Intel Skylake / Kaby Lake

[moonbogg]

Buy the HEDT platform so that if you buy one and change your mind, you don't need to rip and replace your motherboard

I love this idea, seriously. For the first time we can have one board and swap CPUs from quad to up to 10 cores. I like it. That was a sweet move by Intel. Actual upgradeability for once. I got a feeling that for a high end gaming rig the sweet spot might be a 6 core rig for the Intel camp this time around.

 

[IntelUser2000]

4.5Ghz vs. 4.2Ghz is only a 7% boost, not 10%. If 7700K can only overclock to 4.7-4.8Ghz, it means Intel just pre-overclocked the more mature 14nm 6700K cores for the consumer.

I agree 100% with this. Perhaps we'll see 2-3% improvements since they were talking about KBL having "key" performance changes.

Also, no chip on sale running on air(regardless of company, architecture, pipeline stages, process) has went above 5GHz. The highest-ever clocked chips go slightly over 9GHz, with extreme cooling like LN2. There's some fundamental barrier in clocks that no one seems to be able to penetrate(and no one wants to explain). Both 4790K and 6700K leaks claimed 5.2-5.5GHz but never panned out in reality. They both ended up in the 4.8GHz range. All the fancy technologies like High-K, Strain, FinFET pale in comparison with what we used to get on plain boring shrinks before that.

Power 6/8: 5GHz fastest, 5.6GHz claimed but only in labs
FX-9590: 5GHz Turbo
DC/SKL: 4.8GHz overclock

The stock 5GHz chips like Power and FX only does that with insane 200W+ TDP levels. Ever since Prescott CPU manufacturers struggled mightily to bring CPU performance much above that level. Clock speed barrier is the reason.

You might see a 4.8GHz chip one day, but with zero overclocking headroom. Well, if we are fortunate enough we might see that chip with 5GHz Turbo.

 

[imported_ats]

I agree 100% with this. Perhaps we'll see 2-3% improvements since they were talking about KBL having "key" performance changes.

Also, no chip on sale running on air(regardless of company, architecture, pipeline stages, process) has went above 5GHz. The highest-ever clocked chips go slightly over 9GHz, with extreme cooling like LN2. There's some fundamental barrier in clocks that no one seems to be able to penetrate(and no one wants to explain). Both 4790K and 6700K leaks claimed 5.2-5.5GHz but never panned out in reality. They both ended up in the 4.8GHz range. All the fancy technologies like High-K, Strain, FinFET pale in comparison with what we used to get on plain boring shrinks before that.

There is a fundamental limit as Delay(Transistor) --> 0 of the intrinsic RLC delays. A lot of paths fundamentally are sitting at ~50% RLC delay and ~50% switching delay. You can cut both by further hyper pipelining but that tends to run into diminishing returns rather quick as the DFlop delay ends up taking more and more of the cycle time (and gets even worse as you push higher frequencies because a decent chunk is frequency independent). As it is, for a 20 FO4 cycle time, you generally only have say 15 FO4 of logic and wire time. If you want to go down to say 10 FO4 cycle time, you end up with 5 FO4 of logic/wire delay and as you push Freq higher Flop delay may increase to 6 or 7 FO4 due to static uncertainties like jitter et al. And as you push things like process density things like RLC become more of a concern since the RLC scaling gets worse with smaller wires and smaller wire gaps.

 

[Arachnotronic]

I love this idea, seriously. For the first time we can have one board and swap CPUs from quad to up to 10 cores. I like it. That was a sweet move by Intel. Actual upgradeability for once. I got a feeling that for a high end gaming rig the sweet spot might be a 6 core rig for the Intel camp this time around.

The Basin Falls platform will actually last for 3 generations before the socket is changed, so anybody buying it should have upgrade path to >10 cores down the line once Cannonlake-X/Icelake-X arrive.

 

[daniel1926]

There is a fundamental limit as Delay(Transistor) --> 0 of the intrinsic RLC delays. A lot of paths fundamentally are sitting at ~50% RLC delay and ~50% switching delay. You can cut both by further hyper pipelining but that tends to run into diminishing returns rather quick as the DFlop delay ends up taking more and more of the cycle time (and gets even worse as you push higher frequencies because a decent chunk is frequency independent). As it is, for a 20 FO4 cycle time, you generally only have say 15 FO4 of logic and wire time. If you want to go down to say 10 FO4 cycle time, you end up with 5 FO4 of logic/wire delay and as you push Freq higher Flop delay may increase to 6 or 7 FO4 due to static uncertainties like jitter et al. And as you push things like process density things like RLC become more of a concern since the RLC scaling gets worse with smaller wires and smaller wire gaps.

Can you explain this a in a little less technical language for those of us not in the engineering domain?

 

[IEC]

Can you explain this a in a little less technical language for those of us not in the engineering domain?

Translation: There is a fundamental technical limit to current materials and design tradeoffs which is making >5GHz chips a difficult problem to solve.

As a current OC 6700K user, it's hard to get excited unless we see the 7700K break the 5GHz barrier.

 

[RussianSensation]

No, you are misunderstanding what Kaby Lake is. Kaby Lake is the same micro-architecture (more or less) as Skylake, but the actual circuit design is improved and it is manufactured on an enhanced 14nm process. It's not just pre-overclocked/cherry picked Skylake chips.

I am not misunderstanding KBL. If both processors max out at roughly similar overclocks of 4.7-4.8Ghz, the enhanced 14nm process is a marketing gimmick that may translate in 5-10W reduction in power usage. If 7700K can consistently overclock to 4.9-5Ghz or can overclock to 4.7-4.8Ghz on stock voltage, then I will agree with you. Until we see any proof that 7700K is actually a better processor when maxed out than 6700K is, your posts are just conjecture unless you have insider info. Considering how much of an overclocking flop BW-E ended up, I am inclined to be cautious than lap up Intel's marketing claims of enhanced 14nm node. The faster iGPU performance is useless for most of us with dedicated GPUs w/ HDMI 2.0 and all that 4K decoding jazz. Besides, anyone who cares about high quality video encoding isn't using iGPU/GPUs for it but MOAR cores. In that case, 7700K will bring nothing to the table as it's still a quad core.

As it stands, per Silicon Lottery:

"As of 5/3/16, the top 59% of tested 6700Ks were able to hit 4.7GHz or greater."

They sell a 4.7Ghz 6700K for $339.99.

We'll see what Kaby Lake brings to the table for desktop overclockers, but for mobile it's a clear win: much higher clocks in the same power envelope + improved media engine.

I didn't say anything about mobile parts. My issue was with a generalized statement of how KBL would bring 10-25% performance boost, while ignoring the most important market segment for this forum - desktop - aka i7 6700K/i5 6600K.

If you want to talk about mobile, it's been 3.5 years since I purchased my Ivy Bridge i7 3635QM laptop. Its spiritual $378 SKU Skylake replacement, the 6820HQ, is only 20% faster (92.8 rating vs. 77.1 rating). By February 2017, my laptop will be 4 years old. You think it's something to celebrate that a mobile Kaby Lake may add another 200-300mhz on top of the 6820HQ after another year? At this pace it's going to take until late 2018 before there is a 45W TDP mobile i7 that's 40-50% faster than my Winter 2013 Ivy.

That's another issue with Kaby Lake -- it's not a new architecture/generation. For someone with a fast laptop, such as mine, it's still worth skipping and waiting until Ice Lake in 2018 or even beyond. Unless gaming or needing a new laptop for better battery, Intel is not providing enough of a CPU increase in the mobile space to warrant upgrading.

It's actually SKX-X, not SKL-X

Anyway, SKX-X should be interesting because it'll be using the Skylake Xeon cores rather than the plain Jane Skylake cores; that means AVX-512 and if David Kanter is right, a larger L2$. I wouldn't be surprised if perf/clock at the core level for SKX-X is higher than SKL-S or KBL-S/KBL-X. [/quote]

By the time AVX 512 is used, we'll be long past the Skylake architecture. Not sure why you are bringing up Intel marketing bullet-points to try and defend the fact that Skylake-X is going to end up 1.5-2 years behind a 4.6-4.8Ghz i7 6700K.

As far as six core mainstream, it looks to me that Intel's goal is to move enthusiasts from the mainstream DT platform to the enthusiast platform. That's why they are doing an interesting, iGPU-less KBL-X SKU as well as offering the SKX-X parts.

I am not denying that. We'll see how successful they are. MicroCenter in the US makes it possible to buy a good $75 Z170 board and a great $105 Z170 board, when paired with a $299.99 i7 6700K. A good X99 board costs $160 and i7 6800K is $399.99. That means the i7 6700K platform would cost $375-405 but the 6-core last generation's architecture with X99 would cost $560 USD. That's a huge difference that could be used towards a better monitor with GSync/FreeSync, stepping up from RX 480/1060 to the GTX1070, or from GTX1070 to the 1080, setting $ aside for a future GPU upgrade, etc. On a value proposition for gaming, the X99 platform loses badly and I don't see how that would change with a 6-core Skylake. Unless Intel purposely delays Ice Lake beyond 2018, games suddenly start benefiting greatly from 6-8 core CPUs, Intel raises prices on the 4C+HT 8700K to $400-450 USD, SKL-X is going to end up as another waste of $ workstation platform.

A reasonable strategy for a 6700K owner would be to make a platform move to Basin Falls (KBL-X/SKX-X platform) and enjoy the ability to use three generations of processors spanning from high TDP/specially binned quad core parts to very high core count SKX-X parts.

I disagree completely. How is it reasonable to sell a perfectly viable fast 6700K gaming platform and upgrade to the same 2-year-old CPU architecture with more underutilized cores, when Ice Lake -- Intel's next major architecture -- is slated for 2018? Given the current pricing premium for X99 over Z170, it stands to reason that i7 7700K + Z270 will continue to enjoy the same $150-200 pricing advantage over the 7800K 6-fore SKL-X.

In hindsight, the 6700K gamer/user would have enjoyed top-of-the-line performance for 1.5-2 years before SKL-X even drops. At that point, might as well wait for Intel's next architecture (see above).

Your other statement about 3 generations of processors I also don't agree with. To me, despite Intel's marketing games, they have had only 4 true architectures post Core 2 Duo/Quad:

1st - 2008 - Nehalem (i7 920)
2nd - 2011 - Sandy Bridge (i7 2600K)
3rd - 2013 - Haswell (i7 4770K)
4th - 2015 - Sky Lake (i7 6700K)
Next is 2018 - Ice Lake. Everything in between is just "marketing generations." In the past when CPUs evolved rapidly, it would have just been the move from Q6600 to Q9550. Today, Intel would call Q9550 a new architecture like Cannon Lake. Marketing gimmick to get something new for OEMs. Kaby Lake is even worse than Q6600 to Q9550 because Intel actually improved the architecture of Kentsfield.

Good times coming for CPU enthusiasts, IMO.

I disagree. Major Intel architectures are exciting. Skylake-X and Kaby Lake-X are the same underlying architectures, slightly refined 14nm node, some more marketing instructions that won't be used for years to come, and just MOAR cores. If someone needs a $1000-1800 CPU for productivity, sure, but for gaming, none of these 6-10 core CPUs have proven their worth yet. I recommended i7 5820K for a while because of how grossly overpriced early i7 6700K + Z170 + fast DDR4 combinations were. But at this time, X99 is a waste of $ for gamers (mostly bragging rights and emotions, not perfomrance). If Ice Lake brings a 15% increase in IPC and is priced reasonably, the same will be true for the next workstation platform. The next workstation platform won't even have PCIe 4.0, right?

Good times was when the Enthusiasts received the flagship platform first (X58 + i7 920/930), nearly a full year before Lynnfield i7 860 launched. You paid early adopter fees for expensive X58 mobos and DDR3 but the processor only cost $284 and you had 1 year of headstart over the mainstream users. Not only that, but the i7 920 overclocked better on stock voltage and with overvoltage than the i7 860/870/875K/880, it allowed higher safety margin for QPI voltage. Today's workstation Xeon CPUs you show no inherent overclocking advantages. In fact, Broadwell-E overclocks much worse than Skylake or even Haswell-E. Therefore, it's also a stretch to assume SKX-X or KBL-X could even match Ice Lake in overclocking headroom.

Good times was also when the Enthusiasts received the refreshed Ivy Bridge (X79 + i7 3930K), less than a year after i7 2600K launch. In constrast, you are saying it's good times when SKX-X will launch 1.5-2 years after i7 6700K? That's an even longer delay than in the past. How is someone supposed to be excited about spending $400-1000 on the same architecture almost 2 years later? HA!

What Intel is doing is slowly killing the workstation platform. It's never been worse. I bet i7 5820K users are patting themselves on the back buying that amazing CPU in August 2014.

I love this idea, seriously. For the first time we can have one board and swap CPUs from quad to up to 10 cores. I like it. That was a sweet move by Intel. Actual upgradeability for once. I got a feeling that for a high end gaming rig the sweet spot might be a 6 core rig for the Intel camp this time around.

To me actual upgradeability would be if I could upgrade to new Intel architectures without needing to change the motherboard. I am hesitant to believe that Intel will support anything beyond Cannon Lake on the next workstation platform. Cannon Lake is likely a repeat of what Ivy was to Sandy or Broadwell was to Haswell - nothing special.

Now imaging dropping cash on a new $250 SKL-X mobo and $400-1000 on a new SKL-X/KBL-X CPU and then in 2018, we get Z370 with PCIe 4.0, naturally M.2 PCIe 4.0 SSDs support, another 10-15% IPC from Ice Lake, and all of that for less $. It practically means the next workstation platform will be 'obsolete' tech in a year unless Intel purposely delays Ice Lake or butchers it somehow. When you purchased your 3930K, it was a refreshed architecture of Sandy and it still managed to come out in the same year as Sandy. You are Moonbogg, you should be pissed as heck that Intel is throwing 2-year-old architecture scraps and on top of it is charging premiums on the mobo and CPU side. But wait, you do get MOAR cores. That's great for 3DMark scores I bet! A $1000 5960X can't even beat a $290 i7 6700. $700 flushed right down the toilet. You think Skylake-X will be any different once Ice Lake shows up? All this time you were bottlenecking your 980Ti SLI with 3930K while you could have just purchased the i7 6700K in August 2015, overclocked it on water to 4.8Ghz and enjoy the same level of CPU performance in games from 2015 to 2018, skipping Skylake-X and Kaby Lake-X entirely. That is exactly why even more delays for the Intel's workstation platform (X58 - 1 year ahead, X79 - same year as Sandy, and now we are 1.5-2 years late!) makes it an even bigger failure for gamers than ever. The only purpose of the workstation platform has become more cores for workstation uses/productivity and the rest are bragging rights of having the "best consumer CPU" (only on paper).

I agree 100% with this. Perhaps we'll see 2-3% improvements since they were talking about KBL having "key" performance changes.

There is also the opportunity cost of waiting. If someone was on Nehalem/Lynnfield (i7 920-960/860) or Sandy/Ivy (2500K/2600K/3770K, etc.), then what was the point of waiting for Kaby Lake an entire year and not grabbing an i7 6700K when prices subsided? If the 2600K/3770K gamer skipped i7 6700K, what does 7700K change about the situation? Might as well wait all the way and get a true next generation Ice Lake architecture. In the meantime, those i7 Sandy/Ivy CPUs overclocked to 4.5-4.8Ghz should be good for 1440P + GTX1070/1080 anyway. By 2018, this gamer could do an entire system overhaul with Ice Lake + Volta/AMD's equivalent. With Intel's new strategy of moving from a 2-year to a 3-year architecture cadence, imho, it makes the most sense to upgrade via newest architecture cycles and skip all the refreshes in-between. That's why the next key year for upgrading for Sandy/Ivy users is 2018 Ice Lake, not 2016 Kaby or 2017 Cannon Lake. Those are all derivatives of the same Skylake cores/architectures. Another option could be to wait for clearance of i7 6700K once i7 7700K launches. If i7 6700K drops to $249.99 at places such as MicroCenter, it will be a slam dunk upgrade.

 

[jpiniero]

Intel really seems to be pushing enthusiasts onto HEDT. You may not have much choice by the time Ice Lake rolls around.

 

[Sweepr]

Strong cores still matter, even in a DX12 (AMD) Gaming Evolved title that seems to love cores:

Strong showing for Skylake-S here, stacks up well next to Haswell-E.

I didn't say anything about mobile parts. My issue was with a generalized statement of how KBL would bring 10-25% performance boost, while ignoring the most important market segment for this forum - desktop - aka i7 6700K/i5 6600K.

Your issue is twisting other people's words to fit your POV. Apparently two other users already told you, but hey, third time's the charm. I was talking about Kaby Lake-U, not desktop. Pardon me for being excited that Intel was able to extract 10%/25% worth of CPU/iGPU almost exclusively from an enhanced process.

 

[mikk]

I disagree completely. How is it reasonable to sell a perfectly viable fast 6700K gaming platform and upgrade to the same 2-year-old CPU architecture with more underutilized cores, when Ice Lake -- Intel's next major architecture -- is slated for 2018?

Icelake isn't slated for 2018. Forget about it. Cannonlake and Coffee Lake will be available in 2018. Although the Basin Falls platform doesn't look appealing to me in case there will be a sixcore Coffee Lake for S1151 half a year after Basin Falls, imho six cores would be a good sweet spot between cores and clock frequency for a mainstream platform. Skylake-X may clock even higher than the upcoming i7-7700k but there isn't much headroom left for a major clock speed improvement I believe.

 

[Sweepr]

New rumor saying Z270 motherboards could arrive in October:

Intel’s product roadmap pointed to the release of the company’s optimized Kaby Lake products sometime in 2016, but so far they haven’t shown up. According to Colorful, however, Kaby Lake motherboards may arrive this fall.

Specifically, Colorful plans to release it Z270 motherboards sometime in October. The company couldn't tell us when Kaby Lake CPUs would come out, though, and it's possible that we will see Kaby Lake motherboards show up on the market before the corresponding processors.

www.tomshardware.com/news/intel-kaby-lake-motherboards-october,32511.html


Some fun bits (and chips?): the same guy who said Broadwell-E 10C/20T was launched because Intel is afraid of Summit Ridge (Zen) now implies i7-7700K is another desperate attempt (sacrificing efficiency), and makes a fuss about its (rumored) 4W higher TDP compared to i7-6700K. Won't provide a link because he doesn't deserve the clicks.

 

[Sweepr]

Some exciting new to share, thanks to NotebookCheck. Intel will introduce quad-cores to their Kaby Lake-U mobile lineup (15W TDP). This means premium ultraportables / convertibles get a singificant boost in terms of multi-thread performance next year.

Intel: Kaby Lake quad-core ULV processors in 2017

www.notebookcheck.com/Intel-Kaby-Lake-Quad-Core-ULV-Prozessoren-in-2017.172185.0.html

The new 4C/8T SKU complements the rather impressive 200-500 MHz clockspeed bump expected for the traditional dual-core SKUs. Mobile Kaby Lake looks promising.

 

[Lepton87]

If you want to talk about mobile, it's been 3.5 years since I purchased my Ivy Bridge i7 3635QM laptop. Its spiritual $378 SKU Skylake replacement, the 6820HQ, is only 20% faster (92.8 rating vs. 77.1 rating). By February 2017, my laptop will be 4 years old. You think it's something to celebrate that a mobile Kaby Lake may add another 200-300mhz on top of the 6820HQ after another year? At this pace it's going to take until late 2018 before there is a 45W TDP mobile i7 that's 40-50% faster than my Winter 2013 Ivy

You cherry-picked that comparison to make the progress look as bad as possible. IvyBridge was a great CPU from the efficiency point of view so much that it's actually more efficient then its successor that is an anomaly and an one-off event. Making that comparison to haswell wouldn't look as bad.

IB-E is way ahead of haswell-e in terms of efficiency and practically on par with broadwell-e.

The same in the mainstream platform. Not even every broadwell SKU is more efficient then 3770K here. I also bought a laptop with the ivy 3630QM and even now this CPU is hard to beat in terms of efficiency,

Now imaging dropping cash on a new $250 SKL-X mobo and $400-1000 on a new SKL-X/KBL-X CPU and then in 2018, we get Z370 with PCIe 4.0, naturally M.2 PCIe 4.0 SSDs support, another 10-15% IPC from Ice Lake, and all of that for less $. It practically means the next workstation platform will be 'obsolete' tech in a year unless Intel purposely delays Ice Lake or butchers it somehow. When you purchased your 3930K, it was a refreshed architecture of Sandy and it still managed to come out in the same year as Sandy. You are Moonbogg, you should be pissed as heck that Intel is throwing 2-year-old architecture scraps and on top of it is charging premiums on the mobo and CPU side. But wait, you do get MOAR cores. That's great for 3DMark scores I bet! A $1000 5960X can't even beat a $290 i7 6700. $700 flushed right down the toilet. You think Skylake-X will be any different once Ice Lake shows up? All this time you were bottlenecking your 980Ti SLI with 3930K while you could have just purchased the i7 6700K in August 2015, overclocked it on water to 4.8Ghz and enjoy the same level of CPU performance in games from 2015 to 2018, skipping Skylake-X and Kaby Lake-X entirely. That is exactly why even more delays for the Intel's workstation platform (X58 - 1 year ahead, X79 - same year as Sandy, and now we are 1.5-2 years late!) makes it an even bigger failure for gamers than ever. The only purpose of the workstation platform has become more cores for workstation uses/productivity and the rest are bragging rights of having the "best consumer CPU" (only on paper).

.

Unless the mainstream Ice-Lake is going to be available in the 6-core flavor I don't think it's going to be a better CPU then skylake-x even for games. Why? There are at least two things that will be different this time around and probably the situation with overclocking should also be different. I don't believe Ice-lake will OC higher to the extent of skylake over broadwell.
1.By the time Ice-Lake launches even games should make reasonable use of more than 4C/8T.
2. Quad-channel memory. This may actually bring the biggest boost in performance. Skylake gains much more performance from fast memory then previous architectures. Part of that is because the uncore is clocked at the core speed so the memory performance is not limited by the uncore speed to the same extend as in haswell-e and broadwell-e. Quad-channel DDR4+ at 4000MHz+ may be a big advantage to skylake-x that should give a very big boost to gaming performance compared to a regular skylake and of course the previous HEDT plarform. I think the gain from this alone can equal half of the IPC improvement from Ice-lake. What's unknown is the overclocking headroom of skylake-x and ice-lake. Which CPU is going to be better may very well come to this but the chances are high that there won't be repeat of the current situation where the overclocked skylake with fast memory outperforms HEDT platform in most games. I think skylake-x should be a sound choice that won't be surpassed by the mainstream Ice-Lake.

 

[jpiniero]

Some exciting new to share, thanks to NotebookCheck. Intel will introduce quad-cores to their Kaby Lake-U mobile lineup (15W TDP). This means premium ultraportables / convertibles get a singificant boost in terms of multi-thread performance next year.

Not a surprise given that earlier "leak", but BTS 17? That's going to be very confusing with this, Coffee Lake (which I guess is just the 2+3e U and 4/6+3e H models), and Cannonlake (2+2 U and Y) all released in about the same time.

 

[frozentundra123456]

Some exciting new to share, thanks to NotebookCheck. Intel will introduce quad-cores to their Kaby Lake-U mobile lineup (15W TDP). This means premium ultraportables / convertibles get a singificant boost in terms of multi-thread performance next year.

Intel: Kaby Lake quad-core ULV processors in 2017

www.notebookcheck.com/Intel-Kaby-Lake-Quad-Core-ULV-Prozessoren-in-2017.172185.0.html

The new 4C/8T SKU complements the rather impressive 200-500 MHz clockspeed bump expected for the traditional dual-core SKUs. Mobile Kaby Lake looks promising.

Whatever happened to the +4e igpu? And no edram at all for desktop, either SL or KL? So many rumored/announced models that never really show up. Lets hope that the 4+2 mobile part actually becomes a reality and is widely used, rather than just vaporware. IDK, I find it hard to get really excited, especially until we see more tests as to whether they can maintain the faster clockspeeds under sustained load. Was also hoping for more widespread use of edram.

 

[Sweepr]

Not a surprise given that earlier "leak", but BTS 17? That's going to be very confusing with this, Coffee Lake (which I guess is just the 2+3e U and 4/6+3e H models), and Cannonlake (2+2 U and Y) all released in about the same time.

My thoughts as well. In 2017/2018 we will have:

- Y
10nm Cannonlake 2C/4T + GT2

- U
10nm Cannonlake 2C/4T + GT2
14nm Kaby Lake 4C/8T + GT2
14nm Coffee Lake 2C/4T + GT3e

- H
14nm Coffee Lake 2C/4T + GT3e
14nm Coffee Lake 4C/8T + GT3e
14nm Coffee Lake 6C/12T + GT3e

On the plus side, if the S lineup is based on Coffee Lake - desktops get standard GT3 + eDRAM across the entire Core (i3/i5/i7) line. That would provide a nice CPU boost to certain applications and a giant leap in terms of iGPU performance compared to KBL-S GT2. I imagine Intel will keep the cheaper and smaller (die) dual-core Kaby Lake-S for Celeron/Pentium. Kaby Lake-U 4C/8T only makes sense if a Cannonlake version would not be possible at similar time frame.

Whatever happened to the +4e igpu?

They are indeed missing from this roadmap. Last time I saw KBL-H GT4e mentioned was last year.

And no edram at all for desktop, either SL or KL?

Kaby Lake-S with eDRAM is probably mobile only, but SKL-S GT4e might be released as a LGA 1151 part later this year, maybe 2017. Biggest hope for more desktop parts with eDRAM is Coffee Lake, that is, if they make desktops SKUs out of it.

 

[DrMrLordX]

It would be really, really weird for them to do Skylake-S GT4e on the desktop in 2017 instead of Kabylake-S GT4e. Why would they even do that?

 

[majord]

There was a Presentation at HotChips yesturday


Inside 6th generation Intel Core code named Skylake:: New Microarchitecture and Power Management

By Jack Doweck. Wonder if a Video for it will surface?, and what details he went into - SL has been a bit of a mystery since lanch

 

[Nothingness]

There was a Presentation at HotChips yesturday


Inside 6th generation Intel Core code named Skylake:: New Microarchitecture and Power Management

By Jack Doweck. Wonder if a Video for it will surface?, and what details he went into - SL has been a bit of a mystery since lanch

I saw the slides, and found nothing interesting in it as far as micro-arch goes.

 

[SAAA]

I saw the slides, and found nothing interesting in it as far as micro-arch goes.

Sad. AMD and IBM release so many interesting detail and here we are knowing 0 more than last year, Skylake for server is still a mistery as well as Kabylake. Aren't they releasing both in less than a year? Not to talk about Cannonlake that supposedly comes out Q3 next year too, even if dual core only.

Things are getting so boring that I might skip everything and see in 2018 how's the situation, I'm thinking six core mainstream and Zen+ fighting each other... today is all quad here quad there, hey this is 8 but costs 1000$!

 

[Paul98]

I want to get a new laptop now that the mobile GTX 1070 is out. I am debating if I just go for one of them now with a 6700HQ, or wait for Kaby Lake. It sounds like it will be a while before a mobile version comes out, and the desktop versions are usually super expensive in laptops. Any suggestions?

right now looking at the Sager NP8173-S, with an upgrade to 32gb of ram, and a 512GB ssd

 

[Arachnotronic]

I want to get a new laptop now that the mobile GTX 1070 is out. I am debating if I just go for one of them now with a 6700HQ, or wait for Kaby Lake. It sounds like it will be a while before a mobile version comes out, and the desktop versions are usually super expensive in laptops. Any suggestions?

right now looking at the Sager NP8173-S, with an upgrade to 32gb of ram, and a 512GB ssd

I'd just go with the Skylake laptops. The improvement likely won't be big enough with Kaby Lake to be worth the wait.

 

[mikk]

Detailed Roadmap: http://www.notebookcheck.com/Intel-Detaillierter-Kaby-Lake-Zeitplan-durchgesickert.172274.0.html

And two Slides: https://benchlife.info/intel-kaby-lake-will-get-100-design-08252016/

 

[ShintaiDK]

3 slides actually from benchlife. Last one that is broken on the page.

 

[mikk]

That's a good increase for just a new stepping, in fact the shift from Broadwell to Skylake wasn't higher afaik. Of course the improvements comes also from the improved 14nm process.