Intel: Chips To Become Slower But More Energy Efficient

[HKZeroFive]

Quote

Intel has said that new technologies in chip manufacturing will favour better energy consumption over faster execution times – effectively calling an end to ‘Moore’s Law’, which successfully predicted the doubling of density in integrated circuits, and therefore speed, every two years.

 

The prognosis comes from William Holt, Intel’s Executive Vice President and General Manager of its Technology and Manufacturing Group, speaking at the International Solid State Circuits Conference in San Francisco, and discussing the new technologies – such as tunnelling transistors (or ‘Quantum tunnelling’) and spintronics – which will define the next stages of evolution in computing.

 

“We’re going to see major transitions,” said Holt. “The new technology will be fundamentally different.” and continued “The best pure technology improvements we can make will bring improvements in power consumption but will reduce speed.”

Quote

Holt elaborated that while Intel recognises the need to consider re-tooling its plants and committing to new technologies in chip production, it hasn’t made a decision about direction yet. Quantum tunnelling, though brought to advanced proof-of-concept by DARPA and the Semiconductor Research Corporation, is currently further from commercialisation than spintronics, which uses quantum mechanical properties of particles as switch facilitators, and which is expected to begin to appear in commercial technology such as graphic chips within 18 months.

 

Holt has stated not just that Moore’s Law is coming to an end in practical terms, in that chip speeds can be expected to stall, but is actually likely to roll back in terms of performance, at least in the early years of semi-quantum-based chip production, with power consumption taking priority over what has been the fundamental impetus behind the development of computers in the last fifty years.

Not surprising, you can't really just jump into new technology expecting to have identical or faster performance. However, I hope that Intel will still continue high-end performance options and not take those away (it's most likely they won't). Personally, I'm not much of an energy efficiency guy and I suspect alot of you share the same view when I simply say I just want faster chips. Makes alot of sense to aim for energy efficiency though if you're a large corporation with server farms that require massive amounts of it.

 

So what do you think? Is Intel going in the right direction?

 

Sources:

 

https://thestack.com/iot/2016/02/05/intel-william-holt-moores-law-slower-energy-efficient-chips/

 

https://www.technologyreview.com/s/600716/intel-chips-will-have-to-sacrifice-speed-gains-for-energy-savings/

[Altecice]

I dont have an issue with this at all, chips are fast enough at the moment... if everybody within 5 years can save 5-10w of power * billions of processors, thats ALOT of energy saved that we dont have to produce and pollute.

[Minsekt]

3 minutes ago, Altecice said:

I dont have an issue with this at all, chips are fast enough at the moment... if everybody within 5 years can save 5-10w of power * billions of processors, thats ALOT of energy saved that we dont have to produce and pollute.

you beat me to it 

 

 

but i totally agree. the chips potencial isnt used fully yet anyways

[KemoKa]

This is AMD's fault. They'd better tan intel's hide with Zen or we're all totally screwed.

[Deli]

When you consider a little chip can product more heat than a soldering iron.......and a high-end gaming system can draw more power than a microwave oven.

[Starelementpoke]

15 minutes ago, KemoKa said:

This is AMD's fault. They'd better tan intel's hide with Zen or we're all totally screwed.

All they have to do is offer at least haswell ipc for cheaper and they´re golden.

 

[That Norwegian Guy]

7 minutes ago, Starelementpoke said:

All they have to do is offer at least haswell ipc for cheaper and they´re golden.

 

Ivy Bridge IPC is good enough as long as there are 6 core options.

[daniellearmouth]

10 minutes ago, KemoKa said:

This is AMD's fault. They'd better tan intel's hide with Zen or we're all totally screwed.

No, it's the fault of the laws of physics.

 

Getting a 14nm process working effectively enough for a new microarchitecture had already proven a challenge for Intel, keen to hold onto the ever-dwindling accuracy of Moore's law.

That's kind of why Skylake and Broadwell are in relatively low stock compared to Haswell. Silicon can only do so much. We would have to start using derivatives like silicon-germanium, which IBM used to create a 7nm chip, as a proof of concept.

 

As a matter of fact, the 5nm process is suggested to be the end for Moore's Law. Any smaller and it would be practically impossible to create it.

 

As for chip performance itself, we're pretty comfortable right now. We've got 18-core chips (with a supposed 22-core one on the way) and commercially viable hexa-core and octo-core chips from Intel that perform brilliantly as it is. Right now, we're not in a massive rush to see a new paradigm in CPU performance until we can kick Moore's law out for good, because it's starting to be Intel's downfall if anything.

[AresKrieger]

Fortunately for me I don't need more performance so eh, either way I'm not that surprised since they are running into thermal issues and lower max clocks due to the denser transistor designs, I do wonder what will happen once they reach the wall on the horizon, despite this I still doubt amd will manage against intel, but who know I can't predict the future. 

 

That being said I still want more performance, but won't hold my breath on amd since I doubt the possibility of getting much more performance than intel can given theoretical limits.

[Coaxialgamer]

Let's just hope they keep making high power chips ; otherwise , i'm going full AMD

[BuckGup]

But the worlds energy usage has actually gone done and flat lined. We used to always build more coal plants and nuclear now we haven't had to build anymore in years. Thanks to energy efficient things like LEDs in houses

[AresKrieger]

Just now, Coaxialgamer said:

Let's just hope they keep making high power chips ; otherwise , i'm going full AMD

AMD will likely not be able to get better performance than intel despite this just a fyi, there are limits to how much performance you can get out of silicon and we are close to that limit. But who knows maybe amd will get into on die cooling to compensate.

[daniellearmouth]

Just now, Coaxialgamer said:

Let's just hope they keep making high power chips ; otherwise , i'm going full AMD

Yeah, uh, good luck with that.

 

AMD aren't exactly going to break the laws of physics just to please you, friendo.

[kuddlesworth9419]

This is OK with me surprisingly. What we really need is people making software that actually leverage's all my 6 cores and 12 threads. At the moment no video game or program actually does this properly. Graphics cards aren't though they really need to start bringing stuff out that is way more powerful then what we have. Even the 980Ti isn't powerful enough.

[Coaxialgamer]

9 minutes ago, daniellearmouth said:

No, it's the fault of the laws of physics.

 

Getting a 14nm process working effectively enough for a new microarchitecture had already proven a challenge for Intel, keen to hold onto the ever-dwindling accuracy of Moore's law.

That's kind of why Skylake and Broadwell are in relatively low stock compared to Haswell. Silicon can only do so much. We would have to start using derivatives like silicon-germanium, which IBM used to create a 7nm chip, as a proof of concept.

 

As a matter of fact, the 5nm process is suggested to be the end for Moore's Law. Any smaller and it would be practically impossible to create it.

 

As for chip performance itself, we're pretty comfortable right now. We've got 18-core chips (with a supposed 22-core one on the way) and commercially viable hexa-core and octo-core chips from Intel that perform brilliantly as it is. Right now, we're not in a massive rush to see a new paradigm in CPU performance until we can kick Moore's law out for good, because it's starting to be Intel's downfall if anything.

they could just make bigger dies while staying on 14 nm a bit longer , instead of making smaller and smaller dies by going 10nm or less.

 

Sure , cooling would become an issue sooner or later , and they would have to move on eventually .

[Deli]

1 hour ago, daniellearmouth said:

......

 

As for chip performance itself, we're pretty comfortable right now. We've got 18-core chips (with a supposed 22-core one on the way) and commercially viable hexa-core and octo-core chips from Intel that perform brilliantly as it is. Right now, we're not in a massive rush to see a new paradigm in CPU performance until we can kick Moore's law out for good, because it's starting to be Intel's downfall if anything.

Mark my words. When Polaris and Pascal come out later this year. There will be tons of cry-babies posting,"My 6700K/6960X is bottlenecking my 1080Ti SLI. What can I do?"

[Poofu]

Kind of a clickbait title when there talking about a whole new type of transistor let alone most likely not in silicon. Quantum tunnelling and spintronics are completely new fields and will not catch up to modern chim manufacturing process until they have matured enough.

[daniellearmouth]

Just now, Coaxialgamer said:

they could just make bigger dies while staying on 14 nm a bit longer , instead of making smaller and smaller dies by going 10nm or less.

 

Sure , cooling would become an issue sooner or later , and they would have to move on eventually .

They could, but you'll eventually end up with a chip bigger than even the Xeon E5 2699 v3.

 

And besides, like I mentioned earlier, Skylake and Broadwell are on the 14nm process. How many people run a Broadwell chip? Or a Skylake chip?

By comparison, how many people run a Haswell or Ivy Bridge system? I can guarantee you it's a lot more, and it's more than just because they're older.

 

Producing chips on the 22nm process is much less of a problem than 14nm, since heat is more easily managed, the CPUs are still powerful enough, and unless you don't mind having a processor the size of a small plate, we need to shrink to smaller and smaller transistors at some point.

3 minutes ago, Deli said:

Mark my words. When Polaris and Pascal come out later this year. There will be tons of cry-babies posting,"My 6700K/6960X is bottlenecking my 1080Ti SLI. What can I do?"

If they are using the PCIe 3.0 bus, don't bet on it.

Yes, they seem like huge paradigm shifts in GPU performance, but don't bet on them being so powerful that your rig literally would not be able to handle them.

[daniellearmouth]

Just now, valdyrgramr said:

Rage Fury Maxx

That is such a 90s name for something.

[Jahramika]

1 hour ago, Altecice said:

I dont have an issue with this at all, chips are fast enough at the moment... if everybody within 5 years can save 5-10w of power * billions of processors, thats ALOT of energy saved that we dont have to produce and pollute.

Why are they fast enough in your perspective? Because Intel has no competition at the moment? If Zen pops out bad ass close all this power savings can go away to win?

[AresKrieger]

15 minutes ago, Jahramika said:

Why are they fast enough in your perspective? Because Intel has no competition at the moment? If Zen pops out bad ass close all this power savings can go away to win?

More so programs don't even utilize current technology to it's full extent would be my guess, also I highly doubt zen will win on performance regardless, AMD is likely going to focus on price competition not the high end market which will be good for pricing competition but will not affect intel's performance plans.

[patrickjp93]

1 hour ago, KemoKa said:

This is AMD's fault. They'd better tan intel's hide with Zen or we're all totally screwed.

Our current processors aren't even remotely being pushed to their limits in consumer and prosumer workloads. A 4790K at full tilt on all 4 cores is capable of (4 cores * 256bits wide/32 bits of precision * 2 operations per cycle * 4.2*10^9 cycles per second = ) 268.8 Gigaflops or Gigainops for integer work. Do you operate under the delusion that most software we run actually pushes this chip? Most bottlenecks we see in software today come from bad coding and forced legacy support over allowing newer instructions to do the same work faster. We are far from screwed. The innovation will continue, and performance will increase even if clock speeds don't. Kaby Lake and Cannonlake will also double consumer compute power once again with AVX 512. It's up to software makers to program well enough to take advantage of SIMD instructions like AVX, period.

[patrickjp93]

1 hour ago, Poofu said:

Kind of a clickbait title when there talking about a whole new type of transistor let alone most likely not in silicon. Quantum tunnelling and spintronics are completely new fields and will not catch up to modern chim manufacturing process until they have matured enough.

It's hardly clickbait. This is going to be employed at the 7nm node. We are within 5 years of seeing CPUs made with at least one of these two technologies if not both. Let's not forget Intel was first to FinFET in commercial products and has a penchant for bringing the bleeding edge designs out first. They'll be using both 3D FinFET and FDSOI at 10nm, and IBM will be using the same at 14nm for the Power 9 chips which should enter production by the end of the year.

[patrickjp93]

1 hour ago, AresKrieger said:

AMD will likely not be able to get better performance than intel despite this just a fyi, there are limits to how much performance you can get out of silicon and we are close to that limit. But who knows maybe amd will get into on die cooling to compensate.

Performance comes from architecture designs, not silicon. We're about to double our compute performance per core with AVX 512. If Intel wanted it could make 1024-bit vector units too in the same silicon. That said, software has to leverage those new instructions for the improvement to be seen. That's the big reason everyone seems to think Intel's performance has stagnated in consumerville. The problem is software, plain and simple.

[asim1999]

I think this is good for both consumers and enthusiasts:

For consumers: less heat output and longer battery life on laptops as well as quieter operation

Enthusiasts: they can overclock higher due to lower TDP so lower heat and power Also, they could use less cooling or have better OCs with same cooling

So its a win-win in the long run