Intel Skylake / Kaby Lake

[daperl]

@mikk

FYI, there's a mainboard BIOS update:

https://downloadcenter.intel.com/download/25969/BIOS-Update-KYSKLi70-86A-

 

[sze5003]

That deals over :\

Damn I would have jumped on that.

 

[IntelUser2000]

I doubt they did compare a much lower clocked 45W GT4 against a 65W desktop GT3 with much better thermal solution.

45W Iris Pro 6200 gets about 2800 points in 3DMark11, and 65W version gets 3000. Almost no difference.

2800 vs 3200 = nothing

Would end up at 3400-3500 with 65W.

Actually Broadwell GT3e 65W is 10-30% off to a OC variant of GTX 750 in some newer games, so it's quite close to a default 750. GT4e 65W should beat this, probably even the OC variants.

If you put the same CPU, the GTX 750 pulls way ahead. Even the 740 is faster in some cases. All at lower power!

http://pclab.pl/art64766-5.html

Load - Idle = GPU power

Iris Pro 6200 = 80/84W for 5775C
GTX 750 = 77/84W

Equal/lower power with faster performance!

Given that the 3dmark scores are lower than the scores from nucblog it is certainly not the best 3dmark11 score NUC6i7KYK can do. For you 3dmark may be all what matters, the real question how it performs in a game.

1. Worthiness of a product is determined by competitiveness. Iris Pro 6200 was not competitive with Maxwell. Pascal improves perf/watt by 70%. That's all despite using a seperate PCB(since its a video card) and GDDR5 memory not a proprietary eDRAM and interface. The original Iris Pro 5200 was used by no one because it fell behind discrete graphics boards in perf/watt and perf/$.

2. Only available on a super expensive CPU. For gaming, GTX 750 level of performance is usually matched with a CPU that's at the same "class". Being on a $500 CPU means its essentially GTX 750 class at $500. Remember how mid range GPU went from $299 to $499 to $699? Well, this is the same thing. But they don't make it obvious. You went from a $4 motherboard addition to having to buy $200 minimum cost CPUs that some call it "APUs".

Not all are good, but certain 3DMark benchmarks are accurate for its time. 3DMark06 was roughly telling how iGPUs performed back in the days. 3DMark11 tells roughly how iGPUs perform now. 30% faster in thermal/power stabilized 3DMark11 score for Iris 540 versus HD 520 turns out to be 30% average gain for gaming benchmarks. It's quite accurate.

Of course, if you start using ridiculously low resolutions and settings like TH did(or super old games like HL Lost Coast) with their original Iris Pro 6200 "review", 3DMark11's may not reflect it well, since gaming component of 3DMark at that level of performance runs at far more "graphically intensive" frames per second.

 

[mikk]

45W Iris Pro 6200 gets about 2800 points in 3DMark11, and 65W version gets 3000. Almost no difference.

What is your source? There is a power difference between 48 EUs and 72 EUs, you cannot extrapolate it from a GT3. Otherwise you could argue with a GT2 and say that there is no difference between 45W and 95W.

2800 vs 3200 = nothing

Could you read properly?

Would end up at 3400-3500 with 65W.

Given that the slowest 45W GT4e SKU is able to score 3600 and that a 65W version probably has a higher GPU Turbo than the slowest 45W GT4e SKU means, you are wrong.

 

[Rngwn]

And here we have a 3600 score.

http://www.ocstation.com/cpu-processors/intel-nuc-nuc6i7kyk-review/all

As an important note, this review only use single-channel RAM. Still wondering how much a second stick of RAM would affect the performance.

 

[mikk]

Means the score could be even higher. Maybe that's why their Firestrike score is lower than from the Nucblog.

 

[dark zero]

Only 10-15% higher due Dual Channel. But since is not an API, I don't expect more improvement.

 

[mikk]

+10-15% would result in a score of slightly over 4000. I'm not sure if this is true for an edram version in 3dmark11. Sounds a bit too high. 65W SKUs with Iris Pro 580 can go up to 1150 Mhz, +200 Mhz or potentially +21% more GPU power.

 

[mikk]

http://www.notebookcheck.com/Test-Acer-Aspire-V3-372-57CW-Subnotebook.165512.0.html
http://geizhals.de/acer-aspire-v3-372-57cw-schwarz-nx-g7bev-010-a1417406.html?hloc=at&hloc=de


i5-6267U with Iris 550 Graphics, 8GB Dualchannel, 256GB SSD, 13.3" 1920x1080 IPS for 719€ is quite good. This device is configured as 35W says Notebookcheck, although from the Furmark+Prime95 load HWinfo screen CPU package doesn't draw over 27W.

 

[frozentundra123456]

Yea,very nice.

 

[Sweepr]

Some of the new Gigabyte motherboards that should be announced at Computex:

http://mb.zol.com.cn/583/5832748.html

Ps: Nice find mikk. Now we need a similar Iris Pro 580 design win.

 

[Abwx]

http://www.notebookcheck.com/Test-Acer-Aspire-V3-372-57CW-Subnotebook.165512.0.html
http://geizhals.de/acer-aspire-v3-372-57cw-schwarz-nx-g7bev-010-a1417406.html?hloc=at&hloc=de


i5-6267U with Iris 550 Graphics, 8GB Dualchannel, 256GB SSD, 13.3" 1920x1080 IPS for 719€ is quite good. This device is configured as 35W says Notebookcheck, although from the Furmark+Prime95 load HWinfo screen CPU package doesn't draw over 27W.

They should explain how a 13"3 laptop with IGPU has 27W CPU power package while draining as much as 49.5W at the main and is throttling at 44.5W...

 

[The Stilt]

They should explain how a 13"3 laptop with IGPU has 27W CPU power package while draining as much as 49.5W at the main and is throttling at 44.5W...

On AMD FX-8800P reference system with TDP & PPT set to 27W, the sustained power draw from the wall is 40.5W while the peak is 42.75W during P95 28.9 and Furmark :sneaky:

Meanwhile the battery I2C telemetry (running on battery power) shows discharge rate of 37.2W sustained and 38.8W peak.

 

[dark zero]

Still waiting for Skylake C, that is the only way to see Intel reaching GTX 950 levels on their Iris Pro.

 

[The Stilt]

How wide is the eDRAM interface on Broadwell / Skylake and at what clocks the eDRAM runs at?

 

[IntelUser2000]

Could you read properly?

Personal attacks doesn't change the fact that its overpriced + for a GPU 10% gains are nothing. For a CPU that's very good. Did we forget that Polaris/Pascal is coming with far better efficiency compared to predecessors(which were already better)? Value of the product is often judged by how well its doing compared to the competition. Fury and Fury X is good by itself, but not when Nvidia is there with 980 Ti. Anandtech words: "There are no bad products. Only bad pricing".

Given that the slowest 45W GT4e SKU is able to score 3600 and that a 65W version probably has a higher GPU Turbo than the slowest 45W GT4e SKU means, you are wrong.

*Refer back to above.

i5-6267U with Iris 550 Graphics, 8GB Dualchannel, 256GB SSD, 13.3" 1920x1080 IPS for 719€ is quite good.

This system is ok, because it'll be priced right if we consider 1 Euro often translates 1:1 to US dollars.

They should explain how a 13"3 laptop with IGPU has 27W CPU power package while draining as much as 49.5W at the main and is throttling at 44.5W...

Come on buddy, you know better. It's at LOAD. With MAXIMUM screen brightness.

Furthermore, ask yourself this:
-Does SSD use power?
-Does motherboard PCB(which includes non-main components) use power?
-Does memory use power?
-Does maximum power use equal average power at load?

Intel CPUs since Sandy Bridge has the capability to briefly go above burst for a while. Now people argue that this is thermal(read: heat) only. But thermals are usually related to power use. And Intel manuals state that power considerations are taken into effect. Battery life tests show that the CPUs act just like the manuals state, being over a long period limited in power use to TDP limits, not maximum power aka V x I.

How wide is the eDRAM interface on Broadwell / Skylake and at what clocks the eDRAM runs at?

On Broadwell, it uses the same one as Haswell. 1.6GHz with 50GB/s bi-directional bandwidth(100GB effective, does that make it 256-bit width per way?). Skylake, considering its available on low power SKUs where idle power is different, likely uses the 2nd generation version with 1/4th standby power(active is likely same at 3W).

 

[ShintaiDK]

How wide is the eDRAM interface on Broadwell / Skylake and at what clocks the eDRAM runs at?

You can see below.

Same with Haswell too. The controller is just moved with Skylake/Kaby Lake so its in the system agent now and wont need cache tags from the L3.

http://pc.watch.impress.co.jp/docs/column/kaigai/20140311_638982.html

 

[SAAA]

Kabylake is coming

Model has advanced to 142 and stepping to 9 from Skylake's 94/78 stepping 3.
There should be some process improvements but it still looks like ~0-5% better IPC. Intel not too worried by Zen? Well mobile not till next year for sure...

 

[Phynaz]

On AMD FX-8800P reference system with TDP & PPT set to 27W, the sustained power draw from the wall is 40.5W while the peak is 42.75W during P95 28.9 and Furmark :sneaky:

Meanwhile the battery I2C telemetry (running on battery power) shows discharge rate of 37.2W sustained and 38.8W peak.

Thank you

 

[CHADBOGA]

Kabylake is coming

Model has advanced to 142 and stepping to 9 from Skylake's 94/78 stepping 3.
There should be some process improvements but it still looks like ~0-5% better IPC. Intel not too worried by Zen? Well mobile not till next year for sure...

Things are set in stone this close to release, so it doesn't matter whether or not Zen exceeds expectations.

If it does, then Cannonlake would be Intel's opportunity to address it.

 

[AtenRa]

Things are set in stone this close to release, so it doesn't matter whether or not Zen exceeds expectations.

If it does, then Cannonlake would be Intel's opportunity to address it.

Desktop CannonLake may not even be a 2017 product.

 

[CHADBOGA]

Desktop CannonLake may not even be a 2017 product.

So there would be a period where Kabylake is Intel's product up against Zen.

 

[AtenRa]

So there would be a period where Kabylake is Intel's product up against Zen.

It seems Kaby and ZEN will launch very close to each other within Q4 2016.

Intel will also release Broadwell-E at the end of May.

 

[jpiniero]

Yeah, you have to figure that Intel is going to re-arrange Kabylake's release of the various parts so that the equiv Cannonlake part would come out a year later. It's probably only U and Y in 2016.

 

[Sweepr]

AnandTech: The Intel Skull Canyon NUC6i7KYK mini-PC Review

The desktop computing market as a whole has been subject to severe challenges over the last few years. The ultra-compact form-factor (UCFF) PC market that emerged with the introduction of the Intel NUCs (Next Unit of Computing) has been one of the few bright spots. PC gaming has been one of the few other markets that has withstood the overall issues. The small size of UCFF PCs usually made discrete GPUs difficult to integrate, and iGPUs have not impressed the gaming crowd. Therefore, the market has not seen many products targeting the gaming market while also being compact. This year, we have a new entrant in that category - Intel's Skull Canyon NUC, the NUC6i7KYK, places a 45W TDP Core i7-6770HQ with Iris Pro graphics in a chassis around twice the size of the standard NUC.

www.anandtech.com/show/10343/the-intel-skull-canyon-nuc6i7kyk-minipc-review