Intel Skylake / Kaby Lake

Page 137 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Mar 10, 2006
11,715
2,012
126
Looks like 100mhz increases in base and single-core turbos compared to Broadwell. The dual core turbo sees a bigger boost of 200mhz on the i5s. iGPU also sees a boost of 100mhz.

This might suggest improvements/refinements in the 14nm process.
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,710
182
106
Its a chain upgrade and yes. 3570K will be going out. Also getting a Skylake NUC as HTPC.

Ok, I see. So what's the reason for the upgrade? If it's a chain upgrade (I assume; IB->Haswell, Haswell->Skylake) the gains will be minimal.

Skylake NUC could be nice though.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
From 2C/2T/1.6 to 2C/4T/2.6 there s a ratio of 3 TDP wise.

Now you can always state that the higher clocked part is at 15W, and that the rest is well below this rating, 100% sure that you ll find some believers...

Here's my prediction. Carrizo won't come close in CPU performance and you are going to come up with the same lame excuses as always.
 

dahorns

Senior member
Sep 13, 2013
550
83
91
From 2C/2T/1.6 to 2C/4T/2.6 there s a ratio of 3 TDP wise.

Now you can always state that the higher clocked part is at 15W, and that the rest is well below this rating, 100% sure that you ll find some believers...

Uh, wouldn't binning also explain it?
 

Abwx

Lifer
Apr 2, 2011
10,854
3,298
136
Here's my prediction. Carrizo won't come close in CPU performance and you are going to come up with the same lame excuses as always.

Lol...

It wont come close thanks to underspecced TDPs..

The genuine 15W are to be find at the bottom of the line, higher clocked part is roughly 25-28W.

At some point one has to be realistic, even without accounting HT, to make things simple, going to 2.6 from 1.6 will mandate (2.6/1.6)^2 = 2.64x more power..

Although by negating physical laws that govern CPUs you could easily negate this fact, after all that s your only mean left to explain the forcibly unexplainable, at threads lengths...

Armed with such foolproof science you could perhaps also explain us why Intel didnt rate the lower parts at say 7W if 15W was really possible at 2.6, certainly that you know better than they do...

Uh, wouldn't binning also explain it?

That s eventualy possible if they find some that are 7nm parts...

Binning works for a few % voltage, you ll hardly get more than 10% better perf/Watt.
 
Last edited:

JTsyo

Lifer
Nov 18, 2007
11,709
871
126
Amazon now has the i5 6600K available, kind of tempting but waiting to see if Mircocenter pulls a combo sale next week.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Ok, I see. So what's the reason for the upgrade? If it's a chain upgrade (I assume; IB->Haswell, Haswell->Skylake) the gains will be minimal.

Skylake NUC could be nice though.

Because its faster and time is money, even when gaming. Prescripted benchmarks and reality are often very far from one another.

The Skylake one is 20-30% over the Haswell. The Haswell is 10%+ over the IB.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
Because its faster and time is money, even when gaming. Prescripted benchmarks and reality are often very far from one another.

The Skylake one is 20-30% over the Haswell. The Haswell is 10%+ over the IB.
At lower TDPs Skylake is able to show such increments. c't tested the 6700K and at about the same TDP of Haswell it had 40W lower power consumption under load. Performance gains ~5% in CB, Sysmark '14, ~9% in Linpack (helped by DDR4-2133 vs. DDR3-1600), 13% in Luxmark and 18.5% in 3DMark Firestrike.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Ok, so only 20-30% more in MT loads then.

BTW: Moar threads. I wasn't expecting that from you. :p

Actually it's ~50% more in MT applications (overall). ;)
www.hardware.fr/articles/940-19/indices-performance-cpu.html

The GT3 SKUs of the U series should be numbered 6x50; all of the models listed here are 6x00. The initial batch of Skylake-Us may not have GT3e and might be limited to GT2.

Agreed. I would just like to know a bit more about the new 15W GT3e SKU and when it launches (hopefully still in 2015). Waiting for the next leaks.

This might suggest improvements/refinements in the 14nm process.

Looks like a solid tock.
Here's Broadwell-U and Haswell-U in comparison.

http://www.fanlesstech.com/2015/08/exclusive-skylake-u-lineup.html

lAkG4fc.png


kRbHC2z.png


akcAlaq.png

http://www.extremetech.com/wp-content/uploads/2015/01/Intel-15W-SKU.png

Intel-Haswell-dual-core-U-series.png


dahorns said:
Looks like 100mhz increases in base and single-core turbos compared to Broadwell. The dual core turbo sees a bigger boost of 200mhz on the i5s. iGPU also sees a boost of 100mhz.

Core i5 6200U is probably going to be one of the most popular models, nice to see a 200MHz boost to dual-core Turbo coupled with Skylake's better IPC and faster graphics.
Users coming from Haswell-U or older notebooks are going to experience some significant gains.

Taking a look at the Surface Pro 3 and what we should see inside the Surface Pro 4:
- Surface Pro 3 Core i3 4020Y: 1.5GHz dual-core, no turbo, GT2 Gen 7.5 iGPU.
- Surface Pro 4 Core i3 should receive either Core-M (not likely according to leaks) or Core i3 6100U : 2.3GHz dual-core, better IPC, no Turbo, GT2 Gen 9 iGPU.

- Surface Pro 3 Core i5 4300U: 1.9GHz dual-core, 2.9GHz turbo, GT2 Gen 7.5 iGPU.
- Surface Pro 4 Core i5 6300U : 2.4GHz dual-core, 3.0GHz Turbo, GT2 Gen 9 iGPU.

- Surface Pro 3 Core i7 4650U: 1.9GHz dual-core, 2.9GHz turbo, GT3 Gen 7.5 iGPU.
- Surface Pro 4 Core i7 should receive either Skylake-U GT3e (real Iris Pro with eDRAM) or Core i7 6600U : 2.6GHz dual-core, 3.4GHz Turbo, GT2 Gen 9 iGPU.
Note that a Skylake GT2 Gen 9 iGPU should easily be faster than Haswell's GT3 Gen 7.5 iGPU at 15W, but I'm hoping they will postpone the Core i7 model or put the Core i7 6600U in there and include Skylake-U GT3e SKU as an option later on.
 
Last edited:

Fjodor2001

Diamond Member
Feb 6, 2010
3,710
182
106
Already out of stock.

Isn't this low supply issue unusual for Intel? They usually have the sales channels filled at launch.

Has anybody heard what is the reason for this? Making it seem like there is very high demand for Skylake, or Intel still having yield issues on 14 nm at high frequencies...? :hmm:
 
Last edited:

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
At lower TDPs Skylake is able to show such increments. c't tested the 6700K and at about the same TDP of Haswell it had 40W lower power consumption under load. Performance gains ~5% in CB, Sysmark '14, ~9% in Linpack (helped by DDR4-2133 vs. DDR3-1600), 13% in Luxmark and 18.5% in 3DMark Firestrike.

DDR4 2133 is a significant bottleneck for Skylake. You need to use faster memory for much better results. Both for bandwidth and latency.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106
Isn't this low supply issue unusual for Intel? They usually have the sales channels filled at launch.

Has anybody heard what is the reason for this? Making it seem like there is very high demand for Skylake, or Intel still having yield issues on 14 nm as high frequencies...? :hmm:

Its only an issue in the Americas.

Lowest sale area, last to serve. Shouldnt be a surprise to anyone. Its not the first time.
 
Mar 10, 2006
11,715
2,012
126
Isn't this low supply issue unusual for Intel? They usually have the sales channels filled at launch.

Very unusual, yes.

Has anybody heard what is the reason for this? Making it seem like there is very high demand for Skylake, or Intel still having yield issues on 14 nm at high frequencies...? :hmm:

Obviously there is "high demand" relative to supply, but supply seems to be very tight to begin with.

Given that Z170 (built on mature 22nm) is in high supply with no such "shortages," it might be the case that yields on these Skylake chips still aren't where they need to be.
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,710
182
106
Its only an issue in the Americas.

Lowest sale area, last to serve. Shouldnt be a surprise to anyone. Its not the first time.

It's in low supply in some countries in Europe too. E.g. Sweclockers reported on that some time ago.

Also, I don't recall it being usual practise for Intel to have low supplies of desktop CPUs in the Americas at launch of previous CPUs.
 

Freejack2

Diamond Member
Dec 31, 2000
7,751
8
81
I pulled the trigger today and ordered an i5-6600k, Asus Z170 Pro Gaming, Noctua NH-D14, Arctic Silver 5 with Arcticlean, and Mushkin DDR4-3000 ram 2x8gb. The ram is the [FONT=&quot]997205T with the Ridgeback-G2 heat spreader.
I couldn't find heights on the ram, but Noctua says it's compatible.

Good choices? Did I forget anything? (I'm upgrading from i5-2500k and Asus P8p67 pro so I already have drives, video card, ps, etc.)
[/FONT]
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
145
106

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
I pulled the trigger today and ordered an i5-6600k, Asus Z170 Pro Gaming, Noctua NH-D14, Arctic Silver 5 with Arcticlean, and Mushkin DDR4-3000 ram 2x8gb. The ram is the [FONT=&quot]997205T with the Ridgeback-G2 heat spreader.
I couldn't find heights on the ram, but Noctua says it's compatible.

Good choices? Did I forget anything? (I'm upgrading from i5-2500k and Asus P8p67 pro so I already have drives, video card, ps, etc.)
[/FONT]

Why didn't you wait for the 6700k?
 

Freejack2

Diamond Member
Dec 31, 2000
7,751
8
81
Why didn't you wait for the 6700k?
I use it for gaming. And while I see the 6700k has a slight advantage with hyperthreading, at overclocked speeds the difference is minimal. I just don't think the small difference is worth $100.
 

CakeMonster

Golden Member
Nov 22, 2012
1,384
482
136
I ended up buying DDR4-2666 C15 for what will probably be my Skylake build (still debating HW-E).

I figured that since 2666 was significantly cheaper than anything faster, I should not waste that money, or at least in stead put it towards some other kind of hardware. When I see you guys talking about higher speeds giving great results, is this even considering that 3000+ usually has vastly higher latency? From everything I've read earlier my understanding was that latency was equally important. Wouldn't it then defeat the purpose to get these faster (and very expensive) sets?