Intel Skylake / Kaby Lake

[kimmel]

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

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

Making sure we are on the same page. This is about the graphics precision. Not the core precision.

For example....
http://imgtec.com/powervr/graphics/series7xt/

Take a look at the table on that page. Executing the "same" code with less precision would most likely give you better benchmarks as you get double the resources effectively.

GT7800 8 256 (FP32) or 512 (FP16)
GT7900 16 512 (FP32) or 1024 (FP16)

 

[ShintaiDK]

Making sure we are on the same page. This is about the graphics precision. Not the core precision.

Exactly.

 

[DrMrLordX]

Making sure we are on the same page. This is about the graphics precision. Not the core precision.

Same thing. It's going to come down to the compiler environment and the target API.

 

[Space69]

Same thing. It's going to come down to the compiler environment and the target API.

Not entirely.

lowp is not necessary fp16 - IOS and Android doesn't guarantee fp16 - it can be between fp10 and fp16 depending on the hardware implementation.

The same goes for highp which can be between fp24 and fp32 depending on the hardware.

Most compilers/drivers on mobile devices can force highp to lowp if its related to color information eventhou you specify high precission. The reason for this is obvious:

* Lower power consumption
* Higher performance (more fp10-fp16 units)
* Bandwidth reduction

This is different on PC (until now) where all the calculations happens in fp32 space - you can still use lower precission for textures to reduce bandwidth, but the calculation still happens at fp32. But Nvidia, Intel and AMD have also seen that people gladly compare fp16 and fp32 performance, so they have started to use the same approach as mobile. We're back at 2002 regarding quality.

 

[Sweepr]

Now that we know Broadwell-E is going to bump the core count to 10C/20T what do you guys expect from Skylake-E?

In other news, first Skylake-U GT3e Geekbench and GFXBench Car Chase (4.0) scores:

ST: 3361
MT: 7010

https://browser.primatelabs.com/geekbench3/4164598

35.1 FPS @ Car Chase Offscreen

 

[frozentundra123456]

Now that we know Broadwell-E is going to bump the core count to 10C/20T what do you guys expect from Skylake-E?

In other news, first Skylake-U GT3e Geekbench and GFXBench Car Chase (4.0) scores:

ST: 3361
MT: 7010

https://browser.primatelabs.com/geekbench3/4164598

35.1 FPS @ Car Chase Offscreen

Hope you are right about BW-E, but do you have a link for that? No offense, but I will believe it when I see it. 10 cores at the same base clock as the previous 8 would be sweet though. Any idea of the release date?

 

[Sweepr]

Hope you are right about BW-E, but do you have a link for that? No offense, but I will believe it when I see it. 10 cores at the same base clock as the previous 8 would be sweet though. Any idea of the release date?

Originally posted by dooon here.
Broadwell-EP is Q1-2016, probably not far from it.

 

[dark zero]

Waiting for C or R based Skylake i7 machines. It will show up the real power of Iris...

BTW, no i5 HQ based laptops around there?

 

[frozentundra123456]

Well, maybe all the hype about Zen in legit and Intel is making a pre-emptive strike!!!

But I still am waiting for more confirmation. Weren't there rumors that BW-E was going to be cancelled? If true, this is great, but I think I still would rather have six cores on the mainstream platform than 10 cores on HEDT. At least BW-E will only be one generation behind, not 2 like HW-E.

 

[dark zero]

Seems that Intel agents knows even more than the fanboys themselves and they are ready to screw the Dual Cores and finally move to Quads

 

[witeken]

Well, maybe all the hype about Zen in legit and Intel is making a pre-emptive strike!!!

No, the desktop devision is just trying more or less deperately not to let revenues continue falling (sharply) Y/Y.

 

[frozentundra123456]

No, the desktop devision is just trying more or less deperately not to let revenues continue falling (sharply) Y/Y.

It was said in jest. In any case, if intel really wanted to pump some life into the desktop division, I think a mainstream hex core would generate more interest and favorable publicity than focusing all the attention on the niche HEDT platform.

Heck, if it was on the mainstream, we might even see OEM hex cores. But I dont really want to get into this, because I know any mention of a mainstream hex core is going to instigate a war of words with a certain poster.

 

[Arachnotronic]

It was said in jest. In any case, if intel really wanted to pump some life into the desktop division, I think a mainstream hex core would generate more interest and favorable publicity than focusing all the attention on the niche HEDT platform.

Heck, if it was on the mainstream, we might even see OEM hex cores. But I dont really want to get into this, because I know any mention of a mainstream hex core is going to instigate a war of words with a certain poster.

There is evidence in the public domain to suggest that Intel will be moving to octa-cores on the mainstream with Cannonlake.

 

[Dufus]

Speed Shift just raises the bar even higher in the mobile space. Becoming another essentials.

Err woot!

Haswell can already be hardware controlled to run cores at maximum frequency from idle C-States without OS intervention or P-States, and do it in less than 1ms.

 

[Deders]

Err woot!

Haswell can already be hardware controlled to run cores at maximum frequency from idle C-States without OS intervention or P-States, and do it in less than 1ms.

Can it? Where have you seen this? Most places are saying 30ms.

 

[Brunnis]

*Wrong thread*

 

[nvgpu]

http://www.intel.com/content/www/us/en/nuc/nuc-kit-nuc6i5syh.html

http://www.intel.com/content/www/us/en/nuc/nuc-kit-nuc6i3syh.html

Skylake NUCs are official now.

 

[Arachnotronic]

http://www.intel.com/content/www/us/en/nuc/nuc-kit-nuc6i5syh.html

http://www.intel.com/content/www/us/en/nuc/nuc-kit-nuc6i3syh.html

Skylake NUCs are official now.

Can you buy them anywhere, perchance? I know a family member who has wanted to buy a NUC but I advised them to wait for Skylake.

 

[ShintaiDK]

http://www.intel.com/content/www/us/en/nuc/nuc-kit-nuc6i5syh.html

http://www.intel.com/content/www/us/en/nuc/nuc-kit-nuc6i3syh.html

Skylake NUCs are official now.

My HTPC! Tho the one without 2½" support.

Can you buy them anywhere, perchance? I know a family member who has wanted to buy a NUC but I advised them to wait for Skylake.

They are listed many places. But not in stock yet. Maybe next week.

 

[Sweepr]

NotebookCheck: Surface Pro 4 (Core m3) Review

Cinebench 11.5
ST: 0.98
MT: 2.27

Cinebench 15
ST: 88
MT: 206

Octane v1: 19970
*iPad Air 2: 10084

Sunspider 1.0 (lower is better): 122.4ms
*iPad Pro: 189.7ms

PCMark 8 Home Accelerated: 2426

3DMark 11 Performance: 1344

Nice bump from first-gen Core-M in most benchmarks and solid long-term performance.

Whereas the SP3 featured capable integrated graphics hamstrung by a tendency to throttle under load, the SP4's Intel HD Graphics 515 seems to be more competent in this realm, and with a far lower-power chipset, to boot. In 3DMark 2013 Ice Storm, we achieved a very good score of 42046, which is a little below the XPS 13-9343’s 46293 (+10%) and the Surface Book’s 49514 (+18%), but which easily beats the SP3’s handicapped result of 29229 by 30%.

Under full CPU stress, the CPU amazingly maintains a 2 GHz turbo frequency, with maximum temperatures just touching 51 °C—impressive to say the least, and surely part of the reason for the SP4’s dominance over the SP3’s CPU benchmarks. Full GPU stress, meanwhile, produces initial clock rates of 798 MHz, but eventually this falls to between 648 MHZ and 698 MHz with a maximum recorded temperature of 62 °C.

Combined stress obviously exacts a much direr scenario for the TDP- and thermally-constrained tablet, with the CPU dropping down to just 1 GHz and the GPU managing only 599 MHz to 648 MHz with occasional throttles to 149 MHz. Temperatures are once again hovering around the 62 °C mark, which seems to be the effective ceiling for the Surface Pro 4.

Best of all, the arguably substantial benefits of lower power consumption, silent operation, likely superior battery life, and a complete lack of moving parts (which are prone to failure) come with very few practical compromises for the average user. Yes, CPU performance is below that of the Surface Pro 3 Core i5 configuration we previously reviewed—but overall, real-world system performance is probably closer to just around 20% lower (with no incidence of throttling under practical use), which is of limited concern considering how quick everything still feels.

www.notebookcheck.net/Microsoft-Surface-Pro-4-Core-m3-Tablet-Review.153843.0.html

 

[phillyman36]

Forgive me for not knowing but the Skylake nucs seem to have the Intel HD520 graphics. What happened to the Iris Graphics 540 and 550?

 

[Sweepr]

Forgive me for not knowing but the Skylake nucs seem to have the Intel HD520 graphics. What happened to the Iris Graphics 540 and 550?

Only the Core i3 version. NUC6i5SYH is based on Core i5-6260U (Iris Graphics 540) which is 15W Skylake-U GT3e.
BGA 35W/45W Skylake-H GT4e is coming Q1-2016.

 

[dahorns]

Some m3-6y30 benches on the SP4 I just bought:

pre-speedshift

Google Octane (Edge) = ~18,000

Post-speedshift

Google Octane (Edge) = ~22,000
Google Octane (Chrome) = ~22,000

Kraken 1.1 (Edge) = 1,591 ms
Kraken 1.1 (Chrome) = 1,682 ms

Sunspider 1.0.2 (Edge) = 144 ms
Sunspider 1.0.2 (Chrome) = 306.2

GeekBench 3 (32 bit)

Single Core = 2,450
Multi Core = 4,600​

WebXprt 2013 - 1,414 +/- 36

CB R15

OpenGL - 33.72
CPU - 212​

3dmark Ice Storm 1.2

Overall 41,903
Graphics 50,572
Physics 26,190​

GfxBench

Manhattan 3.0, 1080p offscreen - 43.45 FPS
T-Rex 1080p offscreen - 95.9 FPS
Car Chase, 1080p offscreen - 230 fps​

Notebookcheck also has a review up: http://www.notebookcheck.net/Microsoft-Surface-Pro-4-Core-m3-Tablet-Review.153843.0.html

 

[Sweepr]

Some m3-6y30 benches on the SP4 I just bought:

pre-speedshift

Google Octane (Edge) = ~18,000

Post-speedshift

Google Octane (Edge) = ~22,000
Google Octane (Chrome) = ~22,000 (higher is better)
*iPad Pro (Safari) = 20,062

Kraken 1.1 (Edge) = 1,591 ms
Kraken 1.1 (Chrome) = 1,682 ms
*iPad Pro (Safari) = 1,485 ms (lower is better)

Sunspider 1.0.2 (Edge) = 144 ms (lower is better)
Sunspider 1.0.2 (Chrome) = 306.2
*iPad Pro (Safari) = 189.7 ms

WebXprt 2013 - 1,414 +/- 36 (higher is better)
*iPad Pro - 1,007

Notebookcheck also has a review up: http://www.notebookcheck.net/Microsoft-Surface-Pro-4-Core-m3-Tablet-Review.153843.0.html

Those are some respectable numbers for a fanless device, thanks for sharing dahorns. NBC mentioned no throttling under 'practical use', the only significant throttling happened when their ran CPU+iGPU stress tests simultaneously. Is that true from what you've experienced so far?

Added iPad Pro numbers from AnandTech and ArsTechnica just for fun.

Also from NotebookCheck, here's the first 4C/4T Core i5-6300HQ notebook review. Finally some cheaper Intel quad-core notebooks.

 

[dark zero]

Some m3-6y30 benches on the SP4 I just bought:

pre-speedshift

Google Octane (Edge) = ~18,000

Post-speedshift

Google Octane (Edge) = ~22,000
Google Octane (Chrome) = ~22,000

Kraken 1.1 (Edge) = 1,591 ms
Kraken 1.1 (Chrome) = 1,682 ms

Sunspider 1.0.2 (Edge) = 144 ms
Sunspider 1.0.2 (Chrome) = 306.2

GeekBench 3 (32 bit)

Single Core = 2,450
Multi Core = 4,600​

WebXprt 2013 - 1,414 +/- 36

CB R15

OpenGL - 33.72
CPU - 212​

3dmark Ice Storm 1.2

Overall 41,903
Graphics 50,572
Physics 26,190​

GfxBench

Manhattan 3.0, 1080p offscreen - 43.45 FPS
T-Rex 1080p offscreen - 95.9 FPS
Car Chase, 1080p offscreen - 230 fps​

Notebookcheck also has a review up: http://www.notebookcheck.net/Microsoft-Surface-Pro-4-Core-m3-Tablet-Review.153843.0.html

R.I.P Nvidia Shield tablet.... got outperformed brutally by Intel's solution...