Intel Skylake / Kaby Lake

[IntelUser2000]

I suspect something similar:

i7 = 6C/12T Unlocked
i5 = 4C/8T Unlocked
i3 = 4C Locked
Pentium = 2C/4T
Celeron = removed.

We had the linkedin leak that were talking about an "Cannonlake SoC" with 4/6/8 cores. It makes sense with CFL increasing cores to 6 on the top SKU. The die size would be too small with 4 cores so to justify ASPs they would increase it to 8. Going to 8 cores would have minimal benefits and would have hard time selling it to consumers but that's not a technical issue.

 

[LTC8K6]

Hexus had basically the same temps for SL at 4.6 and KL at 4.85, though.

We decided to overclock both chips using the same methodology so as to keep a level playing field. We raised the CPU voltage a fraction, to 1.30V, which we consider a safe limit for long-term running. Both CPUs ran at below 75°C using a single Noctua fan.

4,850MHz is a decent frequency for a quad-core processor. A company, who shall remain nameless, has tested 30 Core i7-7700K processors and reckoned that 'good' chips can do 5GHz with 1.29V whilst bad chips need 1.35V for the same speed. It appears that we have a bad chip because the processor was only semi-stable at 5GHz at 1.35V. We're genuinely interested to see how next month's retail chips perform with our setup.

 

[RichUK]

Golden samples might have been sent, considering all the negative publicity Kabylake and the 14nm+ process has attracted so far. This coupled with Zen's imminent release could well have opened the golden coffers

Even if they did have a golden sample that could achieve P95 stability at 5.1Ghz; 5.1Ghz + 1.44v +H105 doesn't = ~90 degrees.

Their credibility is questionable in my opinion, and misleading.

 

[RichUK]

Sweepr, as you appear to enjoy harvesting info from the webs, you should pull a table together with '7700k overclocks / volts / temps / info source' and any other meaningful metric so we can have a consolidated view.

You'll be doing a good public service that will be very self rewarding.

 

[RichUK]

Hexus had basically the same temps for SL at 4.6 and KL at 4.85, though.

Did you see what version of Prime95 they are using (i didn't), as v27.9 is netting me far better results?

Short test, sustaining a 5Ghz overclock at 1.34v:

 

[jpiniero]

Is that using AVX2 though.

 

[RichUK]

Is that using AVX2 though.

No (is my understanding), it fails when I use v28.7 which i understand uses AVX2 - it fails and generates insane temps, that's why i'm interested in what version these review sites used so i can compare.

IBT and OCCT with AVX linkpack ticked doesn't fail at 5Ghz for me, whereas Prime95 v28.7 almost instantly fails.

 

[jpiniero]

I'd expect AVX2 to use more power than AVX. After all, it's doing 2X the work.

 

[Ajay]

These results with early production samples KabyLake are making me a bit excited about the possibilities of KBL-X come 2nd half of 2017. 112W TDP should produce a very nice CPU (if I don't have the bucks then for SKL-X).

 

[RichUK]

I'd expect AVX2 to use more power than AVX. After all, it's doing 2X the work.

Yeah, it was ruining my VRMs, that's for sure!

The rationale I am developing is if I don't use programs that utilise AVX2, then surely excluding AVX2 from stress testing is valid.

 

[RichUK]

Prime95 v27.9, according to this site, is a valid stress test: http://www.overclock.net/t/1570313/skylake-overclocking-guide-with-statistics

 

[UncleCrusty]

Yeah, it was ruining my VRMs, that's for sure!

The rationale I am developing is if I don't use programs that utilise AVX2, then surely excluding AVX2 from stress testing is valid.

Definitely seems as if the AVX2 dual FMACS have worse stability characteristics. Whether you'll be fully stable in code where AVX2 is never utilized, it is hard to say. I don't have a Haswell+ CPU so I've never tested this, or seen it tested for that matter. I'm not sure how well balanced Intel's pipelines are, but it seems like the AVX2 hardware probably contains the critical path (the stock AVX2 "voltage bump" I've heard about would support this).

 

[LTC8K6]

I didn't see anything about P95, but Hexus had ram problems.

We have an early sample of a Core i7-7700K. Though it is marked with the codes you would expect on a retail chip, we found that it had a problem in running memory faster than 2,133MHz. The same set of G.Skill memory operated at 3,000MHz on the Core i7-6700K just fine. We've yet to put the Kaby Lake chip in a Z270 board, mind.

 

[Nothingness]

I'd expect AVX2 to use more power than AVX. After all, it's doing 2X the work.

AVX already is 256-bit for FP which is what prime95 uses.

 

[lolfail9001]

AVX already is 256-bit for FP which is what prime95 uses.

The point is FMA appears in AVX2... and well.

 

[Sweepr]

Skylake-X and Kaby Lake-X may debut in August 2017

Intel is planning to unveil its next-generation top-end Skylake-X and Kaby Lake-X series CPUs in the third quarter of 2017, succeeding the existing Broadwell-E series with a price range between NT$15,000-57,000 (US$468-1,780). Intel may debut the two series at Gamescom 2017 in Germany in August.

At the same time, Intel is ready to announce the full series of its Kaby Lake processors at CES 2017, starting January 4, 2017. Intel already began shipping many of its Kaby Lake-S series processors including Core i7-7700K in December.

The new Skylake-X and Kaby Lake-X processors will feature a new LGA 2066 socket and support DDR4 memory. The CPUs will pair with Intel's new X299 chipsets.

At CES 2017, Intel is expected to unveil processors including Core i7-7700K/7700, Core i5-7600K/7600/7500, Core i3-7320/7300/7100, Pentium G4620/4500/4560 and Celeron G3950/3930 as well as its low-power models including the Core i7-7700T. Intel's Z270 and H270 chipsets will also make their debut at the show.

www.digitimes.com/news/a20161208PD201.html

 

[witeken]

Intel might not be licensing AMD GPU patents after all since Intel has indefinite access to the Nvidia portfolio that predates March 31 '17.

http://seekingalpha.com/article/402...reement-nvidia-will-continue-well-beyond-2017

 

[mikk]

http://www.hardocp.com/article/2016/12/09/intel_kaby_lake_core_i77700k_ipc_review/5#.WEvpflxsuUl

At the same clock @ 4.5 Ghz, Kabylake consumes less power.

 

[witeken]

http://www.hardocp.com/article/2016/12/09/intel_kaby_lake_core_i77700k_ipc_review/5#.WEvpflxsuUl

At the same clock @ 4.5 Ghz, Kabylake consumes less power.

If you subtract out the idle power, Kaby Lake's using 1.09 - 1.25 times less power. That 1.25x seems a little optimistic, but in any case a good result for KBL for just an improved process.

 

[dooon]

Kaby Lake Retail Box
"FOR A GREAT VR EXPERIENCE"

 

[witeken]

Those dolls, though.

 

[Ajay]

Those dolls, though.

LOZ!

 

[Ajay]

If you subtract out the idle power, Kaby Lake's using 1.09 - 1.25 times less power. That 1.25x seems a little optimistic, but in any case a good result for KBL for just an improved process.

Or, roughly 10-20% less power than Skylake at ISO frequency. A very good result for 'small' process tweak!

Sorry, 'times less power' bugs me

 

[witeken]

A very good result for 'small' process tweak!

At these frequencies I think this is comparable to the regular 14nm shrink from 22nm.

 

[Arachnotronic]

Kaby Lake Retail Box
"FOR A GREAT VR EXPERIENCE"

Haha, Intel is jumping on the "VR Ready" bandwagon like NVIDIA and AMD have.

Anyway, if they already have chips in retail boxes floating around here in December, then these chips PRQ'd a long time ago.