Intel Skylake / Kaby Lake

[witeken]

- iGPU / Gaming
31% faster @ 3DMark 11 GPU
39% faster @ 3DMark Fire Strike Graphics
33% faster @ Just Cause 3 1366x768 Medium
39% faster @ Bioshock Infinite 1366x768 High
42% faster @ Battlefield 4 1366x768 High
51% faster @ Star Wars Battlefront 1366x768 Medium

http://www.notebookcheck.com/Kaby-Lake-Core-i7-7500U-im-Test-Skylake-auf-Steroiden.172422.0.html

Now one detail from Intel's slides caught my attention. Kaby Lake-U's HD Graphics 620 scores 339 pts @ 3DMark Time Spy Graphics (DX12). In comparison, a Geforce GTX 950M scores 281 pts according to NotebookCheck.

Why is Kaby Lake's graphics performance so much improved? It's still Gen9 right (Cannonlake would be gen10) with 24 EUs?

Edit: According to Intel Ark the graphics specs are entirely unchanged (clock speed), so the increase in performance must be due to improved power/efficiency.

 

[LTC8K6]

Why is Kaby Lake's graphics performance so much improved? It's still Gen9 right (Cannonlake would be gen10) with 24 EUs?

Edit: According to Intel Ark the graphics specs are entirely unchanged (clock speed), so the increase in performance must be due to improved power/efficiency.

Here is a good summary of the IGP:

http://www.anandtech.com/show/10610...six-notebook-skus-desktop-coming-in-january/3

 

[R0H1T]

Why is Kaby Lake's graphics performance so much improved? It's still Gen9 right (Cannonlake would be gen10) with 24 EUs?

Edit: According to Intel Ark the graphics specs are entirely unchanged (clock speed), so the increase in performance must be due to improved power/efficiency.

Jumping the gun perhaps, why not have a look at a broader set of numbers? There are (much) better numbers available for Skylake.

 

[SAAA]

Why is Kaby Lake's graphics performance so much improved? It's still Gen9 right (Cannonlake would be gen10) with 24 EUs?

Edit: According to Intel Ark the graphics specs are entirely unchanged (clock speed), so the increase in performance must be due to improved power/efficiency.

There's a large range in graphics frequency, 300MHz base to 1+GHz, so it's easy to say Skylake was downclocking while Kabylake can probably keep 30-40% higher speeds over time (14nm+ stretching his legs here at such low frequencies). Comparing desktop IGPs from 6th to 7th generation would probably show 0% difference unless they allow higher absolute clocks: they aren't limited by power at all.

As side note here... I still don't get why Intel hasn't pushed frequencies well over 1.5GHz on IGPs, at least on pricier parts, when a little overclock can do it on any chip? Pascal jumped 50% over Maxwell thanks to a speed bump alone, so why aren't they even trying?
I feel like Intel's saving improvements left and right to have a lot of margin when competitors do actually compete too much... perhaps Zen will push them to release six cores and GT3/eDRAM on most of the desktop lineup by the time Coffe/Cannon-lake is out.

 

[Arachnotronic]

There's a large range in graphics frequency, 300MHz base to 1+GHz, so it's easy to say Skylake was downclocking while Kabylake can probably keep 30-40% higher speeds over time (14nm+ stretching his legs here at such low frequencies). Comparing desktop IGPs from 6th to 7th generation would probably show 0% difference unless they allow higher absolute clocks: they aren't limited by power at all.

As side note here... I still don't get why Intel hasn't pushed frequencies well over 1.5GHz on IGPs, at least on pricier parts, when a little overclock can do it on any chip? Pascal jumped 50% over Maxwell thanks to a speed bump alone, so why aren't they even trying?
I feel like Intel's saving improvements left and right to have a lot of margin when competitors do actually compete too much... perhaps Zen will push them to release six cores and GT3/eDRAM on most of the desktop lineup by the time Coffe/Cannon-lake is out.

They could do a hand-tuned layout of the iGPU, but as you'll notice from the die shots, the CPU portions are "clean" and look as though they received a lot of hand tuning. The GPU portion is a blurry/blobby mess, suggesting that it relied heavily on synthesis tools.

Intel probably doesn't think it's worth the time/effort/expense of doing a really nice hand-tuned layout for the GPU as it does for its CPUs (the crown jewel and main selling point of its products).

 

[DidelisDiskas]

HD 620 (GT2) delivers around 70-80% the performance of Iris 540 (GT3e), this bodes very well for future Iris 640 (GT3e) parts launching January 2017.

I do wonder if that improvement will scale the same way to the more powerful offerings. The iris 580 was supposed to be a real monster, but turned out a bit of a disappointment the last i heard about it.

I don't get it. If Intel has done a really good job improving performance/features while keeping prices pretty much flat even as AMD has not been all that competitive, why exactly do we need competition? This is a legit question.

What if intel is not really pushing the train as hard as they would if there would be some serious competition. I bet they would do even better if amd was closing in right behind them.

 

[cbutters]

I don't get it. If Intel has done a really good job improving performance/features while keeping prices pretty much flat even as AMD has not been all that competitive, why exactly do we need competition? This is a legit question.

The price is flat and the performance is improving yes, but really... how quickly? Compared to ten years ago things have really slowed. How many times has intel sold you very close to the same 4 core 8 threaded processor for the same 300ish dollars in the last 5 years?, this is exactly what people are talking about when they say we need competition. It means that intel has already been price gouging you for as much as supply and demand will let them for half a decade.

 

[Azuma Hazuki]

You haven't actually answered my question.

Do I need to hold your hand and baby-step you through what happens when cartels and monopolies are allowed to form? Don't be disingenuous; you know very well why competition is necessary in this arena.

 

[Arachnotronic]

Do I need to hold your hand and baby-step you through what happens when cartels and monopolies are allowed to form? Don't be disingenuous; you know very well why competition is necessary in this arena.

Not being disingenuous. Please stop ducking my question with non-answers if you can actually explain why we need AMD around to provide competition.

 

[Arachnotronic]

The price is flat and the performance is improving yes, but really... how quickly? Compared to ten years ago things have really slowed. How many times has intel sold you very close to the same 4 core 8 threaded processor for the same 300ish dollars in the last 5 years?, this is exactly what people are talking about when they say we need competition. It means that intel has already been price gouging you for as much as supply and demand will let them for half a decade.

Laws of physics are the problem here, not lack of competition.

 

[Ajay]

I don't get it. If Intel has done a really good job improving performance/features while keeping prices pretty much flat even as AMD has not been all that competitive, why exactly do we need competition? This is a legit question.

Intel prices are flat because of a) the general weakness in the PC market and b) the are competing with their own previous gen parts and aren't offering much in the way of increased perceived value to consumers. Competition works only when companies try to one up each other on perceived value - which hasn't been the case for some time in the desktop/notebook markets. So there is no functional competition in these markets (unlike, say, mobile phones).

If AMD could offer a better CPU (say, 20% higher throughput for the same price/TDP) and was able to meet demand on said fictional CPU, then Intel would be forced to up the ante (on price, performance or both). Though, Intel has a pretty stellar reputation and name recognition (perceived values) that would command a bit higher price. In any case, we consumers would start to see better performance increases or prices or both (as there would be a functionally competitive market). That would be good news for us.

 

[Ajay]

Laws of physics are the problem here, not lack of competition.

Actually, vis a vis competition, it's all about money - not the laws of physics. Right now, no one can compete very well against Intel in pushing the laws of physics (with more R&D funding) for x86 processors (or, perhaps anything). The cost for getting a significant advantage over Intel is too high for any company to get a reasonable ROI on said R&D.

 

[cbutters]

Laws of physics are the problem here, not lack of competition.

True to some extent if you only want to look at silicon based flat wafer tech, but what if a competitor showed up on the scene who specialized in making CPUs out of graphene or other exotic materials? Or someone stacked silicon together to do some sort of 3d processing? Calling physics the problem is a cop out. Until I see something that has the equivalent of 275 TRILLION transistors and runs on 20 watts (like the human brain) We haven't even come close to tapping out the laws of physics There is always a way to improve performance, perhaps more competition would bring some out of the box answers to the forefront.

 

[witeken]

Laws of physics are the problem here, not lack of competition.

Once I said that I didn't think running in a race made you go faster. If I'm training, then I can also chose to go fast if I want. The person disagreed, and actually now I disagree as well. Humans by their very nature are really competitive. I'm more for cycling, and when someone goes past me I pretty much always have the urge to try to follow him. Take on some challenge. (Of course you have the benifit that you sit on his wheel, but it's just this instinct of competing that I'm getting at.)

I'd say competition definitely makes people go to their limit. I don't know how much this also influences economic activities, but people really like to be the best. So if Intel sees those Power9 benchmarks in front of them, I'd say that is for sure a motivation to see how they can do better.

 

[SpoCk0nd0pe]

Do you think there will be a Kaby Lake with significantly better performance at tasks that produce lots of cache misses?

A desktop version that clocks well and has more L3 cache or eDRAM?

Or maybe my dream Processor: quad core Skylake-X that clocks like a 6700K but with LGA 2011-v3 features, 24 mb L3 cache and 512 mb eDRAM

 

[Arachnotronic]

Once I said that I didn't think running in a race made you go faster. If I'm training, then I can also chose to go fast if I want. The person disagreed, and actually now I disagree as well. Humans by their very nature are really competitive. I'm more for cycling, and when someone goes past me I pretty much always have the urge to try to follow him. Take on some challenge. (Of course you have the benifit that you sit on his wheel, but it's just this instinct of competing that I'm getting at.)

I'd say competition definitely makes people go to their limit. I don't know how much this also influences economic activities, but people really like to be the best. So if Intel sees those Power9 benchmarks in front of them, I'd say that is for sure a motivation to see how they can do better.

By the time Intel sees power9 benchmarks, their product development pipeline is 4+ years deep and there's little they can do to adjust.

 

[Ajay]

Do you think there will be a Kaby Lake with significantly better performance at tasks that produce lots of cache misses?

A desktop version that clocks well and has more L3 cache or eDRAM?

Or maybe my dream Processor: quad core Skylake-X that clocks like a 6700K but with LGA 2011-v3 features, 24 mb L3 cache and 512 mb eDRAM

If the rumors are true, you won't be getting your dream Processor. Skylake-X will be a pretty standard HEDT processor on socket R4 (2066 pins). Kabylake-X will be a pretty standard unlocked quad core CPU that is socket compatible with R4. Intel is just further segmenting Server CPUs and HEDT CPUs.

 

[Sweepr]

http://pc.watch.impress.co.jp/docs/column/kaigai/1017921.html

 

[Ajay]

Anyone know if SKL-X is on Intel 14nm or 14+?

 

[jpiniero]

Anyone know if SKL-X is on Intel 14nm or 14+?

Unknown, it could go either way. Skylake-EP is more likely regular 14nm though because of the density loss.

 

[Arachnotronic]

Unknown, it could go either way. Skylake-EP is more likely regular 14nm though because of the density loss.

Skylake Server is on 14nm+.

 

[jpiniero]

Skylake Server is on 14nm+.

The biggest die is going to be huge if that's true. The Skylake Server cores are much bigger for sure, plus they are throwing in 4 more. The density loss isn't much I think but it'd be noticeable.

 

[SAAA]

The biggest die is going to be huge if that's true. The Skylake Server cores are much bigger for sure, plus they are throwing in 4 more. The density loss isn't much I think but it'd be noticeable.

I'm not sure if the density loss is appreciable at all on the actual die size, for all we know it could be a couple nanometers that don't affect the overall density of the superior layers (they said just larger fin pitch, not that all the node was enlarged equally).

Also even the largest Broadwell at 24 cores is only 2/3 as large as Knight Landing (456mm2 vs 680mm2), both on 14nm, so they have plenty of space to play with. This comparison is also one of the reasons that make me think server Skylake topping at 28cores is unreasonable: they already fab close to 700mm2 dies (with AVX512...) and sell them for less than Xeons but can't make higher core-count parts? Pftt, be ready for 32cores and AVX/larger caches this time, especially if Zen/Power9 just risk to give some actual competition.

 

[Sweepr]

Sneak peek at some Kaby Lake designs coming soon. Ultrabook review has a complete list of notebooks using the new chips, starting from $499 all the way up to $1299.

- Microsoft

As per reports reaching our news desk, Surface Pro 5 will be equipped with the new Kaby Lake processor to provide an enhanced graphics and longer battery life than the previous counterparts.

http://www.pc-tablet.com/microsoft-surface-pro-5-rumors-hint-4k-display-metal-design-kaby-lake-cpu/

- ASUS

Back in May during Computex, ASUS announced the Zenbook 3 laptop. However fast forward to today at IFA 2016, it seems that ASUS is introducing the Zenbook 3 again. If you haven’t purchased your Zenbook laptop yet, you will be pleased to learn that you have made a very good decision.

This is because with this reintroduction, ASUS will be giving the Zenbook 3 an upgrade to Intel’s Kaby Lake processors which is the latest generation, versus the 6th generation Skylake chipsets that would have found its way into the laptop earlier this year, and given all that we’ve heard about Kaby Lake, this can only be a good thing.

http://www.ubergizmo.com/2016/08/asus-zenbook-3-kaby-lake-processors

- Acer

Inside, this is one of the very first laptops to be offered with Intel's new seventh-generation Core i-series processors (sometimes known by the code name Kaby Lake), which Intel says will provide extra power for video decoding and playback, while operating more efficiently for better battery life.

Having had a chance to handle a Swift 7 recently, I can say it feels impressively thin in the hand, and gives the Spectre and other super-slim laptops some real competition.

Thinner than a Macbook 12'' - yet it comes with Kaby Lake-U, not Core M.

www.cnet.com/products/acer-swift-7/preview/

- Lenovo

The Lenovo Yoga 910 is the direct successor of the Yoga 900 introduced last year, which at present is Lenovo’s top-of-the-range offering in the Yoga lineup. The new Yoga 910 will not only offer higher performance (something that is logical to expect from a PC based on a newer CPU), but also a slightly larger 13.9” IPS display panel with either 4K (3840×2160) or FHD (1920×1080) resolution. Thanks to thinner bezel, the larger screen does not affect dimensions of the convertible, and in fact the new model is even a little smaller and thinner (14.3 mm vs 14.9 mm) than its predecessor. Still, it is noteworthy that the Yoga 910 weighs 1.38 kg (3.04 lbs), which is around 80 grams more than the weight of the Yoga 900. When it comes to battery life the UHD model can offer 10.5 hours on one charge (in line with current models that have 3K displays), whereas the FHD promises to work for up to 15.5 hours (which is a massive improvement over current SKUs).

http://www.anandtech.com/show/10627...ake-meets-4k-display-and-ultrathin-formfactor

- Razer

In addition, Razer is updating to the latest Kaby Lake processors, with the Core i7-7500U being the only CPU option. The storage starts at 128 GB for the QHD version, and 512 GB for the UHD model, with the UHD offering a 1 TB PCIe SSD now as well. System memory is also doubled to 16 GB of LPDDR3-1866, which is excellent to see, especially since the price has not changed. The rest of the product hasn’t changed, with the same dimensions and weight, even with the higher capacity battery. Check out our review of the first gen here. It keeps the same starting price of $999 and is available to order now.

http://www.anandtech.com/show/10640/razer-updates-the-razer-blade-and-razer-blade-stealth-at-pax

- Alienware

Alienware 13 13.3-inch
Kabylake Core U / max 32 GB RAM GTX 1060 ? 6.0 lbs / 2.7 kg

http://www.ultrabookreview.com/11702-laptops-nvidia-1060

 

[Ajay]

Skylake Server is on 14nm+.

Thanks. Do you have a source for that?
If SKL-EP is 14+, then I would think it's pretty certain that SKL-X will be too.