Intel Skylake / Kaby Lake

[Sweepr]

But the thing about IPC is that you want all other things to be equal. You don't know if for instance Haswell would also become 10% faster with the faster memory, then the IPC difference would still be almost zero.

If Haswell does not benefit from faster memory, so both aren't bottlenecked, then you're right.

See Raftina's post about this. DDR4-2133 is actually worse.

AnandTech:
DDR4: 2133/15=142.2
DDR3L: 1866/9=207.3
A difference of 65 in favor of DDR3L

Compare these to the "not even close" RAM that Hardware Canucks chose:
DDR4: 2666/13=205.1
DDR3L: 1866/11=169.6
A difference of 39 in favor of DDR4

More memory scaling results:

 

[Sweepr]

The mythical Skylake GT4e mobile Xeon is coming.

At Intel, we’ve been working closely with PC companies for years to deliver high performance workstations that deliver the computing demands of designers, content creators, engineers and architects. If you fall in this camp, you understand that not just any PC will do. You need the right mix of processor performance, memory, graphics and storage that lets you create, test, and deliver solutions. With the increasing popularity of digital creation (4K videos, digital design etc.), more creative professionals and engineers are seeing a need for workstation class capabilities in a portable device. In fact, IDC’s most recent report on mobile workstation usage showed that the quarter ending in June 2015 was the sixth straight quarter of year-over-year mobile workstation unit growth[1].

That’s why Intel is excited to announce plans to bring the power of Intel® Xeon® processors to notebook PCs for the first time. Designed for the professional who needs workstation-class capabilities in a portable device, Intel will soon be launching the Intel® Xeon® Processor E3-1500M v5 Product Family. This family of processors are based on the next gen Skylake architecture and they will deliver high precision computing horsepower in notebook form factors, delivering the right balance of power and mobility. Intel Xeon-based mobile workstations will have key features such as error-correcting code memory that automatically detects and repairs errors on-the-fly that cause data corruption and system crashes for peace-of-mind reliability. These new systems will also enjoy the benefits of the unique hardware-assisted security, manageability, and productivity capabilities of Intel vPro™ Technology. Mobile workstations featuring Intel Xeon will also feature Thunderbolt™ 3 – the USB-C that does it all. And if that’s not enough, Intel will also provide certifications for applications in computer-aided design and engineering, and digital content creation.

This is just scratching the surface of what to expect from the Intel® Xeon® Processor E3-1500M v5 Product Family for mobile workstations. We’re not quite ready to unveil all the details, but as you can see there is a lot for content creators, designers and engineers to get excited about. Stay tuned for more and the rest of the 6th Gen Intel Core family based on Skylake architecture.

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

From 3 months ago:

http://forums.anandtech.com/showpost.php?p=37385371&postcount=538

Also another another Skylake-S memory scaling gaming result: 17.2% faster in Watch Dogs simply by going from DDR4-2133 to DDR4-3333.

www.jagatoc.com/2015/08/skylake-ddr4-overclocking-part-1-dari-ddr4-2133-ke-ddr4-3333

 

[iSkylaker]

Any i7 multiplier locked owner regretting not having bought the equivalent unlocked version? I think I'm waiting a bit more and getting the i7-6700 instead.

 

[LTC8K6]

Well, the 4790 and the 4790K are far apart at stock speeds...

3.6/4.0 versus 4.0/4.4, so in this case, buying the K version gets you something even without an overclock.

We don't know if the 6700 and 6700K will have the same stock speeds, do we?

 

[witeken]

The mythical Skylake GT4e mobile Xeon is coming.

If those GT4s trickle down to the lower end SKUs: RIP AMD APU.

 

[witeken]

How do you guys think Apple's A9(X) chips will compete against Skylake-Y? And graphics?

Intel should really consider using a denser version of their process since Core m doesn't have to go higher than 3GHz anyway, and then they could significantly upgrade the IGP without sacrificing die area. A minus 35% transistor size advantage doesn't help much if the layout doesn't capitalize on it.

 

[Dufus]

We don't know if the 6700 and 6700K will have the same stock speeds, do we?

i7-6700 3.4GHz base, 4GHz turbo.

 

[Fjodor2001]

If those GT4s trickle down to the lower end SKUs: RIP AMD APU.

Perhaps strictly performance-wise. But what about when it comes to TDP? This mobile Xeon with GT4 is 35/45 W TDP. That's a lot for a mobile CPU these days.

IIRC, AMD's GPU cores have better perf/watt that Intel's. So for big iGPUs, I think AMD will have an advantage. Even more so when AMD moves to 14 nm for their APUs.

 

[Fjodor2001]

See Raftina's post about this. DDR4-2133 is actually worse.

I'm not sure I found an answer to Witeken's question there. Does that post really say that Haswell wouldn't also benefit from faster RAM?

Because if Haswell also benefits from faster RAM, then the same RAM speed ought to be used for both Haswell and Skylake when comparing them to derive the IPC increase. I.e. just as Witeken said.

 

[Idontcare]

How do you guys think Apple's A9(X) chips will compete against Skylake-Y? And graphics?

Intel should really consider using a denser version of their process since Core m doesn't have to go higher than 3GHz anyway, and then they could significantly upgrade the IGP without sacrificing die area. A minus 35% transistor size advantage doesn't help much if the layout doesn't capitalize on it.

Not to be overly moribund, but Steve Jobs passed away on Oct. 5, 2011. Essentially 4 yrs before the launch of what we would currently label as the A9(X). To whatever extent Steve's maniacal management approach was what directed (or limited, however you look at it) the creation of the chips that went into the iPhone/iPad/iPod hardware prior to his passing, for sure there will be essentially zero of that direction or guidance embedded in the A9(X) unlike the A8(X) and predecessors.

Furthermore, Jim Keller left Apple mid-2012. The maximum extent of his input/guidance/ingeniousness on the A9(X) at that point would have been limited to the first 9 months or so. Enough to set goals, but not enough to assist in the downselection process for features and opportunities when Samsung's 14nm and TSMC's 16FF+ electrical parametrics really came to be solidified in a PDK.

So the question for us regarding the A9(X) is how much innovation and team culture remain imbued in the A9 team that was present in the A8(X) which clearly trounced the Qualcomm team's abilities at utilizing TSMC's 20nm offerings to maximum effect?

If Jobs and Keller had minimal impact or influence on the A8 team, then we are justified to expect comparable awesomeness from the A9 team.

 

[LTC8K6]

i7-6700 3.4GHz base, 4GHz turbo.

And just 65 watts, if the prelim info is accurate.

http://www.cpu-world.com/CPUs/Core_i7/Intel-Core i7-6700.html

Also a T version at 35W, if anyone is interested.

http://www.cpu-world.com/CPUs/Core_i7/Intel-Core i7-6700T.html

 

[witeken]

Perhaps strictly performance-wise. But what about when it comes to TDP? This mobile Xeon with GT4 is 35/45 W TDP. That's a lot for a mobile CPU these days.

IIRC, AMD's GPU cores have better perf/watt that Intel's. So for big iGPUs, I think AMD will have an advantage. Even more so when AMD moves to 14 nm for their APUs.

How do you know. In any case, extremely unlikely.

Cause we're talking Gen9 here, and 14nm.

By the way, yesterday I read the Reddit AMA with the Intel architect another time, and what was very notable is that Intel isn't interested too much in higher performance, IPC; they do everything with efficiency, performance per watt in mind.

You saw this with Ivy, then Haswell, Broadwell had a stunningly high 2:1 performance for power increase for new features instead of the more usual 1:1.

So I wonder what Skylake will be designed like, cause they ditched the FIVR.

 

[phillyman36]

Has anyone heard anything on when we will see the cpus? Trying to decide whether to take my rig apart now and ebay it. I dont want to tear it down and the find out we wont see any cpus until mid September.

 

[witeken]

Not to be overly moribund, but Steve Jobs passed away on Oct. 5, 2011. Essentially 4 yrs before the launch of what we would currently label as the A9(X). To whatever extent Steve's maniacal management approach was what directed (or limited, however you look at it) the creation of the chips that went into the iPhone/iPad/iPod hardware prior to his passing, for sure there will be essentially zero of that direction or guidance embedded in the A9(X) unlike the A8(X) and predecessors.

Furthermore, Jim Keller left Apple mid-2012.

So the question for us regarding the A9(X) is how much innovation and team culture remain imbued in the A9 team that was present in the A8(X) which clearly trounced the Qualcomm team's abilities at utilizing TSMC's 20nm offerings to maximum effect?

If Jobs and Keller had minimal impact or influence on the A8 team, then we are justified to expect comparable awesomeness from the A9 team.

You can ask the same for Intel: in a few years we'll see the first products (although we already see them with things like Sofia and presumably Kabylake) without any influence from Otellini. For better or worse?

In any case, for mobile (which is part why Krzanich was hired in the first place) it can't get any worse :').

 

[Qwertilot]

I'm not sure I found an answer to Witeken's question there. Does that post really say that Haswell wouldn't also benefit from faster RAM?

Because if Haswell also benefits from faster RAM, then the same RAM speed ought to be used for both Haswell and Skylake when comparing them to derive the IPC increase. I.e. just as Witeken said.

Haswell standard won't benefit from DDR4 because it can't use it It probably shouldn't be counted as an IPC increase, but its going to be a definite plus for a Skylake build going forward so I guess have to consider it somehow.

Mind you, the real competition for Skylake K - Haswell E (/B'well E etc) and the locked quads due shortly - will all use DDR4 anyway.

Suppose its possible that Skylake somehow scales better with faster DDR4 than Haswell E.

 

[-batab-]

You guys are all hardware maniacs (I'm saying as a positive thing so don't be mad at it) but I'm wondering why are you discussing on few % differences.

We know nothing about measurement repeatibility, dispersion of results and secondary factors that can affect the result of different benchmarks of different samples mounted on different PCs. Skylake is better than Haswell, not by much and that's it.

Honestly I'm positively surprise by skylake because as it's pointed out, it has thermic limitations, meaning that a better dissipation could bring much higher frequencies. Also one must consider that Skylake die is probably 60-70 mm² and dissipating 100-200W on such a small surface is not easy at all. Also soldering could be much more difficult than with SB with its huge die.

 

[Dufus]

And just 65 watts, if the prelim info is accurate.

Yes but does that mean thermals (Tjc) are worse? Doesn't necessarily mean it will not use more than 65W after all TDP is based on the base frequency of 3.4GHz while the K is based on 4GHz..

 

[phillyman36]

Anyone care to list what parts they got for Skylake(mobo, memory and cpu for those who could find them)?

 

[coercitiv]

Because if Haswell also benefits from faster RAM, then the same RAM speed ought to be used for both Haswell and Skylake when comparing them to derive the IPC increase. I.e. just as Witeken said.

Some tests are both bandwidth and latency sensitive, like data compression. WinRAR benchmark is so sensitive to memory performance that it can severely skew results in favor of the platform with better memory specs. (make Skylake lose by a 1% or win by 20%) The only way to make it a "fair fight" is to pick both similar speeds and latencies. (see Haswell WinRAR tests, lots of scaling)

Other tests are latency sensitive only (bandwidth is enough for the task at hand even with DDR3 1600). Playing games on a powerful dGPUs can be an example. This case is rather simple, one just needs to ensure both platforms have similar (low) latencies. (see Haswell with multiple GPUs)

Other tests are bandwidth sensitive: iGPU gaming. The platform with higher memory bandwidth will have a tremendous advantage only if both iGPUs are bandwidth limited. Other times advantage is... nil.

On one hand, I don't completely agree with Sweepr, in the sense that I doubt faster memory specs "unleash" Skylake anymore than they do Haswell. I'd like to see some proper testing on this (not just multiple speeds, but multiple latencies for the same speed).

However, I do concur with Sweepr when it comes to signaling strange inconsistencies in reviewing decisions. It all comes down to what we want to compare:

• CPU archs - we should use memory kits with similar performance for both platforms. (similar bandwidth & latency)
• Platform performance - we should use the best each platform has to offer.

The argument that Anandtech used stock speed for Skylake doesn't hold up. They also used DDR3 2400 for their initial Haswell Review. Granted previous platforms could also take advantage of the same memory kit.

 

[Fjodor2001]

Haswell standard won't benefit from DDR4 because it can't use it

Of course, but for Haswell they could have used DDR3 at equal speed to the DDR4 used for Skylake.

 

[ShintaiDK]

Now if I could just get that in my ITX build...

 

[Abwx]

If those GT4s trickle down to the lower end SKUs: RIP AMD APU.

There s nothing to trickle down, and thats why it is used only in high TDP SKUs.

Iris Pro in BDW consume 50W.....

Godavari 7870K GPU is at 40W...

Carrizo GPU is barely at 25W and get 91% of the Iris Pro 3DMark11 score, and that s with regular RAM...

 

[Sweepr]

How do you guys think Apple's A9(X) chips will compete against Skylake-Y? And graphics?

Intel should really consider using a denser version of their process since Core m doesn't have to go higher than 3GHz anyway, and then they could significantly upgrade the IGP without sacrificing die area. A minus 35% transistor size advantage doesn't help much if the layout doesn't capitalize on it.

To heat up the comparison, here's GFXBench scores for Skylake-S GT2:

And here's Apple A8X:

Now obviously Core-M will provide slower scores thanks to lower sustained clocks and we have yet to see how much Apple A9X improves on Apple A8X. There's a rumour out there that only iPad Pro gets a new SoC this while iPad Air 3 will be a minor refresh (with Apple A8X)

If those GT4s trickle down to the lower end SKUs: RIP AMD APU.

Skylake GT3e would do that just fine looking at Broadwell-K performance and considering the architectural improvements from Gen 9, Skylake GT4e is in a league of its own.

I'm not sure I found an answer to Witeken's question there. Does that post really say that Haswell wouldn't also benefit from faster RAM?

Because if Haswell also benefits from faster RAM, then the same RAM speed ought to be used for both Haswell and Skylake when comparing them to derive the IPC increase. I.e. just as Witeken said.

Raftina's post says that AnandTech did not test them in equal conditions, it's not just about speed but latency too. By using the 'highest officially supported speed' excuse + worse than DDR3 latency they actually crippled Skylake relative to Haswell.

Slow DDR4 is lame compared to DDR3.

Strangely most websites out there got better results than them, even some that also did use DDR4-2133 like PC Perspective. At this point I miss old Anand Lal Shimpi reviews.
And then there's this:

7-Zip 9.2
- AnandTech: Skylake 1% slower than Haswell per clock
- Hardware.fr: Skylake 7% faster than Haswell per clock
- Hardware Canucks: Core i7 6700K 11.8% faster than Core i7 4790K with lower turbo clocks

POV-Ray
- AnandTech: Skylake 9.7% faster than Haswell per clock
- HardOCP: ''POV-Ray is a ray tracing for creating high quality graphics. We are using the benchmark included in the software and using its multicore ability. Again we see Skylake stretch its legs. Skylake rewards us with a 16% decrease in render time compared to Haswell, a 21% decrease compared to Ivy Bridge, and a 27% decrease compared to Sandy Bridge.'' (per clock, fixed 4.5GHz)

WinRAR
- AnandTech: Skylake 0.1% slower than Haswell per clock
- Hardware.fr: Skylake 10.85% faster than Haswell per clock
- Lab501: Core i7 6700K 9.4% faster than Core i7 4790K with lower turbo clocks

I can't find an explanation as to why their results are so much worse than others.

The argument that Anandtech used stock speed for Skylake doesn't hold up. They also used DDR3 2400 for their initial Haswell Review. Granted previous platforms could also take advantage of the same memory kit.

I agree. IMHO they should have tested both in the best conditions like many websites did, fast DDR3-2000+ for Haswell and DDR4-3000+ for Skylake (after all these are 'K' chips).

 

[ShintaiDK]

There s nothing to trickle down, and thats why it is used only in high TDP SKUs.

Iris Pro in BDW consume 50W.....

Godavari 7870K GPU is at 40W...

Carrizo GPU is barely at 25W and get 91% of the Iris Pro 3DMark11 score, and that s with regular RAM...

You keep making up numbers.

 

[cantholdanymore]

Not to be overly moribund, but Steve Jobs passed away on Oct. 5, 2011. Essentially 4 yrs before the launch of what we would currently label as the A9(X). To whatever extent Steve's maniacal management approach was what directed (or limited, however you look at it) the creation of the chips that went into the iPhone/iPad/iPod hardware prior to his passing, for sure there will be essentially zero of that direction or guidance embedded in the A9(X) unlike the A8(X) and predecessors.

Furthermore, Jim Keller left Apple mid-2012. The maximum extent of his input/guidance/ingeniousness on the A9(X) at that point would have been limited to the first 9 months or so. Enough to set goals, but not enough to assist in the downselection process for features and opportunities when Samsung's 14nm and TSMC's 16FF+ electrical parametrics really came to be solidified in a PDK.

So the question for us regarding the A9(X) is how much innovation and team culture remain imbued in the A9 team that was present in the A8(X) which clearly trounced the Qualcomm team's abilities at utilizing TSMC's 20nm offerings to maximum effect?

If Jobs and Keller had minimal impact or influence on the A8 team, then we are justified to expect comparable awesomeness from the A9 team.

Wow IDC, that's a very insightful post right there. I work for a major wireless operator and you'll be surprised who is ( almost) calling the shots in this company. I will say that the maniac Jobs's nature is pretty much embedded in Apple.