Samsung: We can get to 5 nm without "any major technical difficulties"

[dahorns]

Yeah, ARM/Qualcomm/Apple/TSMC/Samsung have been very successful in the mobile arena, true. But as I said, on desktop process tech has not brought as much improvement lately.

You must be ignoring the laptop market, which I would contend is properly part of the mobile arena. Huge gains have been made there. Also, the gains in desktop have been less about absolute performance, but more about decreasing the size and energy consumption of the desktop. What once took massive full towers, can now be placed in micro atx or mini itx form factors. To me, that is a pretty significant change.

 

[jdubs03]

You must be ignoring the laptop market, which I would contend is properly part of the mobile arena. Huge gains have been made there. Also, the gains in desktop have been less about absolute performance, but more about decreasing the size and energy consumption of the desktop. What once took massive full towers, can now be placed in micro atx or mini itx form factors. To me, that is a pretty significant change.

I would have to agree. As has been said before (I forgot who said it) Intel's recent strategy has been to focus on energy savings and small IPC increases rather than their former strategy (last decade) of focusing primarily on performance.

My i5-2430M laptop at the beginning got about 3.33 hours of runtime (now I have 85% of the original capacity, but seems like I'm down to like 2.5hrs of runtime). My friend just bought a Yoga 2 Pro with an i7-4510U, performs a bit better (at least in geekbench), yet the TDP is way lower and lasts about twice as long. And that is only a 2 generation difference.

Just from that example, at least in the laptop space there has been fundamental change in product performance.

Bay Trail was obviously a real nice jump above Saltwell and ushered them to a pretty defensible position, but it seems like Intel is content with Airmont not being much of an advance to all of ours dismay (well maybe aside from a few anti-Intel peeps in here). Willow Trail is the next jump, and it better be good because that will be their last shot at the mobile area in terms of high-end. If that fails, Core M will have to come down in price or companies may migrate away.

This may not be really on-topic but this article seems pretty interesting:
http://blogs.wsj.com/digits/2015/02/26/apples-focus-in-israel-chip-design/

Maybe Apple will move away from ARM and just start their own in-house IP for logic. Or maybe they're just looking for more designers to tweak ARM's A72 core as far as it can go per release. Either way if it wasn't apparent from before, Apple is serious about logic design.

 

[Lepton87]

Maybe Apple will move away from ARM and just start their own in-house IP for logic. Or maybe they're just looking for more designers to tweak ARM's A72 core as far as it can go per release. Either way if it wasn't apparent from before, Apple is serious about logic design.

They already use a custom ARM core so you mean a custom ISA? Why would they go through all the trouble of porting the software for no apparent gains? The royalties paid for using ARM
ISA must be extremely tiny for apple if they are paid per ARM core.

 

[Fjodor2001]

I would have to agree. As has been said before (I forgot who said it) Intel's recent strategy has been to focus on energy savings and small IPC increases rather than their former strategy (last decade) of focusing primarily on performance.

Here's an interesting question: Let's assume Intel instead had decided to continue perusing increasing performance as the primary goal. What gains do we think actually would have been possible?

Isn't the reason that Intel focuses mostly on lowering TDP on desktop also that process tech itself lately has been able to provide more gains in that area, and less in performance such as frequency increases. And that is regardless of whether of that Intel would have chosen to prioritize?

I don't think we would have seen 5-6 GHz base frequency CPUs from Intel at 22 nm and ~95 W TDP even if increasing performance would have been their primary goal.

 

[ShintaiDK]

The node is one part. The design is another. And the designs isnt made for 6Ghz quadcores at 130W. Remember the design strategy 2% performance may cost no more than 1% power increase.

 

[Fjodor2001]

The node is one part. The design is another. And the designs isnt made for 6Ghz quadcores at 130W. Remember the design strategy 2% performance may cost no more than 1% power increase.

Sure, but given that Intel would have focused primarily on performance, do you think we'd see substantial gains at 30-50% per generation, counting both IPC and frequency increase while at unchanged TDP? I think not. I think part of the reason Intel has focused on lower TDP on desktop as well (and not only mobile where lower power consumption matters more) is that it's been their only option, given the characteristics of later process tech nodes.

 

[ShintaiDK]

It is what endusers and OEMs want. The desktop is simply dead. And honestly, its a good thing.

I wouldnt expect some miracle increases even if they focused on the desktop and raw performance only. Legacy software and coding isnt going away anytime soon.

 

[Hulk]

You've still got to admire Samsung.


Once that behemoth machine corporation decides to do something - nothing is gonna stop them from doing it.

Not reality, or physics, or anything else!

 

[Enigmoid]

Sure, but given that Intel would have focused primarily on performance, do you think we'd see substantial gains at 30-50% per generation, counting both IPC and frequency increase while at unchanged TDP? I think not. I think part of the reason Intel has focused on lower TDP on desktop as well (and not only mobile where lower power consumption matters more) is that it's been their only option, given the characteristics of later process tech nodes.

Its a lot harder to increase performance at a given power level than it is to reduce power at a given performance level given the power curve.

Considering we've just barely seen 30-50% efficiency improvements per generation on mobile, this kind of performance gain would have been pretty much impossible. It gets exponentially harder to increase IPC as well.

 

[Fjodor2001]

Its a lot harder to increase performance at a given power level than it is to reduce power at a given performance level given the power curve.

Yes, lately that has been true. But check out the performance increase per watt increase during 1970-2005 or so, primarily driven by frequency increase but also IPC improvements:

http://www.intel.com/pressroom/kits/quickreffam.htm

Yes, the TDP increased too. But at the current stage TDP increase much more per performance increase compared to before. Sad but true...:'(

 

[jdubs03]

They already use a custom ARM core so you mean a custom ISA? Why would they go through all the trouble of porting the software for no apparent gains? The royalties paid for using ARM
ISA must be extremely tiny for apple if they are paid per ARM core.

I doubt that Apple would adopt their own ISA (I was just throwing the possibility out there), so it is likely that they'll continue to pursue the strategy of customizing ARM cores. But, they obviously have a further agenda when they're continuing to acquire logic design firms. You're right though, the amount that Apple has to pay to license the design is a pittance.

 

[Enigmoid]

Yes, lately that has been true. But check out the performance increase per watt increase during 1970-2005 or so, primarily driven by frequency increase but also IPC improvements:

http://www.intel.com/pressroom/kits/quickreffam.htm

Yes, the TDP increased too. But at the current stage TDP increase much more per performance increase compared to before. Sad but true...:'(

Its always been easier to decrease power at a given performance level than to increase performance at a given power level.

Thats the way the performance - power graph works.

Historically this was the case as well, though large increases in frequency could be made and there was a lot of low hanging fruit in increasing IPC. Not so much anymore.

 

[Idontcare]

No major technical difficulties for Samsung, he's just waiting for TSMC to figure those out

LOL!

 

[cbn]

That because desktop is a dying market.

Dying? Or a market that will be reinvigorated in the future when there are more competitors?

 

[cbn]

And FinFET should be able to get them to 7 nm without major issues

That is amazing considering Intel is expect to switch to a xtor type that is different from FinFET for 10nm.

 

[cbn]

Yeah, ARM/Qualcomm/Apple/TSMC/Samsung have been very successful in the mobile arena, true. But as I said, on desktop process tech has not brought as much improvement lately.

I think in order for us to see node development focusing on higher drive currents, the ARM world would have to develop a cpu core design that is able to not only deliver high frequency, but high IPC as well.

Right now Apple has the high IPC, but they don't have high IPC and high frequency together.

 

[Idontcare]

Dying? Or a market that will be reinvigorated in the future when there are more competitors?

The form factor is the problem.

If I can get my "good enough" computing in a mobile laptop format then why would I ask my boss, my parents, my wife, etc. for desktop compute form factor?

Yes there are people who can't get good enough computing in a mobile form factor (myself included), but we are a minority, not a majority. And that makes for a form factor that is set to experience low growth, if not contraction (i.e. "dying") for the foreseeable future.

 

[Idontcare]

That is amazing considering Intel is expect to switch to a xtor type that is different from FinFET for 10nm.

FinFET is the xtor geometry, III-V is the material. Geometry and material are decoupled and are not required to go hand-in-hand.

You can have III-V planar xtors, you can have Si/SiO2 FinFet transistors. Or you can have III-V Finfet xtors. Etc.

I would be surprised if Intel abandons the FinFET geometry as early as the 7nm node, but I wouldn't be surprised if they adopt III-V materials from which they build the xtor.

 

[cbn]

It is what endusers and OEMs want. The desktop is simply dead. And honestly, its a good thing.

We also have to consider the current price to performance ratio on Intel desktop has stayed basically the same (on almost all levels) for the last four years. The major exception would be iGPU (which I question the relative value of).

 

[cbn]

The form factor is the problem.

If I can get my "good enough" computing in a mobile laptop format then why would I ask my boss, my parents, my wife, etc. for desktop compute form factor?

Yes there are people who can't get good enough computing in a mobile form factor (myself included), but we are a minority, not a majority. And that makes for a form factor that is set to experience low growth, if not contraction (i.e. "dying") for the foreseeable future.

I bet there are lots of young people in lots of countries (especially emerging markets) who would love better than "good enough" performance.

But how cost effectively can ARM can get there is the main question?

Development of CPU core design focusing on both high IPC and high frequency vs. the process node cost (xtor design, materials science, lithography) to support that higher frequency?

 

[ehume]

Yes, lately that has been true. But check out the performance increase per watt increase during 1970-2005 or so, primarily driven by frequency increase but also IPC improvements:

http://www.intel.com/pressroom/kits/quickreffam.htm

Yes, the TDP increased too. But at the current stage TDP increase much more per performance increase compared to before. Sad but true...:'(

We're at the top of the S-curve now. The stuff you're quoting comes from the stem.

 

[mrmt]

Dying? Or a market that will be reinvigorated in the future when there are more competitors?

Most corporate applications today run comfortably on notebook computers, and in two generations we will be able to say the same on tablet computers. If you can have your computer in your backpack or in your drawer, why bother with a cumbersome desktop?

Desktops will be mostly restricted to gamers and workstation guys. Everyone else will go towards small form factors a la NUC or mobile form factors like tablets, notebooks or convertibles.

 

[Fjodor2001]

Most corporate applications today run comfortably on notebook computers, and in two generations we will be able to say the same on tablet computers. If you can have your computer in your backpack or in your drawer, why bother with a cumbersome desktop?

Desktops will be mostly restricted to gamers and workstation guys. Everyone else will go towards small form factors a la NUC or mobile form factors like tablets, notebooks or convertibles.

The problem is not performance, but display size and ergonomics of input devices. You cannot beat a 24" monitor with mouse and proper keyboard if you actually want to get something done. Tablets are useless in comparison, but good for casually consuming media.

 

[mrmt]

The problem is not performance, but display size and ergonomics of input devices. You cannot beat a 24" monitor with mouse and proper keyboard if you actually want to get something done. Tablets are useless in comparison, but good for casually consuming media.

Mobile computers like notebooks and computers have keyboards and mouse, and you can put a bluetooth keyboard on your tablet. You can also send the video signal from your tablet to a big screen, and with notebooks or convertibles you can just plug the thing on the big monitor.

 

[MisterMac]

Mobile computers like notebooks and computers have keyboards and mouse, and you can put a bluetooth keyboard on your tablet. You can also send the video signal from your tablet to a big screen, and with notebooks or convertibles you can just plug the thing on the big monitor.

IBM also has sub-10nm nanotube transistors ready.
Just not for mass logic production.

Market does not have infrastructure in place for " seemlees streaming of screen signals and input\output control".

And that ain't happening for a long while - despite the fact it's been technicly feasible for what 10 years?