked to what? 2.4GHz? Krait 400 is roughly on-par at 2.26GHz in terms of performance and it seems that Snapdragon 805 will pack 2.5GHz variants, built again on 28nm HKMG.

Comparing clock frequency is completely meaningless. Which benchmark shows you that Silvermont is behind Snapdragon 800 and 805, and at which power? Geekbench 3.0 supports your "roughly on par" statement, but other benchmarks show Intel reasonably ahead.

Here are some from Anandtech's reviews of Bay Trail and Galaxy Note 3:

http://www.anandtech.com/show/7376/samsung-galaxy-note-3-review/4
http://www.anandtech.com/show/7314/intel-baytrail-preview-intel-atom-z3770-tested/3

Mozilla Kraken (lower = better):
- Galaxy Note 3: 7,674 ms
- Bay Trail FFRD: 4,686 ms (64% better)

Google Octane v1 (higher = better):
- Galaxy Note 3: 3,867 ms
- Bay Trail FFRD: 6,219 ms (61% better)

AndEBench - Native (higher = better):
- Galaxy Note 3: 16,802 ms
- Bay Trail FFRD: 21,235 ms (26% better)

Sunspider 1.0 (lower = better):
- Galaxy Note 3: 590 ms
- Bay Trail FFRD: 567 ms (4% better)

Of course, a few benchmarks are inverted, but I think Intel's job with optimizing things like Java runtime are far from finished, and there's still room for massive optimization.

As for expectations for a frequency boost from Snapdragon 805, don't assume that Intel doesn't also have a frequency bin improvement planned for Bay Trail as well. I expect more benchmark comparisons this year, with Intel even more prepared on the software side with Android to battle it out with Qualcomm.

The question is whether Silvermont architecture overall can keep up. I think the data so far says it can.