Intel CPUs in 7-10 years

[CakeMonster]

I'll just quote one of IDC's posts from another thread because I'm curious about this:

Only Intel's management and decision makers had the wisdom and fortitude year's ago to see to it that they were smartly investing in the R&D that they would need some 7-10 yrs later (now).

Has anyone followed what has been stated publicly about what an Intel CPU will look like in 7-10 years? I assume Intel have planned as they always have, but if we discount the experimental stuff, do we know anything about what can reasonably expect in this time frame?

 

[witeken]

I expect CNTs (5nm), and some other cutting edge technologies:

Not sure about the design side. I'd guess a continuation of the FLOPS/clock improvements and other things they increase every Tock like OoO window.

 

[cytg111]

CNT is potentially revolutionary I understand .. Is there any qualified guesstimates on the speed advantages? Are we going to see 10GHz+ chips in the foreseeable future?

 

[ShintaiDK]

CNT is potentially revolutionary I understand .. Is there any qualified guesstimates on the speed advantages? Are we going to see 10GHz+ chips in the foreseeable future?

CNT doesnt fix the issue why we cant have 10Ghz chips. We need async clocks or something around in the cores for that to happen. Or something else in the design to fix it. Else you get metastability.

Example of Ivytown:

 

[witeken]

CNT is potentially revolutionary I understand .. Is there any qualified guesstimates on the speed advantages? Are we going to see 10GHz+ chips in the foreseeable future?

I think you can find an answer somewhere here.

 

[cytg111]

Thx, good reading.

Apparantly it is not only when you approach zero kelvin that the universe starts to act weird, operating on the nano scale is apparantly a giftbag on its own ie "Remote joule heating"
http://phys.org/news/2012-04-carbon-nanotubes-weird-world-remote.html
it is like "hidden variables" all over again..

 

[frozentundra123456]

Still ramping up 14nm.

Edit: OK, I will be serious. The difficulties in 14nm production really show how difficult it is to work at this level, so I think it is pretty much impossible to know what will be on the market in 7 to 10 years. I am hopeful that there will be some big gains with alternative substrates to silicon, but otherwise I see gains in performance becoming increasingly difficult and expensive, and eventually even for intel the cost of a new node may be more than the benefit.

 

[Homeles]

CNT doesnt fix the issue why we cant have 10Ghz chips. We need async clocks or something around in the cores for that to happen. Or something else in the design to fix it. Else you get metastability.

Intel has PLLs on 14nm capable of 7.5GHz. 10GHz is well within range. 7-10 years puts us at 5nm or so.

 

[Triskain]

A PLL does not make a processor. I've seen PLL Designs that reached the 40GHz range on a bog-standard 65nm TSMC node. The Clock speed of today's MPUs is not even close to being limited by the underlying process technology.

 

[NTMBK]

I'd guess a continuation of the FLOPS/clock improvements

Actually I'd expect that one to taper off after Skylake. They've hit parity with the Xeon Phi now in SIMD width- they'll probably only increase at the same rate as the Phi.

 

[biostud]

Some kind of nanotube heat dispenser.

 

[Fjodor2001]

We will have approximately 1.08^7 = 1.71 => 71% faster Intel CPUs in the early 2020s. Power consumption on the low performance SKUs will be a lot lower though. That is if going by the track record for that last few years.

 

[Nothingness]

Actually I'd expect that one to taper off after Skylake. They've hit parity with the Xeon Phi now in SIMD width- they'll probably only increase at the same rate as the Phi.

Agreed. SIMD widening has limited impact on performance of typical user code, and does not even benefit all HPC code bases.

 

[you2]

I expect a strong focus in mobile arena. Currently high end snap-dragon are 1/2 to 2/3 of a mobile I-3. In 5 years I would expect baytrail to produce processors faster than the current mobile i5 while maintaining the general power draw required of a mobile processor. This comments ignoret he fact that snapdragon/tegra GPU are killing intel so we will also see quite a bit of improvement in this areana.

 

[Fjodor2001]

I think the more interesting question is: What will Intel and the rest of the semiconductor industry do after 7-10 years? There is no plan after 5 nm, so chances are Moore's law is dead, and nobody has any idea where to go after that. Unless something disruptive happens...

 

[Bubbleawsome]

I think the more interesting question is: What will Intel and the rest of the semiconductor industry do after 7-10 years? There is no plan after 5 nm, so chances are Moore's law is dead, and nobody has any idea where to go after that. Unless something disruptive happens...

At that point I expect it would be a race to find a new material and create a new Ghz war. maybe optical after that. I don't know anything about optical though.

 

[Burpo]

 

[Homeles]

A PLL does not make a processor. I've seen PLL Designs that reached the 40GHz range on a bog-standard 65nm TSMC node. The Clock speed of today's MPUs is not even close to being limited by the underlying process technology.

Yeah, I'm not particularly sure why I said that.

 

[TechFan1]

Haha, the first thing that popped into my head when I saw that baby picture was chip implants that interface with your brain. The true post PC era.

 

[dealcorn]

Anything 7-10 years out is experimental with unproven economics for manufacture. The bottom 2/3 of Arnold Frisch's Seeking Alpha article on Intel's next disruptive technologies suggests that silicon photonics may be disruptive sooner than expected (http://seekingalpha.com/article/2309465-what-are-intels-next-disruptive-technologies?uprof=46). I believe Xeon Phi will be the introductory vehicle during either 2015 or 2016 but it is fundamental to Moore's observation that technology rolls down-market.

 

[Idontcare]

Anything 7-10 years out is experimental with unproven economics for manufacture.

Unfortunately this is true in part because the economics are in large part determined by the specific IC being manufactured and the market's (largely unknown) demand for that IC 7-10yrs down the road.

One could argue that GF's 32nm process tech was largely an economic failure, in part due to the 'dozer-based ICs that AMD designed and tried to sell with it. No need to look a decade forward to see examples of failed technology and product parings in today's world.

 

[dmens]

Intel has PLLs on 14nm capable of 7.5GHz. 10GHz is well within range. 7-10 years puts us at 5nm or so.

We had ALU's running at 7ghz with Prescott on 90nm. Netburst forever!

 

[witeken]

I think the more interesting question is: What will Intel and the rest of the semiconductor industry do after 7-10 years? There is no plan after 5 nm, so chances are Moore's law is dead, and nobody has any idea where to go after that. Unless something disruptive happens...

I think there are plenty of ideas. Just take a look at sites like ExtremeTech and IEEE Spectrum. They don't even must have something to do with the actual transistors. I don't think it will end at the 5nm node.

 

[NTMBK]

My theory- Larrabee finally comes to fruition, and Intel replaces integrated graphics with x86 cores plus texture units. A consumer SoC has a handful (about 2) high frequency, out of order, very wide CPU cores for single threaded performance, and dozens of in-order simple cores handling multithreaded workloads like graphics. Of course this will meed a fairly major rethink of OS scheduling techniques.

 

[Fjodor2001]

My theory- Larrabee finally comes to fruition, and Intel replaces integrated graphics with x86 cores plus texture units. A consumer SoC has a handful (about 2) high frequency, out of order, very wide CPU cores for single threaded performance, and dozens of in-order simple cores handling multithreaded workloads like graphics. Of course this will meed a fairly major rethink of OS scheduling techniques.

I.e. Intel will follow AMD's strategy, and focus on "compute cores", whether CPU of GPU cores.