Intel Skylake / Kaby Lake

[Tovarisc]

DDR4-2133 vs DDR4-2666 from TweakTown Review

www.tweaktown.com/reviews/7274/intel-skylake-core-i7-6700k-cpu-z170-chipset-gt530-review/index9.html

We're not even talking about DDR4-3000+
Skylake-S loves fast DDR4 kits and faster memory helps Core i7 6700K beat Core i7 4790K despite lower Turbo clocks.

Can't take AnandTech's review seriously after seeing this.

Seeing more these tests where different DDR4 frequencies get compared makes me think that I should invest into 2666MHz kit for that performance boost.

 

[jpiniero]

So my i5-2500k is still good and upgrading wouldn't be worth it ?

Yeah, esp if it's overclocked. Unless you are somehow unhappy with the frame rates.

 

[Sweepr]

So the early leaks claiming they got Skylake to 5.2 GHz on air is like fake, or an extreme golden chip? None of the reviews got it above 5 Ghz, and only one above 4.8 Ghz.

At least a few reviewers mentioned they could get it to 5GHz+ with watercooling, others will update the results soon (some reviews were rushed), BIOS also needs some tuning.
Now considering most reviewers got 4.7-4.8 GHz with little effort (after Broadwell-K's 4.2-4.3 GHz) I would say those who made made a fuss about 14nm problems and Skylake-S not being able to match Devil's Canyon overclocking potential should be a bit embarassed.

Seeing more these tests where different DDR4 frequencies get compared makes me think that I should invest into 2666MHz kit for that performance boost.

That makes sense, DDR4-2666 is the minimum I'd pick for Skylake-S.

 

[Midwayman]

I disagree. At stock a Core i7 6700K is 38% faster than Core i7 2600K in CPU-bound games according to Hardware.fr, which means that it will lose even when OCed to 5GHz.
Sure, 5GHz Sandy Bridge is still an awesome chip but 4.6-4.8GHz Skylake-S will be noticeably faster when you are CPU limited.

How often do you play games at 640x480 or 1024x768 these days? While interesting in an academic sense, real world application matters a whole lot more. If you really do, I guess have fun playing at 400fps or whatever.

 

[Sweepr]

How often do you play games at 640x480 or 1024x768 these days? While interesting in an academic sense, real world application matters a whole lot more. If you really do, I guess have fun playing at 400fps or whatever.

Hardware.fr tested at 1080p, difference is they actually test some CPU demanding games, not prescripted GPU-limited benchmarks.

Also from Tech Report:

Doesn't mean you should ditch your current SB/IB/Haswell build but in some titles you can put the extra CPU grunt to use.
For 120 Hz gaming every little bit helps too.

 

[Ajay]

We don't actually have proper memory scaling data for Skylake yet, but my guess is we won't see a much different picture than what Haswell-E painted.

I hope the scaling will be a bit better as SKL is only dual channel and has less L3$.

 

[coercitiv]

That makes sense, DDR4-2666 is the minimum I'd pick for Skylake-S.

But is it the bandwidth or just the improved latency that's really needed?

 

[crashtech]

It's encouraging to see the overclocking results come in fairly well from so many different reviewers. Looks like a reasonable average expectation is 4.7GHz, which, if we can recall, is somewhat better than the original Haswell.

 

[Burpo]

"Intel's Skylake is the king of quads, but not the upgrade you waited for"

http://www.digitaltrends.com/processor-reviews/intel-core-i7-6700k-review/#ixzz3hyuBZogd


Somewhere I saw a screenshot running at 200bclk & 23X.. Can't find it now.. Anybody else see it?

 

[Carfax83]

So when is Skylake-E supposed to be available?

 

[escrow4]

HA! Retailing here for $559 i7 6700K, and overnight the 5820K went up slightly to $559. $559 and it doesn't even have the top end iGPU. What a waste.

 

[Arachnotronic]

So when is Skylake-E supposed to be available?

1H 2017 probably.

 

[Absolute0]

So the early leaks claiming they got Skylake to 5.2 GHz on air is like fake, or an extreme golden chip, or a brief suicide run? None of the reviews got it above 5 Ghz, and only one above 4.8 Ghz.

Suicide run. They dialed the clock back 200 Mhz for a Pi calculation.

Overclocks look relatively good, after broadwell I was worried but on water 4.8 Ghz seems pretty likely. Better than HW and IB. My Ivy didn't even like to go about 4.4 Ghz with a Noctua D14 on it. Was very sad. Chip just got smoking hot and peetered out.

Will be excited to see new tests with better memory and see what happens as DDR4 gets faster.

 

[inf64]

It seems my prediction was pretty much accurate in every regard, I only overstated the IPC jump (I expected 8-10% and AT measured 5.7% for Haswell to Skylake). I suppose Skylake is an "OK" tock but when compared to D.C. Haswell it's pretty much insignificant perf. increase. . Those who shelled out $$$ for BDW-C parts virtually get no speedup apart from slightly higher OC room and essentially bought next-gen product (even though that same BDW-C is also marginally faster than Haswell except in a few very specific cases).

 

[Carfax83]

1H 2017 probably.

Are you serious? It seems the time lag between the mainstream parts and the enthusiast parts are getting longer and longer..

I want Skylake-E to be my next platform upgrade, but that's a long way off considering I'm on the aging x79 platform.

What about PCI-E 4.0 support?

 

[Sweepr]

It seems my prediction was pretty much accurate in every regard, I only overstated the IPC jump (I expected 8-10% and AT measured 5.7% for Haswell to Skylake). I suppose Skylake is an "OK" tock but when compared to D.C. Haswell it's pretty much insignificant perf. increase. . Those who shelled out $$$ for BDW-C parts virtually get no speedup apart from slightly higher OC room and essentially bought next-gen product (even though that same BDW-C is also marginally faster than Haswell except in a few very specific cases).

Hardware.fr results were a bit better though, in line with your prediction. At fixed 4GHz Skylake-S was:
- 6.7% faster in applications, though Haswell results look a little inflated next to Sandy Bridge (15.7% faster on average).
- 12.1% faster in games. Considerably faster than Haswell in some CPU limited titles, and a hair slower than Broadwell-K per clock due to its huge eDRAM (on average).

Meanwhile Core i5 6600K was 8% faster than Core i5 4690K in applications and 10% faster in games (same 3.5-3.9GHz).

Will be fun to see how Skylake performs with eDRAM next year.

 

[moonbogg]

Are you serious? It seems the time lag between the mainstream parts and the enthusiast parts are getting longer and longer..

I want Skylake-E to be my next platform upgrade, but that's a long way off considering I'm on the aging x79 platform.

What about PCI-E 4.0 support?

Yeah, me too. Its the only way forward really and seems we get to wait a while. By that time I think big Pascal will have been out a while.

 

[Arachnotronic]

Are you serious? It seems the time lag between the mainstream parts and the enthusiast parts are getting longer and longer..

I want Skylake-E to be my next platform upgrade, but that's a long way off considering I'm on the aging x79 platform.

What about PCI-E 4.0 support?

Yeah, it sucks

I don't think Skylake-E will bring PCIe 4.0 support, either. We will see more PCIe 3.0 lanes, though.

 

[Fjodor2001]

At least a few reviewers mentioned they could get it to 5GHz+ with watercooling, others will update the results soon (some reviews were rushed), BIOS also needs some tuning.

5+ Ghz with water cooling, yes. But previously it was mentioned 5.2 GHz on air, which is something completely different. So that was likely fake then.

Interesting that it is now confirmed that Intel 14 nm overclocks worse than 22 nm (and even 32 nm), as expected by some. Others that refused to realize that are now put to shame.

 

[Burpo]

:X

 

[inf64]

Hardware.fr results were a bit better though, in line with your prediction. At fixed 4GHz Skylake-S was:
- 6.7% faster in applications, though Haswell results look a little inflated next to Sandy Bridge (15.7% faster on average).
- 12.1% faster in games. Considerably faster than Haswell in some CPU limited titles, and a hair slower than Broadwell-K per clock due to its huge eDRAM (on average).

Meanwhile Core i5 6600K was 8% faster than Core i5 4690K in applications and 10% faster in games (same 3.5-3.9GHz).

Will be fun to see how Skylake performs with eDRAM next year.

I like that hardware.fr review more than AT one. It has more details and seems like they put more time into it.

Anyway, Skylake will eventually push out all BDW and HSW parts from the market as it should. It feels more like a tick than a tock, but since Haswell did set a bar very high it is not that easy to chase IPC as most low hanging fruit has already been picked up. eDRAM will help a bit next year but this would only show in select few workloads, just like in the case of BDW-C parts.

 

[IntelUser2000]

5+ Ghz with water cooling, yes. But previously it was mentioned 5.2 GHz on air, which is something completely different. So that was likely fake then.

They stated that 5.2GHz on air wasn't stable and that it needed to be 5GHz to run benchmarks.

Based on the average of the review its even lower than that. Its similar to DC where it claimed 5GHz but ended up lower.

Interesting that it is now confirmed that Intel 14 nm overclocks worse than 22 nm (and even 32 nm), as expected by some. Others that refused to realize that are now put to shame.

Only the die-hard Intel people didn't think so.

To reviewers: No, its not worth it to upgrade from a 2600K. The 2600K pricing was $317, this is $350. The price has increased. And you have to change the platform. Its a $600 upgrade for CPU+Memory+Motherboard. How can you justify that for a 30% improvement? Give me additional 30%(multiplicative, not additive) and maybe I'll think about it. Most people would have done it if 6700K was 15% faster than 4790K, not 5-7%.

Another thing: Someone keeps claiming that Sandy Bridge had similar gains in IPC to Haswell. Numerous reviews are pointing out Sandy Bridge was THE chip in the last few years. That Intel graph about "IPC gains" are quite inaccurate.

 

[Ajay]

5+ Ghz with water cooling, yes. But previously it was mentioned 5.2 GHz on air, which is something completely different. So that was likely fake then.

Interesting that it is now confirmed that Intel 14 nm overclocks worse than 22 nm (and even 32 nm), as expected by some. Others that refused to realize that are now put to shame.

Wait, 4.6-4.8 GHz is worse than Haswell? Really?

 

[Sweepr]

5+ Ghz with water cooling, yes. But previously it was mentioned 5.2 GHz on air, which is something completely different. So that was likely fake then.

HKEPC's 5.2GHz shot was not 24/7 stable, just stable enough to get a CPU-Z Validation. In case you missed, TweakTown also managed to get Skylake-S to 5.2GHz for CPU-Z Validation.

Interesting that it is now confirmed that Intel 14 nm overclocks worse than 22 nm (and even 32 nm), as expected by some.

4.7-4.8GHz 24/7 stable is better than Haswell did at launch, 14nm doom appologists are indeed in the halls of shame now.

 

[IntelUser2000]

4.7-4.8GHz 24/7 stable is better than Haswell did at launch, 14nm doom appologists are indeed in the halls of shame now.

4 years and we need an entire platform change to get 30%.

I really think you being "right" pales in comparison to what the real issue is, that we've got tiny increases in the last few generations.

This also confirms that average IPC gain per year is 5-7%. After Sandy Bridge the gains practically collapsed.