Intel Skylake / Kaby Lake

[Sweepr]

Interesting, Intel announced mobile Skylake Xeons:

http://blogs.intel.com/technology/2015/08/bringing-intel-xeon-to-notebook-pcs/

And if the leaked slides are right then it will be a quad-core GT4e part @ 35-45W TDP. Small and efficient powerhouse.

Granted, the results are looking better, but the inconsistency still bothers me, anand's results in particular. Especially since it appears you need very fast, expensive ram to see much benefit.

The difference is not absurd though. Comparing G.Skill 2x 8GB DDR4 kits at Newegg, an extra $15 buys you DDR4-2666 and $25 more DDR4-3000. From what we saw I think it's worth the extra cost if you're building a Skylake-S gaming PC.

Seems that AMD is finally being left behind hard in the benchmarks they were great. They got stuck on figthing Sandy... maybe Zen is the very last move.

Right now there's a huge gap between Intel and AMD in terms of performance per clock. PCLab included results for both Core i7 6700K and FX8350 operating at 4.7GHz in their review so it's easy to make an architecture comparison. According to their results, Skylake-S is:

34,3% faster @ Assassins Creed Unity
60,4% faster @ Arma III
60,3% faster @ Battlefield 4 Multiplayer
64% faster @ Counter Strike Global Offensive
32.5% faster @ Crysis 3
104% faster @ Far Cry 4
114% faster @ GTA V
85% faster @ The Witcher 3
60.6% fater @ Watch Dogs
87.4% faster @ Project Cars
133% faster @ Starcraft 2
139.1% faster @ Total War Attila

Overall: Skylake-S 81.2% faster than Vishera per clock.

 

[Sweepr]

Still nobody post here this review?
http://www.purepc.pl/procesory/test...56600k_skylake_mocne_cztery_rdzenie?page=0,33

Nice find. PurePC chose DDR4-2666 CL13 for their Skylake-S system.
Another round of Sandy Bridge vs Ivy Bridge vs Haswell vs Skylake results at equal clock (4.5GHz).

Here we go, Core i5 6600K vs Core i5 4690K operating at 4.5GHz, Skylake-S is:

15.9% faster @ Battlefield 4 SP
10% faster @ Crysis 3
21.3% faster @ Far Cry 4
16.5% faster @ Watch Dogs
16.5% faster @ Arma III
13.2% faster @ TES Skyrim
14.5% faster @ Starcraft 2
18.2% faster @ GTA V

Overall: Skylake-S 15,8% faster than Haswell per clock. Even better than PCLab results (below).

Here's a quick clock per clock (IPC) analysis, PCLab has Haswell and Skylake-S at equal clocks in their overclocking results.

Core i5 6600K vs Core i5 4690K at fixed 4.5GHz, Skylake-S is:
16.2% faster @ Battlefield 4 MP
10.9% faster @ Counter Strife Global Offensive
6% faster @ Crysis 3
22.7% faster @ Far Cry 4
13% faster @ GTA V
12.5% faster @ The Witcher 3
9.2% faster @ Watch Dogs
14.7% faster @ Project Cars
14.8% faster @ Starcraft 2
24.5% faster @ Total War Attila

Who said Skylake-S was slower for games again? You have some explaining to do AnandTech.

Edit: Overall gaming performance per clock (IPC) from 14 games/tests @ 1080p Ultra:

Sandy Bridge to Haswell: 13.5%
Haswell to Skylake: 12.8%
Skylake beating Broadwell with eDRAM here.

Right now there's a huge gap between Intel and AMD in terms of performance per clock. PCLab included results for both Core i7 6700K and FX8350 operating at 4.7GHz in their review so it's easy to make an architecture comparison. According to their results, Skylake-S is:

34,3% faster @ Assassins Creed Unity
60,4% faster @ Arma III
60,3% faster @ Battlefield 4 Multiplayer
64% faster @ Counter Strike Global Offensive
32.5% faster @ Crysis 3
104% faster @ Far Cry 4
114% faster @ GTA V
85% faster @ The Witcher 3
60.6% fater @ Watch Dogs
87.4% faster @ Project Cars
133% faster @ Starcraft 2
139.1% faster @ Total War Attila

Overall: Skylake-S 81.2% faster than Vishera per clock.

 

[shady28]

Nice find. PurePC chose DDR4-2666 CL13 for their Skylake-S system.
Another round of Sandy Bridge vs Ivy Bridge vs Haswell vs Skylake results at equal clock (4.5GHz).

Here we go, Core i5 6600K vs Core i5 4690K operating at 4.5GHz, Skylake-S is:

15.9% faster @ Battlefield 4 SP
10% faster @ Crysis 3
21.3% faster @ Far Cry 4
16.5% faster @ Watch Dogs
16.5% faster @ Arma III
13.2% faster @ TES Skyrim
14.5% faster @ Starcraft 2
18.2% faster @ GTA V

Overall: Skylake-S 15,8% faster than Haswell per clock. Even better than PCLab results (below).

What memory did they choose for their SB / Haswell systems?

That is not exactly an enlightening test if they used overclocked RAM on one box and just generic DDR3-1600 on the other (Skylake as I understand it is rated for DDR4-2133, just as Haswell is rated for DDR3-1600).

These graphs are on a 4770K using a GTX 780. The differences are likely to be greater on more modern 4790K / GTX 980 Ti setups (where they are far less likely to go GPU bound at any point).

I'm trying to find a more recent set of tests using a more powerful GPU, but for now consider the % increase from using DDR3-2400 vs DDR-1600 on Haswell :

http://www.corsair.com/en-us/blog/2014/march/haswellrealworld

 

[inf64]

What memory did they choose for their SB / Haswell systems?

That is not exactly an enlightening test if they used overclocked RAM on one box and just generic DDR3-1600 on the other (Skylake as I understand it is rated for DDR4-2133, just as Haswell is rated for DDR3-1600).

These graphs are on a 4770K using a GTX 780. The differences are likely to be greater on more modern 4790K / GTX 980 Ti setups (where they are far less likely to go GPU bound at any point).

I'm trying to find a more recent set of tests using a more powerful GPU, but for now consider the % increase from using DDR3-2400 vs DDR-1600 on Haswell :

http://www.corsair.com/en-us/blog/2014/march/haswellrealworld

Nice find, Haswell scales just as good as Skylake with faster DDR3 memory .

 

[ClockHound]

Wondered if I softened the phrasing too much to adequately describe my utter contempt for the pathetic excuse of a computer that laptops have been/are/will be. Need to work on being more direct with my opinions.

Warning: Enthusiast Rant starts now.

To minimize the limitations, being the antithesis of an enthusiast computer, costs multiple thousands - even the uber lappys still suffer from a terrible thermal envelope, pathetic screens and Playschool style keyboards. And none that exceed the broken format barrier. The only desktop replacement is a better, faster desktop.

The laptop is a betrayal of the Personal Computer, a once noble device that could be tailored to one's budget, needs and tastes - assembled and serviced with one's own hands from carefully selected components.

The laptop is the icon for bland look-alike computing appliances - trading function for portability - performance for form factor - excellence for mass marketing drivel - the laptop is a celebration of conformist mediocrity, all at a higher price. That sort of banal treachery can make a corporation rich.

/rant

Such is the cost of portability.

That's a heavy cost for something so light. ;-)

 

[Sweepr]

I can find plenty of tests where Haswell barely scales with faster DDR3 in games (not sponsored by Corsair).

PCLab's review favours Haswell according to AnandTech's speed/latency calculation...

Normally in our DRAM reviews I refer to the performance index, which has a similar effect in gauging general performance:

DDR3-1600 C11: 1600/11 = 145.5
DDR4-2133 C15: 2133/15 = 142.2

As you have faster memory, you get a bigger number, and if you reduce the CL, we get a bigger number also. Thus for comparing memory kits, if the difference > 10, then the kit with the biggest performance index tends to win out, though for similar kits the one with the highest frequency is preferred.

PCLab
DDR4-2666/16 = 166 (Skylake-S)
DDR3-1866/9 = 207 (Haswell)

...and Skylake still manages 12.8% better performance per clock overall in 14 games. Right now we have at least 4 reviews with 1080p gaming tests where Skylake is beating Haswell by >10% per clock (Hardware.fr, Eurogamer, PCLab and PurePC). No need to justify your purchase inf64, Haswell is still an awesome gaming chip (and will be for years to come), it's just that there's something better out right now.

 

[shady28]

^^^^^
Yeah those above benchmarks were run on old video (58xx AMD) cards at resolutions where they were GPU limited not CPU limited. If you're going to examine an edge case of CPU-limited games, stick with that edge case. You need to find benchmarks where they were clearly CPU limited with a GPU that has plenty of gas left - like running 1080p games on a 980 Ti.

Here's another couple of examples, the first one is clearly not CPU limited but rather GPU limited like those charts in the earlier post.

The 2nd chart shows SLI where it is no longer GPU limited, but mostly CPU limited. That is a 30% increase in minimum and max FPS on Haswell going from 1333 to 2666.

 

[Sweepr]

DDR4 speeds start at 2133MHz. The gains using faster than DDR3-2133 memory with Haswell are less impressive.

And once again, Skylake doesn't need to be at an advantage (better memory speed/latency) to beat Haswell by >10% in CPU limited gaming as shown above.

 

[shady28]

DDR4 speeds start at 2133MHz. The gains using faster than DDR3-2133 memory with Haswell are less impressive.

Then why do most of the charts from your post mostly compare DDR4-1600 to DDR4-3000?

Skylake is rated for DDR4-1600 Low Power or DDR4-2133.

Anything else is overclocking the memory controller. That's fine and all, but if you're doing a comparison to another CPU then do apples to apples and overclock them both.

And once again, Skylake doesn't need to be at an advantage (better memory speed/latency) to beat Haswell by >10% in CPU limited gaming as shown above.

 

[Sweepr]

That's fine and all, but if you're doing a comparison to another CPU then do apples to apples and overclock them both.

We won't know for sure until someone does what I proposed earlier in the thread, multiple tests comparing Haswell vs Skylake using different DDR3/DDR4 clocks/latencies. PCLab is my review of choice for now, they tested 14 CPU intensive games at 1080p and I found the memory selection fair to both platforms (despite a possible Haswell advantage if we take AnandTech's word on the matter).

 

[inf64]

I have no doubt that Skylake is a better chip but the point is the increase (once you tweak both with faster memory and OC them) is really not something anyone could notice. Given how prices will go down and motherboards will become also more reasonably priced, it's more logical to buy Skylake than Haswell. The disappointing part is that we are really getting ever smaller performance increases with each new generation of chips, the low hanging fruit has been picked up...

 

[podspi]

What I don't get is...if Skylake scales so well with memory speeds, why did Intel validate it for such (relatively) slow memory?

For the non-k skus, aren't people going to be SOL, or can non-k skus still oc the men controller?

 

[shady28]

We won't know for sure until someone does what I proposed earlier in the thread, multiple tests comparing Haswell vs Skylake using different DDR3/DDR4 clocks/latencies. PCLab is my review of choice for now, they tested 14 CPU intensive games at 1080p and I found the memory selection fair to both platforms (despite a possible Haswell advantage if we take AnandTech's word on the matter).

That we can agree on. I don't dispute that Skylake might have a memory scaling advantage vs Haswell, IB, and SB, but comparing OC memory Skylake to non-OC memory SB/IB/HW on CPU limited game setups doesn't really say anything - just look at the boost haswell got in BF4 when it was CPU limited on a 780 Ti SLI rig.

If someone is running games on a 980 Ti at 1080p they can probably get a 0-20% FPS boost by overclocking DDR3 on those systems, just as Skylake can.

Even C2D could get more FPS with the settings low by using OC memory - this is a 7% increase in avg FPS and 20% in minimum on an E6550 :

 

[LTC8K6]

Well, they are overclock friendly chips and chipsets. They are meant to be overclocked.

Has anyone compared the 4790K and the 6700K with the same decent overclock and the fastest ram each will handle?

We know intel specifies low end ram for both, but both will run with much faster ram.

 

[Sweepr]

Well, they are overclock friendly chips and chipsets. They are meant to be overclocked.

Has anyone compared the 4790K and the 6700K with the same decent overclock and the fastest ram each will handle?

We know intel specifies low end ram for both, but both will run with much faster ram.

PCLab tested Core i7 6700K and Core i7 4790K at 4.7GHz, you can check out the results here.
Both chips could handle faster RAM but at least they didn't limit them to 'highest officially supported speeds' like AnandTech and some other reviews did (PCLab used DDR4-2666 CL16 for Skylake and DDR3-1866 CL9 for Haswell).

 

[Shivansps]

What I don't get is...if Skylake scales so well with memory speeds, why did Intel validate it for such (relatively) slow memory?

For the non-k skus, aren't people going to be SOL, or can non-k skus still oc the men controller?

Well, yes, but if you can BCLK overclock them, then is not a real problem as you can increase DDR4 speeds.

 

[PaulIntellini]

Skylake has higher bandwidth for L2 and L3 cache.
So it may be able to use more DRAM bandwidth to prefetch to these caches.

how is memory scaling for i5 6600K ? This part has only 6MB L3 cache.
Maybe it's time for Intel to increase the L3 cache size to 10 MB.

 

[coercitiv]

I can find plenty of tests where Haswell barely scales with faster DDR3 in games (not sponsored by Corsair).

It may be a good idea for you to take a step back and look again at available data. At this point you are effectively setting yourself up for confirmation bias, picking up results from tests that cannot be objectively compared. Earlier in this very thread Keysplayr posted a link to a guru3D review which should have at least raised a big question mark regarding memory scaling:

We can look at the Bioshock 720p framerates and notice some nice scaling, but we can also look at the Tomb Raider results and see no scaling whatsoever. We have a healthy jump in memory performance (2133->3200) and the FPS meter doesn't even budge. Could it be this situation is similar to the one in the Anandtech's Haswell DDR3 scaling review?

Haswell with 780Ti does 133FPS in Tombraider @720p, meanwhile in the memory scaling review it does 47FPS on single dGPU and 90FPS in the Crossfire test. Now imagine if we ran the test again on Nvidia's or AMD's latest.

 

[Deders]

That we can agree on. I don't dispute that Skylake might have a memory scaling advantage vs Haswell, IB, and SB, but comparing OC memory Skylake to non-OC memory SB/IB/HW on CPU limited game setups doesn't really say anything - just look at the boost haswell got in BF4 when it was CPU limited on a 780 Ti SLI rig.

If someone is running games on a 980 Ti at 1080p they can probably get a 0-20% FPS boost by overclocking DDR3 on those systems, just as Skylake can.

Even C2D could get more FPS with the settings low by using OC memory - this is a 7% increase in avg FPS and 20% in minimum on an E6550 :

What you have to bear in mind is that Core2 architecture was limited by its available memory bandwidth (about 8GB/s). Even the AMDx64 range had twice the bandwidth due to the integrated memory controller.

Intel has done the same thing Ever since Nehalem and the bandwidth issue was not only resolved but it increased with every generation to the point that increasing clock speeds would net a very small difference in games with Haswell as Anandtech showed here: http://www.anandtech.com/show/7364/memory-scaling-on-haswell/7

Skylake on the other hand seems to show much bigger differences.

I do agree though the tests should be re-done with modern games and a 980ti.

 

[Dufus]

Well, they are overclock friendly chips and chipsets. They are meant to be overclocked.

Has anyone compared the 4790K and the 6700K with the same decent overclock and the fastest ram each will handle?

We know intel specifies low end ram for both, but both will run with much faster ram.

HWBot is a good place to look.

Cinebench 11.5
i7-4790K     CPU 6.21GHz, DDR3 2850MT/s CL9,  13.73pts   
i7-6700K     CPU 6.16GHz, DDR4 3419MT/s CL14, 15.07pts

XTU
i7-4790K     CPU 5.84GHz, DDR3 2817MT/s CL5,  1535marks    
i7-6700K(ES) CPU 5.30GHz, DDR4 3738MT/s CL15, 1800marks

XTU likes that DDR4.

 

[Dresdenboy]

Skylake has higher bandwidth for L2 and L3 cache.
So it may be able to use more DRAM bandwidth to prefetch to these caches.

You mentioned an important part of DRAM performance: Prefetcher (at different levels). And there are even more influencers like buffer sizes, predictors (e.g. to open mem pages), etc.

So if we only look at bandwidth, we might miss something, as depending on the benchmark all this tech might cause better and in corner cases less than expected increases vs. older gen CPUs.

 

[steve wilson]

How much faster? G.Skill has 3600Mhz kits, but they are outrageously expensive at $400 for 8GB (2x4GB)

I don't really know to be honest. But the chip seems to work better with faster RAM. I was just wondering if it's worth upgrading soon, or waiting 3 or 4 months for faster/cheaper RAM? I'm coming from an i5 2500k at 4.3ghz.

On a side note does anyone know if the NH-U14S Noctua CPU cooler will fit on the z170 board?

 

[AtenRa]

What memory did they choose for their SB / Haswell systems?

That is not exactly an enlightening test if they used overclocked RAM on one box and just generic DDR3-1600 on the other (Skylake as I understand it is rated for DDR4-2133, just as Haswell is rated for DDR3-1600).

They used 1600MHz at 8-8-8-24 for DDR-3 and 2666MHz 13-14-14-39 for DDR-4.

 

[phillyman36]

I don't really know to be honest. But the chip seems to work better with faster RAM. I was just wondering if it's worth upgrading soon, or waiting 3 or 4 months for faster/cheaper RAM? I'm coming from an i5 2500k at 4.3ghz.

On a side note does anyone know if the NH-U14S Noctua CPU cooler will fit on the z170 board?

The mounting holes are the same as 1150. Yes its fits. Im going to use a NH-U14s with my Asus z170 Deluxe.(just waiting to find a cpu)

http://www.noctua.at/main.php?show=compatibility_gen&products_id=64&lng=en#LGA1151_Asus

 

[ClockHound]

On a side note does anyone know if the NH-U14S Noctua CPU cooler will fit on the z170 board?

It is. Direct from the horse's fins:

"Yes. Due to the heatsink mounting mechanism being identical on LGA1150, LGA1151, LGA1155 and LGA1156, the cooler also supports LGA1151."

http://www.noctua.at/main.php?show=faqs&step=2&products_id=64&lng=en#30