Intel Skylake / Kaby Lake

Page 644 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Dayman1225

Golden Member
Aug 14, 2017
1,152
973
146
they're interesting because they can hit really high frequencies and deliver some really good performance with reasonable power draw.

And such a thing is why Intel's 10nm process is somewhat exciting for arm SoCs, no?

IIRC >3ghz 250uw(microwatts) per mhz so <1w, is what they claimed
 
Mar 10, 2006
11,715
2,012
126
And such a thing is why Intel's 10nm process is somewhat exciting for arm SoCs, no?

IIRC >3ghz 250uw(microwatts) per mhz so <1w, is what they claimed

Yeah. I mean, we don't really have a competitive comparison with, say, TSMC 7nm or upcoming Samsung 7nm, but being able to do high frequency at low power for a reasonably complex design is really interesting, too.

Intel's processes have generally had clear performance leadership at higher leakage levels/power levels, but in the ultra-low power smartphone space, it's far less clear how Intel fares and unless you are one of the fabless customers who have done test chips on Intel processes and foundry processes, it's just tough to know how that comparison works out.
 
  • Like
Reactions: Phynaz

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
Intel has been renowned for their clocklike Tick-Tock manufacturing prowess, delivering a new process node with 2x density every 2 years, for many nodes, reliably, leading the industry in performance, density and yields.

Since 14nm, Intel has suffered a lot of delays and simply has failed to deliver on its manufacturing capability. Brian Krzanich' promise of relentless pursuit of Moore's Law has not come true.

TSMC and Samsung, on the other hand after the 20nm hiccup, have delivered on 16nm and 10nm, with no signs yet of any 7nm problems for TSMC.

Intel seems to have been overly aggressive with the choice of SAQP with DUV. The MMP of 36nm forces use of SAQP with DUV which leads to increased mask count, complexity and hence higher probability of creating defects due to more processing steps. TSMC and GF have chosen 40nm as MMP for a reason at 7nm. That allows them to avoid SAQP and lower mask count / complexity. Samsung has chosen to go only EUV at 7LPP with a MMP of 36nm and CPP of 54nm which is identical to Intel 10nm specs. The decision to go EUV only has pushed out their 7nm by over a year behind TSMC .

https://www.semiwiki.com/forum/content/4502-changing-foundry-landscape-trends-challenges.html
https://www.semiwiki.com/forum/cont...alfoundries-discloses-7nm-process-detail.html

Intel's problem is one of arrogance. They believe that if they are having crippling, show-stopping problems, its competitors have no hope of delivering on what they promise.

I think Intel might have expected EUV to be ready earlier and taken an over aggressive approach on their 10nm process specs. With EUV being delayed they seemed to have run into complexity / yield issues using SAQP with DUV. But from the magnitude of the delays I think the problems are larger than just yield. It also looks like Intel missed performance targets which were quite moderate to start with (25% perf increase vs 14nm). Intel and Samsung seemed to have gambled hard on EUV and we are seeing the former struggle and the latter delayed.
 
Last edited:

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
Intel seems to have been overly aggressive with the choice of SAQP with DUV. The MMP of 36nm forces use of SAQP with DUV which leads to increased mask count, complexity and hence higher probability of creating defects due to more processing steps. TSMC and GF have chosen 40nm as MMP for a reason at 7nm. That allows them to avoid SAQP and lower mask count / complexity. Samsung has chosen to go only EUV at 7LPP with a MMP of 36nm and CPP of 54nm which is identical to Intel 10nm specs. The decision to go EUV only has pushed out their 7nm by over a year behind TSMC .

Intel was probably betting on EUV being just in time for mid 10nm insertion, allowing them to attack crucial ( at decision time, what about now? ) mobile markets with superior products early with lower yield and then move on to EUV in those critical masks for their core business chips.
With hindsight it was horrible mistake, but one with business case attached I guess?
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
I think Intel might have expected EUV to be ready earlier and taken an over aggressive approach on their 10nm process specs. With EUV being delayed they seemed to have run into complexity / yield issues using SAQP with DUV. But from the magnitude of the delays I think the problems are larger than just yield. It also looks like Intel missed performance targets which were quite moderate to start with (25% perf increase vs 14nm). Intel and Samsung seemed to have gambled hard on EUV and we are seeing the former struggle and the latter delayed.

I wonder if it was Intel or ASML who was overly optimistic.
 

StinkyPinky

Diamond Member
Jul 6, 2002
6,761
777
126
7800X has way more PCI-E lanes, supports 128gb of quad channel ram, and has AVX-512, so it will have some advantages over 8700K.

7800x only has 28 lanes. Which is more than the 8700k will likely have but hardly way more.
 

Eddward

Member
Apr 10, 2012
56
19
81
Remember, I'm responding to this:


That is only 3 fabrication process shrinks (65 nm to 45 nm, 45 nm to 32 nm, and 32 nm to 22 nm) unless you ignore Haswell refresh. That is out of what, 17 process shrinks. And it is only true if you ignore other chips.

Yes, Tick-Tock has happened, but not that often in the history of processors. Intel was fine before tick-tock, will be fine after tick-tock, and AMD has done fine without tick-tock. It just isn't necessary. It wasn't an unpredictable confusing mess before and it won't be now.

Buy a motherboard compatible with your processor. End of story. That wasn't that messy.

Tick-Tock did serve to give the the impression that Intel was turning out new and improved products like, ahem, clockwork. This aided buyer's perceptions of them, but ultimately raised expectations that eventually couldn't be met. As said, they are fine without that artificial cadence.
Well, lets say in 2012 we had a pretty good idea what we can expect in 2015 - 14nm Skylake - a new architecture on an improved process. Yes, it could have failed, but it was working for quite some time and Intel was sending really strong messages to the audience on each conference how is tick tock doing well. And it worked fine until 2015, more or less, basically since 2006.
Now, they are starting to mix marketed generations / architectures / nodes / chipsets. Not well planed, just "glued together" what is possible and what is ready. Basically we know almost nothing what will happen in 12 months.
Situation with Kaby Lake-Coffee Lake is just the beggining and I can image that a lof of Kaby Lake users are not very happy now. Their expensive future proof CPUs are basically absolete in a few month after launch. (if purchased during summer then even worse).
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,682
136
Their expensive future proof CPUs are basically absolete in a few month after launch. (if purchased during summer then even worse).
They weren't future proof to begin with. A core count increase was clearly in the works and discussed in advance in the enthusiast forums. The successful product launch from the competition should have been the last sensible warning sign that the time of quads was at an end.

More importantly I don't see how Intel maintaining their clockwork cadence would have made this transition any easier: did you honestly expect Intel to let the market know in advance that 6c/12t CPUs are coming while also inviting customers to purchase their newly launched 4c/8t product? Had they kept their (presumably) original plan to launch CFL in Q1 2018 would the summer Kaby Lake purchase be any better?
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Situation with Kaby Lake-Coffee Lake is just the beggining and I can image that a lof of Kaby Lake users are not very happy now. Their expensive future proof CPUs are basically absolete in a few month after launch. (if purchased during summer then even worse).

Well I thought about it but the 6-core rumors were already known End of 2016 so it wasn't a complete surprise. And that Zen would be a 8-core was also known. So kaby-lake buyers knew the risks, at least enthusiast that are informed. The uninformed won't even know about CFl.
 

epsilon84

Golden Member
Aug 29, 2010
1,142
927
136
Their expensive future proof CPUs are basically absolete in a few month after launch. (if purchased during summer then even worse).
I don't get this logic. Did their Kaby Lake CPUs suddenly get slower and less 'future proof' because of upcoming CPU releases? No, they are still just as fast and as 'future proof' as they were originally.
 

majord

Senior member
Jul 26, 2015
433
523
136
mKBI6Ba.jpg
 

dullard

Elite Member
May 21, 2001
24,998
3,325
126
Well I thought about it but the 6-core rumors were already known End of 2016 so it wasn't a complete surprise. And that Zen would be a 8-core was also known. So kaby-lake buyers knew the risks, at least enthusiast that are informed. The uninformed won't even know about CFl.
I suck at internet searches, so someone could probably find earlier references. But here is one from July 2016 mentioning 6-core Coffee Lake:
http://wccftech.com/intel-14nm-coffee-lake-10nm-cannonlake-2018/

Coffee Lake was known since April 2016, if not earlier: http://www.oregonlive.com/silicon-forest/index.ssf/2016/04/intels_new_president_launches.html

Intel working on mainstream 6 core and 8 core products was known in Oct 2015: http://hexus.net/tech/news/cpu/86879-intel-go-beyond-quad-core-cannonlake-processor-family/

So, yes, these were known to all who really cared. And those who don't really care will get whatever they see at the local computer shop when they buy a their next computer and don't care about 4 cores vs 6 cores.
 

R81Z3N1

Member
Jul 15, 2017
77
24
81
Just get this if you really need a model M, it is even cheaper than some modern mechanical keyboards.

http://www.pckeyboard.com/page/product/UNI0P4A

What they need to do is charge more for the inclusion of the ps2 ports, put it on a gaming model, but have the base model use usb, or for higher tier models list it as an upgrade. Or even put a thunderbolt port.

I agree that the back panel is prime rel-estate and should be used for connectors and items that are more relevant than ancient ports. I would even be happy if they just allowed a motherboard header for it, let folks use that. Granted nowadays the front ports are becoming more important, folks are complaining that they have to use the back panel to plug anything in anymore.
 

Eddward

Member
Apr 10, 2012
56
19
81
They weren't future proof to begin with. A core count increase was clearly in the works and discussed in advance in the enthusiast forums. The successful product launch from the competition should have been the last sensible warning sign that the time of quads was at an end.

More importantly I don't see how Intel maintaining their clockwork cadence would have made this transition any easier: did you honestly expect Intel to let the market know in advance that 6c/12t CPUs are coming while also inviting customers to purchase their newly launched 4c/8t product? Had they kept their (presumably) original plan to launch CFL in Q1 2018 would the summer Kaby Lake purchase be any better?
Launching in Q1 or H1 '18 would do the job better I think, at least from psychological point of view. IMHO the best launch window for 6 core was in January, ~16 months after Skylake. But OK, it is as it is.

I don't get this logic. Did their Kaby Lake CPUs suddenly get slower and less 'future proof' because of upcoming CPU releases? No, they are still just as fast and as 'future proof' as they were originally.
Well, the performance is there, but not the value. Even 6 years old Sandy Bridge still has some of its value but in case of Kaby Lake this will be much much shorter story. Shortest in a decade I assume.
 

dullard

Elite Member
May 21, 2001
24,998
3,325
126
Launching in Q1 or H1 '18 would do the job better I think, at least from psychological point of view...but in case of Kaby Lake this will be much much shorter story. Shortest in a decade I assume.
Intel was on roughly a 13 month average cadence for quite a while. But Kaby Lake was late by a few months. While mobile Kaby Lake Launched Aug 2016, desktop was delayed until Jan 2017. So, Kaby Lake will have a slightly shorter time frame than others, but not much shorter. Had Kaby Lake desktop launched with mobile, it would be a very typical distance between generations. The Sandy Bridge 2700K had even a shorter gap, although admittedly the 3770K isn't much of an upgrade.
  • 8700K: Oct 2017?. 9 months later
  • 7700K: Jan 2017. 17 months later (came later than their normal summer/fall)
  • 6700K: Aug 2015. 14 months later
  • 4790K: June 2014. 12 months later
  • 4770K: June 2013. 14 months later.
  • 3770K: Apr 2012. 6 months later (came earlier than their normal summer/fall)
  • 2700K: Oct 2011. 17 months later
  • 875K: May 2010
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
http://wccftech.com/intel-core-i7-8700k-msi-z370-krait-benchmarks-overclocking-leaked/

Coffeelake as expected will be a overclocking beast easily crossing 5 Ghz. Looks like a good AIO should be able to hit 5.2- 5.3 Ghz. This is the precise reason that 7800x will be destroyed by a 8700k. >10% higher max clocks and better gaming IPC. AMD might have to cut prices to respond to Coffeelake until Pinnacle Ridge arrives in 2018. The next few years are going to be golden for the PC industry with competition unseen since the Athlon K7 / K8 days.
 
  • Like
Reactions: Lodix