Intel Skylake / Kaby Lake

[Sweepr]

I wonder if there are other differences between desktop/mobile Skylake cores and server Skylake cores beside AVX-512.

 

[cbn]

Intel is getting worse about aligning HEDT and DT, not better.

For people that want more than four cores, I'm sure Intel is working on the fix.

Personally, I think that Xeon-D is pretty sweet and would like a high power version of that for a Skylake mini-HEDT.

 

[Sweepr]

That's what I'm worried about. Old roadmaps talked about Skylake in 1S is 2H 2016, but lately all I've seen is Broadwell-E in 1Q 2016. Given how quick Broadwell is being replaced on the desktop I was hoping we'd still see SL-E in 2016, but that's not looking promising.

Intel is getting worse about aligning HEDT and DT, not better. We'll be on CannonLake Refresh by the time SL-E rolls around.

Yup, Broadwell-EP 4S and Broadwell-EX might not be too far from Cannonlake-Y/U launch. Perhaps server cores (starting with server Skylake) are different in more ways than we know right now.

 

[cbn]

I am liking this:

-QuickAssist encryption and compression offloads
-Skylake+ FPGA
-Cannonlake graphics & media transcode

Under "Optional Integrated Accelerators"


Originally, they were talking about Haswell being the generation with FPGAs and integrated accelerators.

Also its interesting that they think of "Cannonlake graphics & media transcode". Something special is happening here. Perhaps with Cannonlake successor, the iGPU would be general purpose enough to use it as an accelerator. Maybe a real "Larrabee" with Airmont cores?

That would be interesting in light of AMD launching APU servers and Nvidia a member of OpenPOWER Consortium.

Regarding Nvidia, I am thinking they would first use NVLINK to connect Power architecture with Pascal, but after that they integrate their GPU with Power (at the high end, for the strong single thread) in addition to ARM (at the lower end) for their own type of APUs.

Then Intel (using strictly FPU within the x86 cores) competes with Nvidia's and AMD's instruction set strategy (which I am guessing would be different and try to bridge FPU between CPU and iGPU)?

P.S. Power6 launched in 2007, Power7 launched in 2010, Power 8 launched in 2013....that means Power9 (whatever that is) would come in 2016 if trend of 3 years per generation continues.

 

[Ajay]

If the roadmap proves accurate, you can see the 1S workstation Broadwell release at 2016, and 1S Skylake release at ~mid 2017.

1S Workstation is exactly HEDT.

Yes, but I just noticed on that Roadmap that the Basin Falls 1S Workstation platform is still in planning (not under development!). So those interested in Skylake-E may have to wait a bit longer. I'm guessing this is a one time hiccup due to the merging of the E5/E7 series.

 

[FlanK3r]

some leaked benchmarks


Geekbench single core


Geekbench multithread


Cinebench R11.5 sinle


Cinebench R11.5 - multithread


Cinebench R15-single core


Cinebench R15 multithread


http://www.cpu-monkey.com/en/

 

[Aristotelian]

some leaked benchmarks


Geekbench single core


Geekbench multithread


Cinebench R11.5 sinle


Cinebench R11.5 - multithread


Cinebench R15-single core


Cinebench R15 multithread


http://www.cpu-monkey.com/en/

I'm a bit surprised by these. I was expecting or rather hoping for a consistently dominant improvement over previous generation(s) and when a 39xx at a lower clock speed (even on a single core test) does better...

Perhaps it's too early to tell.

 

[FlanK3r]

for Cinebench 15 it could be very big L3 size cache as bonus for 3970x

 

[mikk]

some leaked benchmarks

Already known. We can't verify these benchmarks, it seems odd that they have scores from all Skylake SKUs already.

 

[Kallogan]

i hope those benchs are wrong cause it's friggin disappointing.

Well i wouldn't be surprised if skylake is all about igpu

 

[ShintaiDK]

some leaked benchmarks

http://www.cpu-monkey.com/en/

Those benchmarks have already been debunked as complete nonsense.

And its quite clear, look on CB15 ST for example. And then compare with say...Anandtech numbers.

 

[mikk]

Intel Corporation Skylake Client platform
Intel Core m7-6Y75 @ 2.40 GHz
http://browser.primatelabs.com/geekbench3/2637682


A first Geekbench entry of Skylake-Y. Does it mean base frequency will be 2.40 Ghz for m7-6Y75?

 

[Sweepr]

Intel Corporation Skylake Client platform
Intel Core m7-6Y75 @ 2.40 GHz
http://browser.primatelabs.com/geekbench3/2637682


A first Geekbench entry of Skylake-Y. Does it mean base frequency will be 2.40 Ghz for m7-6Y75?

That would be twice Core M-5Y71's base clock. Very impressive if true (big if here).
Intel needs a fast Skylake-Y not only because of the 16FF+/14nm ARM tablet contenders (late 2015 & 2016) but also to allow Broxton to significantly improve Atom's CPU performance (not closing the gap with the Core line too much).
As I said previusly Broadwell-Y was a nice first attemp but I think Skylake-Y will be a more serious contender for fanless premium tablets and convertibles.

 

[MrTeal]

That would be twice Core M-5Y71's base clock. Very impressive if true (big if here).
Intel needs a fast Skylake-Y not only because of the 16FF+/14nm ARM tablet contenders (late 2015 & 2016) but also to allow Broxton to significantly improve Atom's CPU performance (not closing the gap with the Core line too much).
As I said previusly Broadwell-Y was a nice first attemp but I think Skylake-Y will be a more serious contender for fanless premium tablets and convertibles.

Very impressive if it's at 4.5W. That's 40% higher than the 5Y70 and essentially the same score at an i5-5300U.
However, the 5Y70 and 5300U are the same die. Without knowing what kind of TDP limits were in place for the Skylake-Y chip, absolute performance numbers don't mean a whole lot.

 

[Ken g6]

I'm a bit surprised by these. I was expecting or rather hoping for a consistently dominant improvement over previous generation(s)

Don't forget to account for clock speed differences in the single core tests. When I do, I get a consistent 9% IPC improvement for the 6700K. Actually, it's so consistent I'd call it suspiciously consistent. :hmm:

 

[piasabird]

Where did they get benchmarks? I wonder if they have actual production retail parts in their final form yet. Often early release parts don't have fully functional BIOS versions.

 

[IntelUser2000]

Very impressive if it's at 4.5W. That's 40% higher than the 5Y70 and essentially the same score at an i5-5300U.
However, the 5Y70 and 5300U are the same die. Without knowing what kind of TDP limits were in place for the Skylake-Y chip, absolute performance numbers don't mean a whole lot.

This is actually terrible. According to GeekBench, the Skylake Core M is mere 10% better than current Macbooks. Also, the Asus T300 Chi's Core M 5Y71 gets pretty close to the score too. The problem is, while Macbook is pretty good in GeekBench, it sucks in everything else, like applications people use: http://www.anandtech.com/show/9136/the-2015-macbook-review/9

If you look at how vast majority of the 5Y70/5Y71 performs, you need about 20-30% to be at an ok level. If you want it really good, Skylake Core M needs to be 20-30% faster than Intel's misleading, hyped, preliminary benchmarks.

Or lower the price of the CPU to $40.

 

[Sweepr]

This is actually terrible. According to GeekBench, the Skylake Core M is mere 10% better than current Macbooks. Also, the Asus T300 Chi's Core M 5Y71 gets pretty close to the score too. The problem is, while Macbook is pretty good in GeekBench, it sucks in everything else, like applications people use: http://www.anandtech.com/show/9136/the-2015-macbook-review/9

If you look at how vast majority of the 5Y70/5Y71 performs, you need about 20-30% to be at an ok level. If you want it really good, Skylake Core M needs to be 20-30% faster than Intel's misleading, hyped, preliminary benchmarks.

Or lower the price of the CPU to $40.

Actually Core M's biggest problem is sustained performance. If Skylake-Y's base clock doubles (2.4GHz, which remains to be seen) then it will greatly improve performance all around if you do more than quick benchmarks runs with your computer.
Geekbench runs last less than a minute, the fact that a Skylake-Y (ES?) is already posting better scores than some of the best Broadwell-Y designs out there is great and indicates peak performance will go up too but I'm more interested to see how it performs under common apps.

 

[IntelUser2000]

Actually Core M's biggest problem is sustained performance. If Skylake-Y's base clock doubles (2.4GHz, which remains to be seen) then it will greatly improve performance all around if you do more than quick benchmarks runs with your computer.

http://browser.primatelabs.com/geekbench3/search?page=29&q=Core+M+5Y71

Most of the results indicate that its already running at the frequencies you are suggesting.

 

[Sweepr]

http://browser.primatelabs.com/geekbench3/search?page=29&q=Core+M+5Y71

Most of the results indicate that its already running at the frequencies you are suggesting.

Did you expect anything different?
Geekbench is not indicative of sustained performance (where higher base clocks would help), the Broadwell-Y scores there are mostly running above 2GHz. Skylake might improve max Turbo clocks and bring ~10% IPC bump on top of that but I doubt we will se more than >20% improvement in quick benchmarks like Geekbench or Javascript browser stuff.

 

[mikk]

http://browser.primatelabs.com/geekbench3/search?page=29&q=Core+M+5Y71

Most of the results indicate that its already running at the frequencies you are suggesting.

There is a big fluctuation in these Geekbench scores from M-5Y71 because of the low base and high Turbo. Especially multithreading, I see lots of results in the 4000-5000 range which is much slower. Base clock on that SKL-Y seems to be set at 2.4 Ghz, therefore it should result in a much more consistent performance in real world. On the other side 2.4 Ghz base seems too good to be true. But we will see.

 

[MrTeal]

There is a big fluctuation in these Geekbench scores from M-5Y71 because of the low base and high Turbo. Especially multithreading, I see lots of results in the 4000-5000 range which is much slower. Base clock on that SKL-Y seems to be set at 2.4 Ghz, therefore it should result in a much more consistent performance in real world. On the other side 2.4 Ghz base seems too good to be true. But we will see.

The Core M chips are challenging since they are so dependent on proper cooling to maintain full turbo. I was just using the 2711/5706 numbers of the Intel Skylake Client Platform vs the 1918/3232 from the Intel Broadwell Client Platform hoping they were the most similar, but that could be completely off the mark.

 

[IntelUser2000]

The Core M chips are challenging since they are so dependent on proper cooling to maintain full turbo. I was just using the 2711/5706 numbers of the Intel Skylake Client Platform vs the 1918/3232 from the Intel Broadwell Client Platform hoping they were the most similar, but that could be completely off the mark.

Remember this?: http://www.anandtech.com/show/8515/quick-look-at-core-m-5y70-and-llama-mountain

What about this?: http://www.pcper.com/reviews/Processors/IDF-2014-Intel-Shows-Core-M-5Y70-Performance-Numbers

2.77 points in Cinebench!

Not a single selling device gets this, not even the Asus T300 Chi which seems to be using 9W on the SoC level to do so. How many popular internet sites hyped the Core M again? Anand's article about "thermals" affecting performance is an excuse, that's all. They should have said that BEFORE hyping with pre-release Intel-sanctioned articles.

All nonsense straight from Intel. It makes perfectly skeptical about any Core M claims after that fiasco.

 

[dahorns]

Remember this?: http://www.anandtech.com/show/8515/quick-look-at-core-m-5y70-and-llama-mountain

What about this?: http://www.pcper.com/reviews/Processors/IDF-2014-Intel-Shows-Core-M-5Y70-Performance-Numbers

2.77 points in Cinebench!

Not a single selling device gets this, not even the Asus T300 Chi which seems to be using 9W on the SoC level to do so. How many popular internet sites hyped the Core M again? Anand's article about "thermals" affecting performance is an excuse, that's all. They should have said that BEFORE hyping with pre-release Intel-sanctioned articles.

All nonsense straight from Intel. It makes perfectly skeptical about any Core M claims after that fiasco.

(1) Intel has been fairly silent on Skylake-Y. We aren't talking about Intel hype, but rather random leaked benches.

(2) Pretty sure the Asus T300 does get those numbers, as I remember a few sites reporting ~2.8 for the T300 when docked with its keyboard. But, the keyboard dock gives it extra thermal room and allows it to stay at max turbo a lot longer, operating well above the 4.5W TDP.

(3) Core-M (Broadwell) scored fairly closely to the U-series parts in short benchmarks.

(4) If Skylake allows Core-M to offer sustained performance at 2.4 Ghz or above, it'll be a massive improvement over the Broadwell version. We should also expect that Intel's SOC process will be much more refined, allowing for better tablet implementations.

 

[MrTeal]

On the pcper preview you linked, the 45997 in Ice Storm matches pretty well with what Anand got.

In Cinebench, the 2.77 for Llama Mountain is probably correct. The XPS13 gets 2.83 with the same silicon but 3MB L3 vs 4MB in the 5Y70. Considering Llama Mountain was running at 6W that kind of score seems reasonable if it was kept cool and could boost to 2.6GHz consistently.