Intel Skylake / Kaby Lake

[utahraptor]

So is 390 the actual coffee lake mobo with 370 a crappy stop gap? You have to wait months for the real thing?

 

[TheF34RChannel]

So is 390 the actual coffee lake mobo with 370 a crappy stop gap? You have to wait months for the real thing?

That's one way of taking things out of context and for a spin

So no, the mobo next year still only has the unimportant and unimpressive small additions over this year's and I'm genuinely surprised that this needs to be repeated every two pages so henceforth I'll refrain from dealing with repeat questions. No offense, just not a fan of repeating which is a me issue

 

[Ajay]

I fully suspect we will see second-wave X299 motherboards with substantial heatsinks on the power delivery to overcome this.

Absolutely agree, just don't know when the second wave are coming. Anyway, X299 seems like a good platform overall, but not the best overclocking i9 CPUs right now.

 

[StefanR5R]

(on the question of whether X299 boards which are specialized to Kaby Lake-X can be justified on Kaby Lake-X's overclocking merits)

The problem encountered by AT appears to be that the board's power delivery couldn't keep up with the current required for a 5GHz OC for benchmarks longer than 5 minutes or so.

...In this case it seems we are hitting thermal limits for the power delivery, as explained by Igor Wallossek over at Tom's Hardware...

@IEC, I think Ian Cutress' bad Kaby Lake-X 5.0 GHz overclock (which nevertheless made it into the headline of Ian's article ) requires further analysis. Ian reported voltages, but neither power draw nor CPU package temperature, VRM temperature, and PWM controller temperature.

Igor Wallossek stated that Skylake-X overclocks with up to 250 W power draw would always be limited by CPU package temperature, not by the board's power delivery.

Furthermore, Ian stated that he got a slightly unstable overclock at 5.1 GHz/ 1.35 V, and got it stable at 5.0 GHz/ 1.25V. Tom's Kaby Lake-X review cited some binning data from a motherboard manufacturer:
http://www.tomshardware.com/reviews/intel-core-i7-7740x-kaby-lake-x-cpu,5107-8.html
According to those data, Ian's specimen was between middle and low (assuming that Ian's criteria for a stable overclock are comparable to the motherboard manufacturer's).

Power consumption of Kaby Lake-X at 5.0 GHz / 1.25 V with all-core load but non-AVX code is probably not more than half of the 250 W which Igor found to be still CPU temperature limited, not VRM temperature limited.

Therefore I have doubts that thermal throttling of the power delivery was limiting Ian's OC. Perhaps the CPU throttled. Could his choice of AVX offset have played a role in Blender and in Handbrake? (Obviously not in Chrome compile though.)

Could another limit of the board's power delivery, short-term peak current, have played a role though? Power delivery to Kaby Lake-X differs from Skylake-X; the former receives Vcore from the board, the latter receives Vccin and then employs another on-package voltage regulator. Vcore is quite a bit lower than Vccin. (And at same power, lower voltage requires higher current of course.)

PS,
the English translation of Igor Wallossek's article lost some of the edges of his criticism of Intel's use of TIM from the original German article. IMO the German original made it more clear that Igor sees this as more relevant than VRM temperatures in practice. And that's with Skylake-X; it should be even more applicable to 4-core Kaby Lake-X.

 

[formulav8]

next gen Optane

z270 says that as well iirc correctly.

 

[formulav8]

That's one way of taking things out of context

z370 looks to be a re-badged z270 as mentioned by some awhile back as a stop gap to release CFL earlier than expected because of AMD. The z390 is obviously not ready yet so z370 is probably a stopgap. It at least looks that way so far.

 

[dullard]

z270 says that as well iirc correctly.

Z170 does not support Optane. Z270 supports Optane. The leaked 370 slide above says it now supports "next-gen Optane" in bold as a change from the Z270 chipset. I haven't seen the term "next-gen" Optane with Z270, although I could be wrong. I suspect that this is Intels Optane-based SSD which Intel claims is 7x faster than their own SSD.

 

[formulav8]

I haven't seen the term "next-gen" Optane with Z270, although I could be wrong.

I think I have which is why I said it. But I could be wrong. Honestly, not much matter right now.

 

[Ajay]

z370 looks to be a re-badged z270 as mentioned by some awhile back as a stop gap to release CFL earlier than expected because of AMD. The z390 is obviously not ready yet so z370 is probably a stopgap. It at least looks that way so far.

The chipset roadmap and CPU roadmaps don't agree. The chipset roadmap only has the z370 through Q2 2018 and the CPU roadmap lists the CNL PCH for 1H 2018. This is getting aggravating - I hope @Sweepr will be able to clarify this.

 

[formulav8]

Agree

 

[TheF34RChannel]

z370 looks to be a re-badged z270 as mentioned by some awhile back as a stop gap to release CFL earlier than expected because of AMD. The z390 is obviously not ready yet* so z370 is probably a stopgap. It at least looks that way so far.

The chipset roadmap and CPU roadmaps don't agree. The chipset roadmap only has the z370 through Q2 2018 and the CPU roadmap lists the CNL PCH for 1H 2018. This is getting aggravating - I hope @Sweepr will be able to clarify this.**

*that indeed seems to be the case!

**yes some clarification would be helpful and most welcome! Perhaps the CNL PCH will still be called CNL despite the lack of corresponding CPUs. I don't know, it's peculiar the way it is. It should be just one chipset; that'd make more sense imo. The answer to this may also answer my earlier question about what the heck these 6C are that are getting to get released alongside this CNL PCH.

 

[formulav8]

Me no know honestly, someone should though, to some degree.

 

[TheF34RChannel]

Me no know honestly, someone should though, to some degree.

My thoughts exactly! And additionally, if next year's 6C are better I'd like to know beforehand so I can choose to wait for those or not. I couldn't care less about the boards - unless they lock out CPUs (impossible).

 

[Sweepr]

@TheF34RChannel I believe they will update the 4-core and 6-core models by the time Cannon Lake PCH arrives (including 35W options). This slide provides different production windows for the Coffee Lake-S SKUs, depending on the PCH. Also, they might release another enthusiast chipset with the updated 300 Series features, but it will coexist with Z370. This future platform could have an important advantage, support for IMVP9 CPUs (Ice Lake) - of course it's only speculation at this point.

 

[TheF34RChannel]

@TheF34RChannel I believe they will update the 4-core and 6-core models by the time Cannon Lake PCH arrives (including 35W options). This slide provides different production windows for the Coffee Lake-S SKUs, depending on the PCH. Also, they might release another enthusiast chipset with the updated 300 Series features, but it will coexist with Z370. This future chipset could have an important advantage, support for IMVP9 CPUs (Ice Lake) - of course it's only speculation at this point.

This is the most interesting post today, thanks!

This would also put me in a tough spot and perhaps a holding pattern until certain things get confirmed - quite unhappily because of a any potential additional wait.

- Ice Lake (not interested) compatibility (and then by default Tiger Lake (interested), albeit speculation, with next year's chipset is interesting and compelling.

- a better 6C in 2018 sounds really good, yet I wonder in what capacity - i.e. it's doubtful they can squeeze a higher frequency out of it so where will the advantage lie? Power saving? In the end it seems an unnecessary part to me.

It leads to these two questions:

1 is the entire 2017 line-up a stop gap?

2 should I wait for the 2018 boards and CPU? (no one can answer this, I know.)

 

[formulav8]

This future chipset could have an important advantage, support for IMVP9 CPUs (Ice Lake)

That sir, would be very nice if true.

 

[TheF34RChannel]

That sir, would be very nice if true.

It would turn Z370 and corresponding CPUs into a an approximate three month band-aid though. Personally I wouldn't mind because 1 CPUs have a long longevity and 2 when I do upgrade a CPU substantially I prefer a new board as well.

We need more info about what's landing in 2018

 

[jpiniero]

I don't think Icelake is going to be compatible even if they had planned for it at one point. I would think they would want to pair Icelake with Tiger.

1 is the entire 2017 line-up a stop gap?

Yes

2 should I wait for the 2018 boards and CPU? (no one can answer this, I know.)

No at the K level. I really think Core as an architecture is nearly tapped out clock speedwise, so while they will throw an extra 2 cores in there for Icelake you probably won't get much at ST or it will even be slower. And then Tiger would just improve clocks some more but not much if any at higher speeds.

At the non-K level possibly since I do think the mhz/W is going to jump by a large amount. Core M in particular could be pretty nice.

 

[Shivansps]

But what a Z390 or any other 300 chipset has anything to do with IMVP9 ? chipsets are no much more than a PCI-E device.

 

[Arachnotronic]

No at the K level. I really think Core as an architecture is nearly tapped out clock speedwise, so while they will throw an extra 2 cores in there for Icelake you probably won't get much at ST or it will even be slower. And then Tiger would just improve clocks some more but not much if any at higher speeds

FUD. When has Intel released a mainstream part with a ST perf regression from previous?

 

[jpiniero]

FUD. When has Intel released a mainstream part with a ST perf regression from previous?

It'd be a clock speed regression, which obviously would affect absolute ST perf.

 

[Arachnotronic]

It'd be a clock speed regression, which obviously would affect absolute ST perf.

Performance = Clock speed * Perf/clock.

 

[DrMrLordX]

FUD. When has Intel released a mainstream part with a ST perf regression from previous?

Last time they actually did that was . . . hmm. I guess technically there were some situations where the 1.4 GHz Williamette P4 could lose to the 1 GHz PIII (let's not count the1.13 GHz PIII since it was recalled/cancelled). But the 1.8 GHz Williamette came out at about the same time so whatever.

 

[VirtualLarry]

Last time they actually did that was . . . hmm. I guess technically there were some situations where the 1.4 GHz Williamette P4 could lose to the 1 GHz PIII (let's not count the1.13 GHz PIII since it was recalled/cancelled). But the 1.8 GHz Williamette came out at about the same time so whatever.

What about Atom? That had slower ST than prior Intel x86-compatible CPUs. Granted, that was a then-new, lower-power-oriented arch.

 

[DrMrLordX]

Atom wasn't really a "mainstream" CPU though. It was basically a netbook CPU rather than a desktop or even laptop CPU, and it wasn't a flagship of anything but . . . netbooks.