Intel Skylake / Kaby Lake

[IntelUser2000]

Not really. FP16 vs FP32.

How much difference do you think it'll make?

Usually doubled compute power resulted in ~30% improvement. FP16 vs FP32 should be that. That means at best, HD 520 is equal to iPad Pro graphics.

Intel should seriously be worried. SpecCPU2000 results show that with gcc, A9 is 10% faster per clock than Haswell, which means equal to Skylake. With GPU, iPad Pro is going to do at 3-5W that Intel needs 15W to do so.

Yea ok let's assume GT3e for additional gain, but at 10x the SoC cost and still 3-4x the TDP + significantly lower battery life per WHr.

At this point it means for all intents and purposes, TSMC 16 = Intel 14. Die size, transistor performance, cost per transistor. Ok, I would assume Apple SoC design team is far more organized than Intel which is why they can put such a good result. The latter has been going like a blind man navigating through mobile "waters".

 

[lsquare]

So far I only see the K processors on the market. When will the other socketed Skylake processors launch?

 

[einhoernchen]

So far I only see the K processors on the market.

Here in Europe it's the opposite: It's almost impossible to buy a 6700K (except you're willing to buy 100 EUR more than the RSP) but the 6700 and so on are quite easy to get.

 

[Sweepr]

How much difference do you think it'll make?

Usually doubled compute power resulted in ~30% improvement. FP16 vs FP32 should be that. That means at best, HD 520 is equal to iPad Pro graphics.

Adding 30% to the best HD Graphics 515 scores would make A9X only ~10-20% faster @ T-Rex and Manhattan Offscreen, assuming it's twice as fast as A8X. Also there's no 4.5W Core m7 device out right now, I bet these scores come from the 4.5W Core m3 version of Surface Pro 4. Core m7's HD Graphics 515 operates at up to 1GHz, up to ~17.7% higher than the low-end model (same clock as Core i5-6300U's HD Graphics 520), which means it would be pretty damn close to A9X @ GFXBench in equal conditions if your 30% estimate is real.

*Ignoring the 86.5 FPS result a Skylake-Y device got @ Manhattan Offscreen, probably an outlier.

 

[DrMrLordX]

Im currently testing that system, memory can be set at 1333,1600 and 1866MHz. That is the maximum for DDR-3.

Interesting. I'll PM you about that to avoid thread-bloat.

 

[AtenRa]

Adding 30% to the best HD Graphics 515 scores would make A9X only ~10-20% faster @ T-Rex and Manhattan Offscreen, assuming it's twice as fast as A8X. Also there's no 4.5W Core m7 device out right now, I bet these scores come from the 4.5W Core m3 version of Surface Pro 4. Core m7's HD Graphics 515 operates at up to 1GHz, up to ~17.7% higher than the low-end model (same clock as Core i5-6300U's HD Graphics 520), which means it would be pretty damn close to A9X @ GFXBench in equal conditions if your 30% estimate is real.

*Ignoring the 86.5 FPS result a Skylake-Y device got @ Manhattan Offscreen, probably an outlier.

Core i3 6100 has the HD530 at up to 1.05GHz
Core i3 6300 has the HD530 at up to 1.15GHz

 

[dahorns]

Adding 30% to the best HD Graphics 515 scores would make A9X only ~10-20% faster @ T-Rex and Manhattan Offscreen, assuming it's twice as fast as A8X. Also there's no 4.5W Core m7 device out right now, I bet these scores come from the 4.5W Core m3 version of Surface Pro 4. Core m7's HD Graphics 515 operates at up to 1GHz, up to ~17.7% higher than the low-end model (same clock as Core i5-6300U's HD Graphics 520), which means it would be pretty damn close to A9X @ GFXBench in equal conditions if your 30% estimate is real.

*Ignoring the 86.5 FPS result a Skylake-Y device got @ Manhattan Offscreen, probably an outlier.

This is all also ignoring the fact that Intel has been much better at 3dmark, both graphics and physics tests.

 

[Sweepr]

This is all also ignoring the fact that Intel has been much better at 3dmark, both graphics and physics tests.

Good point. If that reflects how Apple's SoCs would perform in a PC environment then I don't see them ditching Intel anytime soon.

 

[mikk]

This is all also ignoring the fact that Intel has been much better at 3dmark, both graphics and physics tests.

What is your source?

 

[dahorns]

What is your source?

Uh, like look right above your post for example?

You could also look here: http://www.notebookcheck.net/Intel-HD-Graphics-5300.125576.0.html

You'll see the Core M graphics scores on icestorm are considerably above the Air 2.

also here, http://www.anandtech.com/show/9117/analyzing-intel-core-m-performance/11

Core M models with more than double Air 2's overall Icestorm score.

And to be clear, i was only referring to Core M vs A8x in my statement. I was not talking about Intel's Atom mobile products. The point is, Apple needs the Ipad Pro to double the Air 2's performance in 3dmark just to almost match the last generation Core M parts. I think the end result is Core M is going to stay ahead in 3dmark and be much closer in Gfxbench (running Fp32 vs Fp16) than before.

 

[Space69]

How much difference do you think it'll make?

Usually doubled compute power resulted in ~30% improvement. FP16 vs FP32 should be that. That means at best, HD 520 is equal to iPad Pro graphics.

I have a hard time to follow you logic here. What have 'Usually doubled compute power resulted in ~30% improvement' anything to do with the precission of the fp calculations?

Are you implying that there's a difference of 30% between using f16 and fp32 for a general pixel shader in the iPad Pro? The number of different units, functions and bandwidth reduction does not equal 30% difference in the shaders I look at.

 

[Sweepr]

TechSpot - Intel Z170 Motherboard Roundup

www.techspot.com/review/1073-intel-z170-motherboard-roundup/page7.html

Overclock3D - Intel i3 6300T and 6320 Skylake Review

Once we move from synthetic benchmarks into the avenues of real-world usage it becomes clear that both of these CPUs might be dual-core but they combine beautifully with the rest of our hardware to give surprising responsiveness. When you consider the other options in our graph are all based around the quadcore/octothread i7-6700K, a CPU that's easily more than double the price of either of these two i3s, then the ability of these little dual-core offerings is even more impressive.

Gaming is absolutely the star of the show though. At lower resolutions the GPU is idle enough that a better CPU makes a difference, but once you move into the realms of 1080P with a little anti-aliasing it's obvious that there is nothing to choose between the stock 3.2GHz Core i3-6300T and a 4.8GHz Core i7-6700K. We don't mean "nothing much considering the price", we mean nothing. At all. So if gaming is the main area of use for your home computer, with some light every day tasks mixed in, you'd be foolish to spend more than you have to just so that the specifications read better than the actual results. These Intel Core i3s are both brilliant processors with plenty of performance at an affordable price, and both win our OC3D Value For Money awards.

www.overclock3d.net/reviews/cpu_mainboard/intel_i3_6300t_and_6320_skylake_review/1

Surface Pro 4 - Core m3-6Y30 Benchmarks

http://surfaceproartist.com/blog/20...ce-pro-4-is-the-only-ipad-pro-youll-ever-need

Compared to Surface Pro 3 Core i3 (AnandTech scores) looks like Intel managed to improve both CPU and iGPU performance while moving down from 11.5W to 4.5W TDP. Cinebench R15 MT score is 42% higher, 3DMark scores almost 50% higher. Not bad for the first fanless Surface Pro.

 

[dahorns]

Compared to Surface Pro 3 Core i3 (AnandTech scores) looks like Intel managed to improve both CPU and iGPU performance while moving down from 11.5W to 4.5W TDP. Cinebench R15 MT score is 42% higher, 3DMark scores almost 50% higher. Not bad for the first fanless Surface Pro.

I disagree. Those 3dmark numbers are very disappointing. They barely get to Broadwell Core-M levels. If they are representative, it looks like I was wrong about Core M beating A9x in 3dmark.

 

[Sweepr]

I disagree. Those 3dmark numbers are very disappointing. They barely get to Broadwell Core-M levels. If they are representative, it looks like I was wrong about Core M beating A9x in 3dmark.

3DMark 11 scores are up considerably in the other review I saw. Also the difference between Skylake-Y and Skylake-U is much smaller here.

Arachnotronic has a Core-m3 Surface Pro 4 and will post some benchmarks, let's see if his scores are similar.

 

[pcslookout]

I think the new Surface 4 processor is pretty amazing!

 

[witeken]

Next, let’s take a look at the platform efficiency, and remove the large 70 Wh of capacity from the equation.

While not the outright leaders in efficiency, the Core i5 Surface Book is the class leader when compared against other high DPI devices. The Core i5 Surface Book has even better efficiency than the Surface Pro 4, despite the Surface Pro 4 using an IGZO panel.

Core i5 Book has larger display, more pixels and no IGZO, yet has better efficiency. I guess battery life will always remain a mystery to me :sneaky:.

 

[Sweepr]

Some Core i7-6600U benchmarks:

Nearly 30% faster than the Core i7-3517U (Ivy Bridge ULV) I'm using @ both ST and MT. And almost twice 15W Carrizo's single-thread score.

Impressive boost by SpeedShift here:

www.anandtech.com/show/9767/microsoft-surface-book-2015-review

 

[Sweepr]

On the tablet comparisons, I’ve installed the OpenGL version of GFXBench. Once again the Surface Pro 4 outperforms everything, although in this test the margin is not quite as high. As with 3DMark, on Windows PCs, GFXBench runs at high precision only due to limitations in OpenGL versus OpenGL ES.

Futuremark’s 3DMark is available on all platforms, although when throwing Windows into the mix we always have to take a bit more caution as the level of rendering precision is not always equal. This is due to the fact that lower precision rendering modes - widely available and regularly used on Android and iOS to boost performance and save on power consumption - are not commonly available on Windows PCs, which forces them to use high (full) precision rendering most of the time.

In any case the Core i5-6300U runs away with this test, easily outperforming the other tablets.

I wonder what's the impact of this @ 3DMark and GFXBench scores. That's not to say A9X doesn't sport a very impressive iGPU, but it's something that needs to be clarified before the comparisons pop-up.
Best score for HD Graphics 515 (Core M) @ Windows OpenGL is 86.5 FPS (almost certainly an outlier), other top-scores are in the ~50 FPS range. For those interested, current top-score for 15W Iris Graphics 540 is 91.6 FPS.

 

[witeken]

So A9X runs 16 bit while Gen9 32 bit? Since when does Ax have 16bit support?

 

[kimmel]

So A9X runs 16 bit while Gen9 32 bit? Since when does Ax have 16bit support?

Since always?

 

[ShintaiDK]

Since always?

Pretty much.

IOS+Android+Windows phone=FP16
Windows uses FP32 as default. Not sure what OSX and (PC)Linux uses. Wouldn't be surprised if it was FP32 as well.

 

[AtenRa]

Heh, Skylake delidding tool. :biggrin:

https://www.youtube.com/watch?v=DwH8aVAMYvE

 

[386user]

no info on NUC yet :S

its not the end of the world...maybe the CAD : USD conversion will improve

 

[DrMrLordX]

Pretty much.

IOS+Android+Windows phone=FP16
Windows uses FP32 as default. Not sure what OSX and (PC)Linux uses. Wouldn't be surprised if it was FP32 as well.

I think the bit-ness of your floating point arithmetic will have more to do with the language/compiler environment than anything else. Being a Java luser, I can tell you that the default for Java is 32-bit floating point arithmetic. Doing half-precision IEEE 754 math in Java is . . . tricky?

Most common compiler environments like gcc/g++ etc. are going to default to 32-bit floats

Heh, Skylake delidding tool. :biggrin:

https://www.youtube.com/watch?v=DwH8aVAMYvE

Nice find! I would totally get one if I were going to get a 6700k . . .

 

[frozentundra123456]

Heh, Skylake delidding tool. :biggrin:

https://www.youtube.com/watch?v=DwH8aVAMYvE

Cool, but not sure I would risk my 300.00 cpu for an extra few hundred mhz. How much more can you get with delidding? maybe 300 mhz?