Intel Skylake / Kaby Lake

[.vodka]

How do you justify claiming it can do 5GHz? So far the HEDT 6c and 8c have had a distinctly lower OC potential than mainstream 4c, even on the same generation (mainstream on average being at least one generation ahead).

14nm 6700k overclocks to about 4.5-4.6GHz and then it becomes problematic to scale higher, either because of a hard wall on your particular sample, or voltage, or heat generation becoming a problem. Delidding+CLU helped. At least it overclocked better than 14nm Broadwell... what a step backwards on that regard vs Haswell.

14nm+ 7700k can do 5GHz prime stable when delidded and using CLU to provide decent heat transfer to a decent heatsink. It's your very same 6700k implemented on a better process with some improvements on the iGPU side. See how process improvements and a better physical implementation help with clockspeeds and heat generation?

Coffeelake will be built on 14nm++ which features higher transistor performance at similar voltages than 14nm+. It's the same Skylake cores implemented on an even better process than the 7700k did, with the difference being that you now have two extra cores and some more L3 cache to feed the whole thing.

Seeing the evidence so far and Intel's by now quite extensive understanding of its Skylake architecture, it's not far fetched to expect 5GHz (or near 5GHz) on a decent heatsink when delidded (Intel won't start soldering its mainstream processors again now...) and using CLU just like on the 7700k. Overengineered motherboards won't have a problem supplying the power needed for those extra two cores. One can always get better cooling if needed.



HEDT has historically overclocked worse than mainstream implementations of the same architecture. HEDT dies are server/Xeon dies, those are much more complex and include lots of extra features... If you increase complexity, clock speed usually goes down. Coffeelake is a mainstream part, therefore this is a lean die without all the extra stuff that would go in a server part.


Yeah, it's a safe guess that 5GHz is a possibility with this.

 

[2blzd]

So who/why would anyone buy a quadcore Kaby-X on x299 HEDT when you can buy a newer hexcore a few months later? I know PCIE lane count is important...but man, having a 6c -1151 part surely undermines a 4c- 2066 part..Especially if clocks are similar.

one of the main reasons I went to x99 was for more cores. Kaby-X is a weird move/part from Intel even more so now with CFL in the picture.

 

[mikk]

Is there any indication whether or not Intel could be officially supporting higher memory speeds with the new chips? This would basically give them a free performance-plus unless they want to continue to artificially limit their platform…

DDR4-2667 support according to leaks.

http://semiaccurate.com/2017/04/24/intel-brief-press-sandy-ex-may-2/

Charlie is an idiot. Sandy Bridge-E lol, it's a clown.

 

[moonbogg]

Maybe he's being cynical because CPU's haven't changed much since Sandy. A 2600K will still do just about everything the new chips can do, and not much slower when OC'd either. 6/12 coffeelake does sound great, but again, I really don't like the idea of thermal paste. I want my cooler to be fully functional and not handicapped by a crappy CPU design that uses paste to transfer heat. I can't tolerate that personally. I'd rather have a slower CPU with solder. Maybe that's just me though. I can't imagine watercooling the chip, overclocking it to 4.8 and watching it hit 80c instantly because the stupid tooth paste can't transfer any heat. Unacceptable.

 

[DrMrLordX]

How do you justify claiming it can do 5GHz? So far the HEDT 6c and 8c have had a distinctly lower OC potential than mainstream 4c, even on the same generation (mainstream on average being at least one generation ahead).

Is that really true? The i7-5775c often topped out at 4.2 GHz (though Fugger had one that did 4.8 GHz, albeit with serious volts), and that's a common stopping point for Broadwell-E. Haswell-E hit 4.5 GHz pretty frequently in its 6c incarnations, while even the 4790k only rarely broke 4.6-4.7 GHz.

Kabylake with an improved process could probably hit 5.2 GHz fairly regularly in a 4c form, so 5 GHz in 6c/12t form is not outside the realm of possibility.

 

[lolfail9001]

How do you justify claiming it can do 5GHz? So far the HEDT 6c and 8c have had a distinctly lower OC potential than mainstream 4c, even on the same generation (mainstream on average being at least one generation ahead).

So far HEDT 6c and 8c had better OC potential than mainstream 4c, just saying.

 

[w3rd]

Q: How many FPS does 500Mhz get you in games, past 4Ghz..?

 

[CakeMonster]

So far HEDT 6c and 8c had better OC potential than mainstream 4c, just saying.

I might have used the wrong wording. What I meant was that HEDT does on average reach significantly lower clock speeds than mainstream 4c.

Which would make launching 6c mainstream a challenge since many of its potential buyers would want the same ST performance as the last 4c mainstream (OC and non-OC) for various uses (in particular gaming).

 

[t0mt0m]

It's Charlie, enough said. Same bucket as WCCF, the trash bucket.

Guess Intel may have not invited him to this event as he blasted Optane? He's been pretty critical of the recent CPUs too irc.
(Which is kinda fair for Optane - the pre-launch PR was overblown & whilst making a nice $$ Enterprise product, trying to also make a consumer product out of it makes a weird product where paying a little more and just getting an SSD makes more sense.)

 

[coercitiv]

I might have used the wrong wording. What I meant was that HEDT does on average reach significantly lower clock speeds than mainstream 4c.

Which would make launching 6c mainstream a challenge since many of its potential buyers would want the same ST performance as the last 4c mainstream (OC and non-OC) for various uses (in particular gaming).

I don't know about your wording, but I can assure you the reasoning is also highly debatable. Loosing 3-4% in ST performance to gain 30-35% MT performance is hardly a challenging selling proposition.

 

[lolfail9001]

What I meant was that HEDT does on average reach significantly lower clock speeds than mainstream 4c.

You worded it right, it's just that you're wrong still.

 

[pantsaregood]

HEDT parts will generally hit lower clock speeds than desktop parts simply because they have more cores.

If each Skylake core Intel produces has a 95% chance of being capable of 4.8 GHz, then (assuming core clock yields are independent - likely not entirely true) a 4 core Skylake SKU would only have an 81.5% chance of hitting 4.8 GHz on every core. A 6 core unit would be down to 73.5%.

Now, these numbers are obviously not representative. There's no way Skylake cores hit 4.8 GHz 95% of the time, but it does explain why CPUs with higher core counts typically have slightly lower headroom.

As for stock clocks - they're so much lower to stay within TDP limits, I would assume.

 

[happy medium]

As of 2/22/17, the top 78% of tested 7700Ks were able to hit 4.9GHz or greater.
https://siliconlottery.com/collections/all/products/7700k49g

I would think the chances of a 7700k hitting 4.8 are ~ 90%.

As of 2/23/17, the top 68% of tested 7600Ks were able to hit 4.9GHz or greater.
https://siliconlottery.com/collections/all/products/7600k49g

 

[sze5003]

I was debating going to a 7700k..Hmm is it 1151 compatible, should be.

 

[crashtech]

I was debating going to a 7700k..Hmm is it 1151 compatible, should be.

What do you mean? LGA1151 spans two generations of chipsets now, the 7700K is certainly compatible with 200 series chipsets, and possibly with 100 series chipsets, depending on BIOS revision.

The new 300 series chipset will likely be backwards compatible, the question here is whether or not Coffee Lake will be compatible with Z270, for instance. I'm leaning towards no.

 

[sze5003]

What do you mean? LGA1151 spans two generations of chipsets now, the 7700K is certainly compatible with 200 series chipsets, and possibly with 100 series chipsets, depending on BIOS revision.

The new 300 series chipset will likely be backwards compatible, the question here is whether or not Coffee Lake will be compatible with Z270, for instance. I'm leaning towards no.

Yea I haven't been keeping up on the latest which is why I have this thread bookmarked. I think coffee lake will change by the time it comes out.

 

[happy medium]

This is a huge thread.........
Question.
Is there any CONCRETE evidence supporting that Z170/Z270 motherboards will NOT support a 6C/12 thread Coffee lake cpu?
Please link me.

thanks

 

[Bouowmx]

SiSoftware result from earlier: http://ranker.sisoftware.net/show_r...d5e3dbecdfe6d4f280bd8dabceab96a680f3cef6&l=en

6-core processor running on Kaby Lake platform.
So, I'd wager, up to motherboard manufacturers to bring old chipsets up-to-date, like normal.

 

[ozzy702]

This is a huge thread.........
Question.
Is there any CONCRETE evidence supporting that Z170/Z270 motherboards will NOT support a 6C/12 thread Coffee lake cpu?
Please link me.

thanks

I'm really hoping that both Z170 and Z270 mobos can support it. I'll buy one day one if that's the case. I have an overbuilt mobo and huge cooler. 4.7ghz six core here I come.

 

[LTC8K6]

As of 2/22/17, the top 78% of tested 7700Ks were able to hit 4.9GHz or greater.
https://siliconlottery.com/collections/all/products/7700k49g

I would think the chances of a 7700k hitting 4.8 are ~ 90%.

As of 2/23/17, the top 68% of tested 7600Ks were able to hit 4.9GHz or greater.
https://siliconlottery.com/collections/all/products/7600k49g

Now factor in 14nm++ gains...

 

[SAAA]

HEDT parts will generally hit lower clock speeds than desktop parts simply because they have more cores.

If each Skylake core Intel produces has a 95% chance of being capable of 4.8 GHz, then (assuming core clock yields are independent - likely not entirely true) a 4 core Skylake SKU would only have an 81.5% chance of hitting 4.8 GHz on every core. A 6 core unit would be down to 73.5%.

Now, these numbers are obviously not representative. There's no way Skylake cores hit 4.8 GHz 95% of the time, but it does explain why CPUs with higher core counts typically have slightly lower headroom.

As for stock clocks - they're so much lower to stay within TDP limits, I would assume.

As of 2/22/17, the top 78% of tested 7700Ks were able to hit 4.9GHz or greater.
https://siliconlottery.com/collections/all/products/7700k49g

I would think the chances of a 7700k hitting 4.8 are ~ 90%.

As of 2/23/17, the top 68% of tested 7600Ks were able to hit 4.9GHz or greater.
https://siliconlottery.com/collections/all/products/7600k49g

Math is fun because it always works, let's dig a bit with these numbers, straight for the magical "5GHz!":

https://siliconlottery.com/collections/lga-1151/products/7600k50g

As of 2/23/17, the top 47% of tested 7600Ks were able to hit 5.0GHz or Greater.

https://siliconlottery.com/collections/all/products/7700k50g

As of 2/22/17, the top 59% of tested 7700Ks were able to hit 5.0GHz or greater.

Note how:
1) 7700K seems to be slightly better binned
2) on average 14nm+ quad cores have ~53% chance to hit our target
3) target -100/200MHz and you are looking at 75-90% respectively, not bad and real world difference is nil beside benches (I'd rather have better RAM or stability than 100MHz more)

Now using pantsaregood's method you can find the virtual % of a single core hitting 5GHz:

x^4=47% → x=83% (i5 estimate)
x^4=59% → x=88% (i7 estimate)

Now let's simulate a Kaby Lake 6 core:

83%^6=32%
88%^6=45%

One out of three or more could reach 5GHz, then...

Now factor in 14nm++ gains...

He took the words out of my mouth, basically I see a good amount of Coffee hitting 5GHz, say ~50% if the process gains eke up to these speeds.
Maybe less than Kaby but it's not a bad trade for 50% more cores...
Obviously quads should reach even higher speeds but you can disable cores and pick the best 4-5 out of 6, or set for 4.8 and still enjoy the best combination on the planet for cores, IPC and clocks.

 

[Arachnotronic]

Coffee Lake is going to be really awesome. Ice Lake will be even more awesome (I expect 8 core for the top mainstream part).

 

[Sweepr]

Intel's latest PCN, including the complete list of Xeon Gold / Xeon Platinum (launch) models:

http://qdms.intel.com/dm/i.aspx/645B281A-2F68-4DB1-A0E9-241C03D8697C/PCN115363-00.pdf

H0 stepping

 

[jwapk]

DDR4-2667 support according to leaks.

A jump of 267Mhz... and that's when we already have stable running DDR4-4000+ modules for a year. I mean they could just validate the processors for the higher memory speeds, but obviously they won't. Just holding back that memory speed because it would only hurt Optane and they can then make another 15% plus with 14nm+++ with a jump to 2800Mhz -.-
If that continues with about 250Mhz/1H we'll reach DDR4-4000 by 2020, 4 years after market introduction.

 

[Ajay]

Intel's latest PCN, including the complete list of Xeon Gold / Xeon Platinum (launch) models:

http://qdms.intel.com/dm/i.aspx/645B281A-2F68-4DB1-A0E9-241C03D8697C/PCN115363-00.pdf

H0 stepping

H0 - really? Geez!