Intel Skylake / Kaby Lake

[mikk]

Thanks for sharing mikk. That's consistent with my expectations, around 15% faster than Haswell per clock.

You can't know this because of Turbo. If there is Turbo enabled on this Skylake, it is much less than 15%. Turbo can go up to 3.9 Ghz on i5-6600k.

 

[Sweepr]

You can't know this because of Turbo. If there is Turbo enabled on this Skylake, it is much less than 15%. Turbo can go up to 3.9 Ghz on i5-6600k.

As Arachnotronic said it's not a lot less than that. Don't forget GB results tend to vary considerably in different runs, Core i5 6600K might perform better in a retail high-end Z170 MB, with or without Turbo.

 

[Sweepr]

Here's a random Core i5 4690K score (3.5-3.9GHz, Z97 board):

ST: 3426
MT: 10850

Core i5 6600K score (3.5 base, Turbo enabled?, Z170 board):

ST: 3967
MT: 12234

15,7% better ST score.
12,7% better MT score.

Not bad at all.

 

[Enigmoid]

Here's a random Core i5 4690K score (3.5-3.9GHz, Z97 board):

ST: 3426
MT: 10850

Core i5 6600K score (3.5 base, Turbo enabled?, Z170 board):

ST: 3967
MT: 12234

15,7% better ST score.
12,7% better MT score.

Not bad at all.

Factor out the memory and consider this is a tick + tock and its terrible.

 

[Sweepr]

Factor out the memory and consider this is a tick + tock and its terrible.

SB vs Haswell is not any better in Geekbench, so why it's terrible? That's the expected improvement we got from the last few Tocks.

 

[MrTeal]

Factor out the memory and consider this is a tick + tock and its terrible.

It's also not any worse than Nehalem vs SB, and SB was a great arch.

 

[Excessi0n]

You can't know this because of Turbo. If there is Turbo enabled on this Skylake, it is much less than 15%. Turbo can go up to 3.9 Ghz on i5-6600k.

???

An i5-4670 has turbo too, you know. If the Skylake was only running at its base clock then there could be a 20%-plus improvement over Haswell. :awe:

 

[CHADBOGA]

Here is a i5-6600K Geekbench score: http://browser.primatelabs.com/geekbench3/2491359


My i5-4670 running 3.7 Ghz @DDR3-2133: http://browser.primatelabs.com/geekbench3/9539

6600K is said to run at 3.5GHz base/3.9GHz turbo.

Perf/clock improvement = 12.9%*(3.7/3.9) = ~12.24%.

Definitely solid for a "tock."

Here's a random Core i5 4690K score (3.5-3.9GHz, Z97 board):

ST: 3426
MT: 10850

Core i5 6600K score (3.5 base, Turbo enabled?, Z170 board):

ST: 3967
MT: 12234

15,7% better ST score.
12,7% better MT score.

Not bad at all.

How or why did a phone CPU benchmark, become the benchmark du jour for desktop CPU's?

Will GeekBench soon extend their benchmarking dominance to the Server world? D:

 

[Arachnotronic]

How or why did a phone CPU benchmark, become the benchmark du jour for desktop CPU's?

Will GeekBench soon extend their benchmarking dominance to the Server world? D:

It's quick and easy, especially since there are a lot of CPUs in the GB database.

Doesn't necessarily make it a good benchmark, but for ball-parking things, it can be useful.

 

[wilds]

GeekBench wasn't a mobile benchmark to begin with if I remember correctly. I could be wrong.

I think we are all interested in the overclocking headroom of Skylake and that will be the deciding factor to upgrade for a lot of us. I would really like to see an overclocked Broadwell-K vs Skylake-K vs older K series chips from nearly a decade back. Should be an exciting release especially if DDR4 will become super cheap.

 

[mikk]

???

An i5-4670 has turbo too, you know. If the Skylake was only running at its base clock then there could be a 20%-plus improvement over Haswell. :awe:

As I told my HSW is running at 3.7 Ghz in both ST and MT, for a IPC comparison this is a reliable base we can use. Not sure what is unclear to you. We can't be sure about the Skylake frequency of course.

 

[Hulk]

Here's a random Core i5 4690K score (3.5-3.9GHz, Z97 board):

ST: 3426
MT: 10850

Core i5 6600K score (3.5 base, Turbo enabled?, Z170 board):

ST: 3967
MT: 12234

15,7% better ST score.
12,7% better MT score.

Not bad at all.

Normalizing scores to equalize clocks in Anand's Haswell review and for Cinebench R10 single threaded shows the following.

Prescott to Conroe 85.6% IPC improvement
Conroe to Nehalem 14.7% IPC improvement
Nehalem to Sandy Bridge 19.6% IPC improvement
Sandy Bridge to Haswell 21.9% IPC improvement

It's only one benchmark so make of it what you will but the fruit seems to be getting harder and harder to reach.

 

[Sweepr]

I haven't seen this slide yet:

Mobile 4C/8T Skylake Xeon with GT4e Graphics?

 

[xthetenth]

Makes sense on a workstation laptop to allow users to avoid a discrete chip if they want.

 

[Headfoot]

Seems like we may have a repeat of 2011 -- Skylake = higher per core performance limited to 4 cores. Zen might push higher core counts for desktop, up to 8 if rumors are to be believed, like Bulldozer but hopefully still with decent per core performance.

It would be interesting if it shook down to 8C/16T Sandybridge level per-core with Zen at the same price as 4C/16T Skylake...

 

[mikk]

35W/45W seems low for a 4+4e SKU.

 

[coercitiv]

35W/45W seems low for a 4+4e SKU.

Haswell 4+3e SKUs are 47W. Considering 14nm power savings, 45W for 4+4e should be doable. The 35W will have slightly lowered clocks.

 

[frozentundra123456]

I haven't seen this slide yet:

Mobile 4C/8T Skylake Xeon with GT4e Graphics?

Like I have said many times, I wish they would just make 4+3e standard for the 35/45 watt quad core mobiles for Broadwell and Skylake. Matter of fact it would be nice to even see some of the standard voltage mobile Broadwell quads in available laptops.

Might be a decent low/mid level gaming machine without a discrete card, especially since AMD doesnt seem to be putting anything more that 19 watt chips in their laptops. Even with AMD's relatively better igpu, I cant see getting a decent gaming laptop in a 19 watt envelope for both cpu and igpu.

Is skylake supposed to be Gen 9 graphics? It would be nice if they just made the igpu more efficient instead of throwing more die space at it. Relative to the cpu, it seems like they are devoting almost as much space to the igpu as AMD, with much poorer results.

 

[Shivansps]

Here's a random Core i5 4690K score (3.5-3.9GHz, Z97 board):

ST: 3426
MT: 10850

Core i5 6600K score (3.5 base, Turbo enabled?, Z170 board):

ST: 3967
MT: 12234

15,7% better ST score.
12,7% better MT score.

Not bad at all.

This is mine at 4.2ghz
http://browser.primatelabs.com/geekbench3/2509175

 

[coercitiv]

Might be a decent low/mid level gaming machine without a discrete card, especially since AMD doesnt seem to be putting anything more that 19 watt chips in their laptops. Even with AMD's relatively better igpu, I cant see getting a decent gaming laptop in a 19 watt envelope for both cpu and igpu.

19W is clearly not enough. One can get 35W APU equipped laptops in Europe from Asus (FX 7600P, FHD TN screen, 16GB RAM), but it seems they decided to skip US market with them.

I still can't really understand why Haswell 4+3e SKUs never made it to Windows laptops.

 

[frozentundra123456]

19W is clearly not enough. One can get 35W APU equipped laptops in Europe from Asus (FX 7600P, FHD TN screen, 16GB RAM), but it seems they decided to skip US market with them.

I still can't really understand why Haswell 4+3e SKUs never made it to Windows laptops.

Have you seen any benchmarks on that FX 7600P?

As readers on this forum know, I am strongly against APUs on the desktop, because the 50 to 100 dollar savings vs a cpu plus discrete is a very poor compromise IMO. You just give up too much performance for the small price difference. Unfortunately intel plus any kind of decent gaming card (say GT850 or above) is in the thousand dollar ballpark. If one could get a decent apu in the 600.00 or less range, that is a significant enough difference to make me think about it.

 

[MrTeal]

19W is clearly not enough. One can get 35W APU equipped laptops in Europe from Asus (FX 7600P, FHD TN screen, 16GB RAM), but it seems they decided to skip US market with them.

I still can't really understand why Haswell 4+3e SKUs never made it to Windows laptops.

I'd have to guess it's just cost and benefit. The 4870MQ (3e) is ~$50 more than a slightly higher clocked 4810MQ(2) according to Intel's list price. When you're talking about a $400 processor in a $1000+ laptop, you're probably going to have a discrete GPU anyway, so why spend the extra on Iris Pro?

 

[R0H1T]

Have you seen any benchmarks on that FX 7600P?

As readers on this forum know, I am strongly against APUs on the desktop, because the 50 to 100 dollar savings vs a cpu plus discrete is a very poor compromise IMO. You just give up too much performance for the small price difference. Unfortunately intel plus any kind of decent gaming card (say GT850 or above) is in the thousand dollar ballpark. If one could get a decent apu in the 600.00 or less range, that is a significant enough difference to make me think about it.

What or how would you describe the said difference, say an i7 ULP level (CPU) perf or something even lower? You know very well that memory bandwidth are the APUs' Achilles heal, especially AMD.

Now with 14nm Excavator (hopefully further improved by then) I expect some decent level of perf from their mobile APUs, though at a slightly higher TDP, & am hoping that they'll integrate gen1 or gen2 HMB straightaway, certainly for the top SKU's :thumbsup:

This is off topic but anyway I've seen posters here say that if AMD did this or that they'd buy their products instantly, if it fits their budget of course. Not trying to start a flame war here but would you really do that, if given a choice between slightly higher TDP & better price?

I say this because I wanna gauge what is needed to sway someone AMD's way, if it's so then there's hope for AMD after all

 

[Sweepr]

Like I have said many times, I wish they would just make 4+3e standard for the 35/45 watt quad core mobiles for Broadwell and Skylake. Matter of fact it would be nice to even see some of the standard voltage mobile Broadwell quads in available laptops.

I think Iri Pro will become a lot more important for them from now on. It started as a niche product with Haswell, now Broadwell brings it to LGA desktops and Skylake will bring GT3e to 15W ''U'' ultrabook chips and a new GT4e SKU.

Also unless I'm missing something GT1 SKUs can't be found in any of the leaks, so perhaps GT2 iGPUs will be the base for Core-based products from now on (which makes sense given that low-end Braswell already packs 16 EUs). That's just speculation but a Gen 9 GT2 iGPU would be one hell of an upgrade from Gen 7.5 GT1 iGPUs currently found in Haswell Celeron/Pentium.

Is skylake supposed to be Gen 9 graphics? It would be nice if they just made the igpu more efficient instead of throwing more die space at it. Relative to the cpu, it seems like they are devoting almost as much space to the igpu as AMD, with much poorer results.

It will be Gen 9. Unfortunately we don't know the architecture details yet but I'm hoping they will solve some of the bottlenecks found in their current solutions. Otherwise it wouldn't make sense to dedicate so much area to graphics.

 

[Sweepr]

This is mine at 4.2ghz
http://browser.primatelabs.com/geekbench3/2509175

Interesting. Stock Core i5 6600K (Turbo enabled?) beats 4C/4T SB @ 4.2GHz by ~24% (single-core score).