Will Intel's focus on future CPU's power efficiency, limit overclocking?

[VirtualLarry]

See this post by Virge:
http://forums.anandtech.com/showpost.php?p=32993694&postcount=37

Speaking about the Pentium M:

The design team actually went in and slowed down paths that were running above Banias' target clock frequencies, because if a path is able to run faster than it should, it means that you're wasting power. The benefit of this is an even more power efficient microprocessor, but the downside is a microprocessor that has a clear clock frequency wall.

Will Intel do the same with IB, or perhaps Haswell? Slow down the critical paths intentionally, so that they won't run above a certain Mhz wall, up to a limit on certain SKUs? All in the name of greater power efficiency? (Which Windows 8 is focusing on too.)

Is this not only going to be the end of distributed computing, but also overclocking?
http://forums.anandtech.com/showthread.php?t=2225768

 

[Kristijonas]

We can look at this as a bad thing, or we can look at it like this:
we already have enough CPU performance, so we can lower power draw and make processors smaller and colder until we can fit a sandy bridge +80% performance CPU into a tablet. Problem?

 

[GammaLaser]

I don't really understand how this is a bad thing. The paths that were slowed were not critical paths but paths that ran faster for no benefit, thus power was wasted. To me this sounds like the design was optimized and balanced so no one path was causing a frequency bottleneck.

 

[Lonbjerg]

Ivy Brigde lowers TDP.
Haswell raises it again to the same levels we have today.

 

[grkM3]

Its higher tdp is from a 30-40% increase of gpu power.intel likes its 95 watt tdp so it will fill that gap with a better gpu.

 

[sm625]

We know that chips like atom and brazos cant really overclock. But we also know that these chips would still be dogs even if they ran at 5.4GHz, so it is hard to say if it matters.

 

[Idontcare]

I don't really understand how this is a bad thing. The paths that were slowed were not critical paths but paths that ran faster for no benefit, thus power was wasted. To me this sounds like the design was optimized and balanced so no one path was causing a frequency bottleneck.

I already posted in the video thread, so I won't copy-paste it here, but you are right, of course. I'm willing to bet you are in the industry? Circuit designer or layout engineer maybe?

For the OP, if Intel wanted to limit overclocking they need simply reduce the range of available clockspeed multipliers.

They could have made the top Sandy Bridge multiplier be 40x for example and then that would have limited it.

As to whether or not their prioritization of lower power consumption will come at the expense of reducing maximum clockspeeds, yes. Absolutely. But not because of the leakage aspects that are underpinning your initial concerns.

The limits come in from the engineered-in reliability of the IC itself and how structure and electrical parameters factor into the device's intrinsic reliability (Gate oxide integrity, TDDB, electromigration, etc).

When you design a process node to be "low power" you don't need to make big fat wires for current density purposes (electromigration), and you don't need to implement robust insulating dielectric materials for signal isolation purposes (speaks to clockspeed-limited cross-talk and leakage).

But if you take such a process node and juice it with voltage to boost the drive-current in a bid to accelerate the switching speeds (i.e. overvolt so you can clock the chip higher) then the associated increases in the electric field placed on the dielectric materials as well as the associated increase in current through the wires and channel of the transistor is going to degrade everything all the more rapidly.

Shortening the operating life of the chip, and making it die sooner.

There is a reason GPU's and CPU's like Atom/Nano/Brazos don't clock to 5GHz. Its one part microarchitecture and one part process node. Each can limit the other, there is always a rate-limiting aspect to the dependence.

 

[Ben90]

Is this not only going to be the end of distributed computing, but also overclocking?

So you can't run distributed computing projects if your computer isn't overclocked a crapload?
My stock laptop didn't have a problem folding, don't see what the problem is.

In all seriousness, I would imagine not being able to overclock really high would increase demand for DC participants.

 

[TuxDave]

Here's my two cents. Slowing down paths that run above frequency to save power has been done ever since I joined Intel. On the other hand, you don't have to be afraid because when slowing down paths, we don't bring it right to the timing wall. There is a safety margin because the more paths you have at the timing wall (the frequency spec), statistics will kick in and cause less chips to make the highest bin. We have enough tough timing paths that will most likely limit the frequency of the chip, we're not in the business of crushing our yield.

 

[Borealis7]

seems more logical to me that Intel would limit overclocking potential due to focus on PROFITS

 

[Lonbjerg]

I think the landscape might change post "haswell".
As Intel are pursuing their ambition of a SoC in PC's...and my gut feeling is that the CPU after "haswell" wil be a fullblown SoC.
Intel never really like third party chip in their designs... SoC would later all that...and close the options for Mobomakers to implement their own overclocking "hacks"...

 

[Arachnotronic]

Says they're improving overclocking in HSW...
http://img.donanimhaber.com//images/haber/31727/intelhaswelldx111_dh_fx57.jpg

 

[Idontcare]

nvm

 

[piasabird]

Well if you have turbo boost technology for Quad core processors, then you need some overhead to speed up because they typically advertise to top end of the turbo boost. Some things that I have seen Intel doing looks like a late reaction to some of their criticism. They went back to the drawing board and put the Intel HD video 3000 on some of their processors, that they should have done when it came out.

Then you look at processors like the 2500k and wonder why there is nothing faster than that at this time? The 2500K has been out for over a year. With no competition, Intel looks like it is sitting on its hands. They are probably working on the next socket and processor, or the next big business oriented chipset and processor. The business side is where they make a lot of their money.

 

[Lonbjerg]

Well if you have turbo boost technology for Quad core processors, then you need some overhead to speed up because they typically advertise to top end of the turbo boost. Some things that I have seen Intel doing looks like a late reaction to some of their criticism. They went back to the drawing board and put the Intel HD video 3000 on some of their processors, that they should have done when it came out.

Then you look at processors like the 2500k and wonder why there is nothing faster than that at this time? The 2500K has been out for over a year. With no competition, Intel looks like it is sitting on its hands. They are probably working on the next socket and processor, or the next big business oriented chipset and processor. The business side is where they make a lot of their money.

Google " Intel tick tock".
It might surprise you.