Replies to post #137578 on Intel Corp (INTC)

[wbmw]

Really? Now you are putting the same argument about why Bay Trail sucks against A7 chips but replaced with Core.

Bay Trail and A7 are pretty comparable designs. The former wins in MT, and is slightly behind in ST. Graphics performance of Bay Trail is slightly behind A7 in GfxBench, yet slightly ahead in GFXBench. Yet I've noticed that many of your posts are skewed in a way that leans towards "Intel Sux", when the data says otherwise.

A8X is a very nice design, and advances Apple's position. But it's also a more aggressive design, at ~3B transistors. The 20nm process makes it manufacturable as a mobile SOC - but it doesn't take much to extrapolate a die size of 120-130 mm2, which is on the high side for mobile SOC's. Hard to imagine pushing the envelope much further. A 4B or 6B design without a process node shrink (which is still a couple of years out, with 16FF+), wouldn't even fit in existing form factors.

Intel could build a larger Atom product, but it's unclear what business prospects they have for chasing Apple in 10" designs. The market seems to have plenty more volume either below that in minis and phablets, or above that in small form factor PC's. Broadwell is a better chip for small form factor PC's, and it should do well. Intel undershot a bit with Silvermont core, but Cherry Trail will build on the momentum of Bay Trail and probably pick up more volume in the mid-range.

I believe Denver performs very well too. And its on 28nm. Don't need no "apple-specific optimizations" or "tri-gate 14nm".

You're one to give Intel hell for thermal characteristics, but nVidia gets a Free Pass? You probably missed the abundant heat problems on Tegra K1-32 on the graphics side, which I'm sure K1-64 inherited. Maybe a dual core CPU with high ST performance is a better trade off for mobile, but Bay Trail holds its own in MT. Denver runs at 2.5GHz, by the way, and a 7-way superscalar pipeline. Where do they go from here? Wider, or faster? Solve the technical problem with more complicated software, or by pushing the thermal envelope...? Cherry Trail should offer similar MT and Gfx performance, but at lower power.

So what you are saying is that Core M is a niche and makes no sense in most markets anyway.

I don't know what the size of the market is, but my point is that Intel has a pretty nice hedge. If the market for small form factor PC's is small or zero, then Intel is protected and completely out of reach in a 15-95W bastion of PC performance. If, on the other hand, the market is trending towards fanless designs in PC form factors, then Core M will protect Intel from ARM encroachment.

Want thin and light? Performance sucks. Want performance? U chips are better. Want affordability? ARM chips and Atoms are for you.

Oh, brother... you say "performance sux", even though it trashes anything else. Why aren't the OEM's fauning over Denver, instead of build innovating 2-in-1 machines based on Core M? Are you going to change your argument of "Core is too expensive" to "Intel is giving them away"?

The other two statements are true. U-series runs at higher power, and delivers better performance. Atom is designed to be lower cost than Core, and competes in price with ARM. It's a great business strategy - what's your problem with it?

4.5W Core M performs same as 6W Core i7 4610Y.

If you're going to run a benchmark like SysMark, then you have to expect the CPU to be running in Turbo mode most of the time. SysMark tests burst scenarios, and includes simulated wait states, as if an end-user were interacting with the PC. Even Haswell-Y running SDP power can achieve Turbo mode operation for burst scenarios. It's a best-case scenario for Haswell.

additional architectural level improvements with Skylake,

>> How much do you expect? 10%? 20%?

I'd expect something on the order of 10-20% IPC improvement, but a disproportionate amount of performance per watt improvement at low power. I think Intel is still moving the needle to the left in terms of power optimization point. I'd like to see Skylake at ~5W exceed Haswell at 15W - not just delivering 2.x GHz speeds in burst scenarios, but also sustainably in much more active scenarios. Ideally, it would look something like this (for top bin):

http://browser.primatelabs.com/geekbench3/compare/1083343?baseline=1080011

[wbmw]

Either (a) 14nm is worse than Intel is claiming (b) Charlie is right about Intel crippling Broadwell to get it out in time

Krzanich said that 14nm is still not where they want it to be in terms of yield. That could affect bin split as well.

But let's think it through. Intel has a cascading product development team, and 14nm has what appears to have been a 6 month impact to schedule, and as much as a 9-month impact to process health and ramp. Plan A is out the window, just like a train getting derailed, with each boxcar hitting the one in front. They could either:

- Plan B: All-hands on deck to fix Broadwell, and let the dominos fall where they may, or...

- Plan C: Cut-loose and double-down on Skylake

It's possible that Charlie's "detective work" has a grain of truth to it. If Intel adopted Plan-C, then they would have worked to get Broadwell out the door by sacrificing productization of either features, or performance. Or both.

And we know that Krzanich has said to expect Skylake in the second half of 2015, even as Broadwell pretty much ramps starting at the tail end of this year, going into early 2015.

Unless he's being overly optimistic about Skylake, it looks to me like he might have gone with Plan-C.