Intel Skylake / Kaby Lake

[Nothingness]

Sure, but a lot of these little enhancements/optimizations add up to something pretty good.

Intel's "big core" teams do a great job and they've delivered something solid with Skylake.

Agreed. But the price is high for the improvements we see. I'm nonetheless impressed this 10 yo uarch is still alive and kicking . When you look at Atom...

 

[Sweepr]

Skylake-H and Skylake-U leaked slides:

With Skylake-H, Intel is going to introduce the first fully DDR4 compatible mainstream notebook platform in the consumer market. There are going to be two series for the Skylake-H lineup, the corporate or main stream 45W and 35W chips alongside the low power 25W chips.

- There will be 25W TDP versions of 4C/8T Skylake-H. Previously Intel's quad-cores started at 37W TDP (most were 47W TDP). Respectable clocks too, 2.0GHz base and 2.8GHz Turbo.
- Xeon E3-1500M v5 will get the mythical GT4e iGPU at 45W TDP.
- Intel will finally release 4C/4T Core i5 mobile chips. 4C/4T Skylake-H will replace 2C/4T Haswell-H chips.

www.xfastest.com/thread-160894-1-1.html

If you missed last page:

 

[Fjodor2001]

You're right, the architecture has been improved across the board from front-end to execution units.

So many architectural improvements, so little actual performance improvement on desktop.

 

[Hitman928]

Sorry, don't have time to search through the whole thread, is there an expected (or better confirmed) release date for Skylake-E?

 

[Sweepr]

Intel® Architecture, Code Name Skylake Deep Dive: A New Architecture to Manage Power Performance and Energy Efficiency

New presentation: https://hubb.blob.core.windows.net/...XCBi+mxwYQnTQ=&se=2015-08-19T01:40:05Z&sp=rwd

Edit:

Skylake front-end can pull 6 instructions/clock out of the uop cache. Jeez.

So Skylake power gates its AVX2 hardware when it's not in use.

Skylake ring fabric is 2X bandwidth at the same power level

Skylake with eDRAM cache get a major rework/upgrade of eDRAM cache architecture. Can even be used to transfer data between IO devices.

 

[Qwertilot]

So many architectural improvements, so little actual performance improvement on desktop.

Eh, for one specific subset of desktop

Anything power limited looks to be boosting a useful amount, which I guess includes the eventual E stuff.

 

[Nothingness]

Intel® Architecture, Code Name Skylake Deep Dive: A New Architecture to Manage Power Performance and Energy Efficiency

New presentation: https://hubb.blob.core.windows.net/...XCBi+mxwYQnTQ=&se=2015-08-19T01:40:05Z&sp=rwd

Edit:

The AVX2 power gating bit is funny, ARM has been power gating their FP/SIMD units for years :biggrin:

 

[Deders]

What bios version are you running? There are some people saying that some of then higher ram when you set the xmp it wont work. It defaults to 2133(have to turn off xmp.)

As far as I know there is only 1 bios version so far.

i have made a bootable win 7 USB..

What else do i need to do?

You'll need to make it with Asus easy installer from the DVD that comes with the board. It's like the Microsoft version but it puts the right driver files on there. Might be downloadable from Asus's website.

Failing that I think you can just use your normal usb 7 with the DVD in the drive and it will pick up the drivers automatically. Although this takes longer because for some reason windows searches the entire 5.77GB of the DVD at the end of the install.

 

[Enigmoid]

Intel® Architecture, Code Name Skylake Deep Dive: A New Architecture to Manage Power Performance and Energy Efficiency

New presentation: https://hubb.blob.core.windows.net/...XCBi+mxwYQnTQ=&se=2015-08-19T01:40:05Z&sp=rwd

Edit:

Is Skylake's PCH on package? The diagram seems to show this.

 

[Fjodor2001]

Eh, for one specific subset of desktop

Anything power limited looks to be boosting a useful amount, which I guess includes the eventual E stuff.

E-stuff is unknown, I think? Otherwise I missed it.

Also, top end / enthusiast desktop does not see much performance improvement. That's why I'm skipping it. Will stay on IB for that.

Fortunately I'm looking for a new Ultrabook (U/Y series). So since that has been somewhat prioritized on Skaylake, I'm quite satisfied with that. However I would like to see some actual data on it though before cheering too loud.

 

[jpiniero]

25W Quads? Nice.

Is Skylake's PCH on package? The diagram seems to show this.

Yeah.

 

[firewolfsm]

Moving PCH on package with the same 15W TDP might actually cause a hit to performance.

It said they're "considering" releasing 2C/GT3e processors. That means it won't happen til January or February. Does anyone have a guess on when skylake laptops should be coming out? Asus should be first on the market.

 

[Fjodor2001]

Does anyone have a guess on when skylake laptops should be coming out? Asus should be first on the market.

Last I heard, actual availability in stores is November 2015:

http://www.kitguru.net/components/c...al-shortage-of-intel-skylake-microprocessors/

actual notebooks powered by “Skylake” processors will only get to the market in November.

 

[Enigmoid]

Moving PCH on package with the same 15W TDP might actually cause a hit to performance.

It said they're "considering" releasing 2C/GT3e processors. That means it won't happen til January or February. Does anyone have a guess on when skylake laptops should be coming out? Asus should be first on the market.

Haswell and Broadwell U (15 and 28W) already have the PCH on package.

This is looking nice for mobile. Finally getting 4C/4T i5s. Probably because since haswell intel has sold virtually no 35W dual core chips (ie the 4200m) with broadwell completely eliminating them.

 

[Sweepr]

Haswell and Broadwell U (15 and 28W) already have the PCH on package.

This is looking nice for mobile. Finally getting 4C/4T i5s. Probably because since haswell intel has sold virtually no 35W dual core chips (ie the 4200m) with broadwell completely eliminating them.

To me 25W mobile quad-cores are one of their biggest achievements this generation.
H series is stronger than ever.

 

[Fjodor2001]

To me 25W mobile quad-cores are one of their biggest achievements this generation.
H series is stronger than ever.

Have they mentioned base and turbo clocks? And what about throttling?

 

[Sweepr]

Have they mentioned base and turbo clocks? And what about throttling?

Leaked slides above. 2.0-2.8GHz for the Core i7 and Xeon.

 

[Fjodor2001]

Leaked slides above. 2.0-2.8GHz for the Core i7 and Xeon.

Xeon? But about the H SKUs? Same?

Anyway, U series are 2.6 to 3.4 GHz at 15 W TDP and 2C/4T. E.g 6600U. Much better ST performance and almost half TDP.

 

[Sweepr]

Xeon? But about the H SKUs?

Same clocks. This thing will demolish Skylake-U 28W and Carrizo in MT CPU tasks.

 

[ShintaiDK]

To me 25W mobile quad-cores are one of their biggest achievements this generation.
H series is stronger than ever.

Ye, 25W puts it into something the masses want. Unlike 35 and 45W.

 

[Fjodor2001]

Ye, 25W puts it into something the masses want. Unlike 35 and 45W.

And moar cores than the higher clocked 15 W SKUs?

 

[Enigmoid]

And moar cores than the higher clocked 15 W SKUs?

Looks like first gen i7m chips a la i7-720qm. Lower turbo than the i5 duals with a moar cores approach.

 

[firewolfsm]

Ye, 25W puts it into something the masses want. Unlike 35 and 45W.

My X220 has a 35W dual core (i7-26XXm) and is an excellent laptop with long battery life. I was waiting for a similar laptop with a 35W quad core but my impression is that the chipset differentiation is what stopped that from happening.

 

[myocardia]

You made this absurd accusation so the burden of proof is on you.

Again, are you seriously admitting that you don't have the slightest idea how someone could intentionally cheat, when comparing two overclockable CPUs to each other, besides RAM speed, to make the performance difference between the two CPUs be higher?

And yes, I called that review a review 'authored' by a shill. Anyone who read it could see what it happened to be. Anyone who knows the definition of the word shill would see that the review to which you continually referred, penned by Richard Leadbetter, and posted @ eurogamer.net, fits the definition. That, like everything else posted by every single member of these forums, is an opinion. If you don't share it, that's fine. I honestly don't mind that you don't.

What my, mikk's and IntelUser2000's benchmarks prove is:

Kid, IntelUser2000's 'benchmarks' (which are actually good estimations, not benchmarks of any type) prove that you don't have the slightest clue how DRAM functions, since every single part of it agrees with what I've been saying since my first post:

It shows on even bandwidth sensitive iGPU test with Anandtech, DDR4 2133 is not faster than DDR3 1866.

To be on par with each other you need:

DDR3-1600 to DDR4-2133
DDR3-1866 to DDR4-2490
DDR3-2100 to DDR4-2800
DDR3-2400 to DDR4-3200

- Latency also matters and Haswell wouldn't perform significantly better using higher clocked memory with much worse latency, putting your lies of payed reviews to rest.

Kid, this will be about the 5th time I've showed in this thread that latency falls as RAM speed rises. CAS Latency matters a little, like I said to begin with, but when comparing different types/architectures of RAM, it doesn't mean much at all (unless of course, you go retardedly far in one direction or the other), as IntelUser2000 said above. Here we go, starting from the very first type of DDR:

First # on the left is the place indicator; the second # from left is bandwidth in GB/sec; the third # from left is DDR RAM speed; fourth # is the most important RAM timing, CAS latency; fifth # is total or true latency.

Yes, I realize that Sleepr won't even glance at this. This is for the people who are actually following along and reading every post, and may not know how to calculate DRAM latency.

DDR(1):
-------------------------------------------------------------------------------------
1. PC3200: 400 DDR @ CL 2= 10.00 ns
2. PC3200: 400 DDR @ CL2½=12.50 ns
3. PC3200: 400 DDR @ CL 3= 15.00 ns
4. PC3700: 466 DDR @ CL 2= 08.58 ns
5. PC3700: 466 DDR @ CL2½=10.72 ns
6. PC3700: 466 DDR @ CL 3= 12.87 ns
7. PC4000: 500 DDR @ CL 2= 08.00 ns
8. PC4000: 500 DDR @ CL2½=10.00 ns
9. PC4000: 500 DDR @ CL 3= 12.00 ns

DDR 2:
-------------------------------------------------------------------------------------
1. PC2-6400: 800 DDR @ CL 4= 10.00 ns
2. PC2-6400: 800 DDR @ CL 5= 12.50 ns
3. PC2-8000: 1,000 DDR @ CL 4= 08.00 ns
4. PC2-8000: 1,000 DDR @ CL 5= 10.00 ns
5. PC2-8500: 1,066 DDR @ CL 4= 07.50 ns
6. PC2-8500: 1,066 DDR @ CL 5= 09.38 ns

DDR3:
------------------------------------------------------------------------------------
01. PC3-10,600: 1,333 DDR @ CL 7= 10.50 ns
02. PC3-10,600: 1,333 DDR @ CL 8= 12.00 ns
03. PC3-10,600: 1,333 DDR @ CL 9= 13.50 ns
04. PC3-10,600: 1,333 DDR @ CL10=15.00 ns
05. PC3-10,600: 1,333 DDR @ CL11=16.50 ns
06. PC3-12,800: 1,600 DDR @ CL 7= 08.75 ns
07. PC3-12,800: 1,600 DDR @ CL 8= 10.00 ns
08. PC3-12,800: 1,600 DDR @ CL 9= 11.25 ns
09. PC3-14,900: 1,866 DDR @ CL 8= 08.57 ns
10. PC3-14,900: 1,866 DDR @ CL 9= 09.64 ns
11. PC3-14,900: 1,866 DDR @ CL10= 10.71 ns
12. PC3-16,000: 2,000 DDR @ CL 9=09.00 ns
13. PC3-16,000: 2,000 DDR @ CL10=10.00 ns
14. PC3-17,000: 2,133 DDR @ CL10=09.37 ns
15. PC3-17,000: 2,133 DDR @ CL11=10.31 ns
16. PC3-17,000: 2,133 DDR @ CL12=11.25 ns
17. PC3-19,200: 2,400 DDR @ CL10=08.33 ns
18. PC3-19,200: 2,400 DDR @ CL11=09.16 ns
19. PC3-19,200: 2,400 DDR @ CL12=10.00 ns
20. PC3-21,300: 2,666 DDR @ CL11=08.25 ns
21. PC3-21,300: 2,666 DDR @ CL12=09.00 ns
22. PC3-22,400: 2,800 DDR @ CL12=08.57 ns
23. PC3-23,400: 2,933 DDR @ CL12=08.18 ns
24. PC3-24,000: 3,000 DDR @ CL12=08.00 ns
25. PC3-24,000: 3,000 DDR @ CL13=08.66 ns
26. PC3-24,800: 3,100 DDR @ CL12=07.74 ns
27. PC3-24,800: 3,100 DDR @ CL13=08.38 ns

DDR4:
-------------------------------------------------------------------------------------
01. PC4-17,000: 2,133 DDR @ CL 12= 11.25 ns
02. PC4-17,000: 2,133 DDR @ CL 14= 13.12 ns
03. PC4-17,000: 2,133 DDR @ CL 15= 14.06 ns
04. PC4-17,000: 2,133 DDR @ CL 16= 15.00 ns
05. PC4-19,200: 2,400 DDR @ CL 12= 10.00 ns
06. PC4-19,200: 2,400 DDR @ CL 14= 11.66 ns
07. PC4-19,200: 2,400 DDR @ CL 15= 12.50 ns
08. PC4-19,200: 2,400 DDR @ CL 16= 13.33 ns
09. PC4-21,300: 2,666 DDR @ CL 13= 09.75 ns
10. PC4-21,300: 2,666 DDR @ CL 14= 10.50 ns
11. PC4-21,300: 2,666 DDR @ CL 15= 11.25 ns
12. PC4-21,300: 2,666 DDR @ CL 16= 12.00 ns
13. PC4-22,400: 2,800 DDR @ CL 14= 10.00 ns
14. PC4-22,400: 2,800 DDR @ CL 15= 10.71 ns
15. PC4-22,400: 2,800 DDR @ CL 16= 11.42 ns
16. PC4-22,400: 2,800 DDR @ CL 17= 12.14 ns
17. PC4-24,000: 3,000 DDR @ CL 14= 09.33 ns
18. PC4-24,000: 3,000 DDR @ CL 15= 10.00 ns
19. PC4-24,000: 3,000 DDR @ CL 16= 10.66 ns
20. PC4-25,600: 3,200 DDR @ CL 15= 09.37 ns
21. PC4-25,600: 3,200 DDR @ CL 16= 10.00 ns
22. PC4-26,400: 3,333 DDR @ CL 16= 09.60 ns
23. PC4-27,200: 3,400 DDR @ CL 16= 09.41 ns
24. PC4-28,800: 3,600 DDR @ CL 17= 09.44 ns

- AnandTech knows this and that's why they normalize clocks/latency

lmfao, so now you're willing to admit, 6 or 7 pages later, that what I pointed out in my first post was correct, that DDR3 1,866 CL 9 and DDR4 2,133 CL 16 are equivalents, making what I said to begin with, that some site using DDR4 2,666 was giving an advantage to the CPU that used DDR4. And no, you may not now pretend that you were the one who said it.

also explains why some websites favoured Haswell instead of Skylake with poor memory selection.

For the 8th or more time in this thread, you could not be more wrong with this statement. The proof is above. Higher speed RAM has lower latency, even when the CAS latency is higher.

leaving behind your precious Core i7 4790K especially when overclocked.

No offense, kid, but only an utter moron would make the above claim...or were you not aware that a K after the name of an Intel CPU denotes the ability to overclock it? Also, as I've noted multiple times throughout this thread, I've got nothing whatsoever against Skylake CPUs. I just was not going to allow you to spread the FUD that you were that in 2015, the better CPU to buy for gaming, for many years to come according to you, is an i5. That does apply to those people who buy new CPUs every year or two, of course, but that wasn't what you had claimed, when I responded.

Deal with it, it's getting boring.

I've been thinking that for many pages, now that you mention it.

 

[Sweepr]

Again, are you seriously admitting that you don't have the slightest idea how someone could intentionally cheat, when comparing two overclockable CPUs to each other, besides RAM speed, to make the performance difference between the two CPUs be higher?

You accused different reviewers of being payed by Intel without evidence only because their results don't match your obvious bias. So now all reviewers that show >10% gains per clock for Skylake in CPU limited games are cheating? Lol, whatever.

Yes, I realize that Sleepr won't even glance at this. This is for the people who are actually following along and reading every post, and may not know how to calculate DRAM latency.

These people will rather trust Crucial than a random poster trying to put Skylake in the worst light possible:

The true definition of latency and the latency equation
At a basic level, latency refers to the time delay between when a command is entered and executed. It's the gap between the two. Because latency is all about this gap, it's important to understand what happens after a command is issued. When the memory controller tells the RAM to access a particular location, the data must go through a number of clock cycles in the Column Address Strobe in order to get to its desired location and “complete” the command. With this in mind, there are two variables that determine a module's latency:

The total number of clock cycles the data must go through (measured in CAS Latency, or CL, on data sheets)
The duration of each clock cycle (measured in nanoseconds)

true latency (ns) = clock cycle time (ns) x number of clock cycles (CL)
Once again:

DDR3-1600 CL8/CL9 DDR3 (used in most Haswell tests) is superior to DDR4-2133 CL15 DDR4 in true latency (used in AnandTech's Skylake review).

Haswell DDR3 1600 CL9 = 1.25 ns x 9 CL = 11.25 ns true latency
Skylake
DDR4 2133 CL15 = 0.94 ns x 15 CL = 14.06 ns true latency
DDR4-2400 CL15 = 12.5 ns latency
DDR4-2800 CL15 = 10.7 ns latency
DDR4-3000 CL15 = 10 ns latency

You need much faster DDR4 to match the real latency of a simple DDR3-1600 CL9 kit, that's a fact, something you might not be used to.

lmfao, so now you're willing to admit, 6 or 7 pages later, that what I pointed out in my first post was correct, that DDR3 1,866 CL 9 and DDR4 2,133 CL 16 are equivalents, making what I said to begin with, that some site using DDR4 2,666 was giving an advantage to the CPU that used DDR4. And no, you may not now pretend that you were the one who said it.

Back to the basics:

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.

Let me help, AnandTech's calculation might be too complex for you which is why you're only making a fool of yourself.
DDR3-1866/CL9 = 207
DDR4-2133/CL15 = 142.2
DDR4-2666/CL15 = 177.73

Skylake would still be at a slight disadvantage with that DDR4-2666 CL15 kit according to AnandTech but, again, it didn't stop it from beating Haswell by ~13% @ PCLab's review. Different people in this thread provided actual benchmarks results where higher clocked memory doesn't necessarily beat lower clocked memory with lower latency and none of what you posted proved otherwise, basically busting your bullshit of pro-Skylake reviews.

No offense, kid, but only an utter moron would make the above claim...or were you not aware that a K after the name of an Intel CPU denotes the ability to overclock it? Also, as I've noted multiple times throughout this thread, I've got nothing whatsoever against Skylake CPUs. I just was not going to allow you to spread the FUD that you were that in 2015, the better CPU to buy for gaming, for many years to come according to you, is an i5. That does apply to those people who buy new CPUs every year or two, of course, but that wasn't what you had claimed, when I responded.

Reported for posting plain lies, never said that. I do think Core i7 6700K is the best gaming CPU at stock. We could also call it the best chip for most games (overall, stock or OCed) due to relatively poor scaling with more cores in most titles today, but Haswell-E's extra cores might make up for the slightly worse IPC/clocks in future titles, which is why I bought one instead of waiting for Skylake in the first place. Skylake Core i5 surprised me for matching previous generation Core i7s in some titles that benefit from more cores (according to PCLab) but I would never recommend it over Skylake Core i7 or Haswell-E for a long-term gaming system.