Intel Skylake / Kaby Lake

Page 303 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
New graphics drivers out, apparently improves DX12 performance. Can someone test Time Spy before and after?

Intel releases new HD Graphics package

Chipzilla has released a new version of its HD Graphics package.

Dubbed the 20.19.15.4501 driver, the software adds Unify colours for different panels and support for 5K3K panel. It also provides content protection for Overlay and non-Overlay drivers.

The 6th-gen camera pipe gets Windows 7 support and will allow for three DVI/HDMI displays on 6th-gen processors and x2 DP mode for type-C systems. Other improvements include a new LACE(Local Adaptive Contrast Enhancement) feature, because you can never have too much lace.

Intel said that the update improves the DX12 performance on some benchmarks. It also selects outputs to be active when more compatible displays are available, and synchronizes the operating system and driver gamma values.

http://www.fudzilla.com/news/graphics/41565-intel-releases-new-hd-graphics-package


Guess who made an appearance on Zauba:

Xeon E5-26xx v5 - Skylake Server

Cr1z87EUEAA2pZe%201_zpsnsd1jmbw.jpg


www.zauba.com/import-skylake-server-hs-code.html

28C/56T + 38.5 MB cache + 165W TDP. :eek:
 
Last edited:

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136

Bu

But even then this has no meaning and is just used out of context to give a bad impression, i mean the max turbo frequency is 3.6GHz, if it needed 35W for a single core at 3.6 then it would need 140W for 4 cores at 3.6GHz, wich is of course ridiculous but that s the kind of things you re implying you and these viral sites by sticking a ST score whith a TDP that is for all cores loaded at 3.2GHz..

FTR at 3.6GHz and a single core it should be within 15W, so the 93 score is also at this TDP, but for sure that it wasnt your intent to highlight the things that way, quite the contrary..
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Bu
But even then this has no meaning and is just used out of context to give a bad impression...
FTR at 3.6GHz and a single core it should be within 15W, so the 93 score is also at this TDP, but for sure that it wasnt your intent to highlight the things that way, quite the contrary..

So now that you've recognized this piece of info actually comes from AMD (another lie exposed) you're back to shifting the goalposts. According to you AMD is downplaying their own chip by telling investors that this score was achieved at 35W TDP when it actually was using only ~15W? Very hard to take any of your posts seriously without rolling my eyes our laughing at this point.

I shouldn't bother, but here's what Bristol Ridge actually scores at 15W TDP inside actual notebooks (hint: nowhere near 93 pts):

- Envy x360 15z m6-ar004dx (FX-9800P): 74 pts
http://www.notebookcheck.com/AMD-FX-9800P-APU-Bristol-Ridge-Benchmarks-und-Specs.173391.0.html

- HP Pavilion 15-aw004ng (A10-9600P): 77 pts
http://www.notebookcheck.net/Bristo...0-9600P-Against-the-Competition.168477.0.html

Can you stop polluting this thread with more nonsense now?
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Based on the FX-9800P score in Cinebench R15, the STAMP boost is not available. FX-8800P scores 82 points with the default configuration (15W TDP, 25W STAMP boost). When STAMP is disabled (i.e true 15W) on FX-8800P it scores exactly the same, 74 points. With default settings STAMP will allow the chip with 15W TDP to boost to 25W for 200 seconds, which is almost enough for Cinebench R15 to complete (at 3400MHz). 25W should be sufficient to keep both the 9800P and 9600P at their maximum advertized boost speed in single threaded mode, but 15W is definitely not. I just did a new DAR with the default configuration (STAPM enabled) and with STAPM disabled. I will plot them to excel and upload the pictures once I'm home.
 

SAAA

Senior member
May 14, 2014
541
126
116
28C/56T + 38.5 MB cache + 165W TDP. :eek:

Soo... any idea on why Skylake server has so little L3 per core? Skylake-X too was leaked as 13.75, so just 1.375MB per core rather than 2.5 as all the other server generations?
I hope this means L4/eDRAM onboard, or much faster caches (less probable, also cutting max capacity by half doesn't sound like a great idea). Anyway it's a further proof that Skylake cores are different from the desktop variant this time, not just for some instructions.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Soo... any idea on why Skylake server has so little L3 per core? Skylake-X too was leaked as 13.75, so just 1.375MB per core rather than 2.5 as all the other server generations?
I hope this means L4/eDRAM onboard, or much faster caches (less probable, also cutting max capacity by half doesn't sound like a great idea). Anyway it's a further proof that Skylake cores are different from the desktop variant this time, not just for some instructions.

It's probably a lower latency cache. Remember larger cache = higher latency and more power consumption.

Edit: Another possibility: mesh interconnect for L3 instead of double ring, would need less cache as well.

Curiously this matches the Skylake-X leak, same 1.375MB/core!

Intel-New-X-family-e1468996249763.jpg
 
Last edited:
  • Like
Reactions: Arachnotronic

railven

Diamond Member
Mar 25, 2010
6,604
561
126
Hey guys, hey, not sure if the proper thread but both are in the name so figured I'd ask here.

I just passed Go and collected money, itching to upgrade my CPU/MoBo/RAM.

Currently:
i7 4970K + MSI Gaming 5 + 16GB DDR3 2133

Can't decide:
Snag a Skylake or Broadwell-E or just wait to see what Kabylake brings to the table (even if just price reductions on the former). Got the itch, so not sure I can wait for Kabylake (Google has no solid release date outside of OEM/Laptop models sometime within the next few weeks/months.

Just got the itch and I needs to scratch it.
 

SAAA

Senior member
May 14, 2014
541
126
116
It's probably a lower latency cache. Remember larger cache = higher latency and more power consumption.

Curiously this matches the Skylake-X leak, same 1.375MB/core!

Indeed I noticed that, yet I'm still expecting some other change... I mean Power added large memories pools long before Intel, also Xeon Phi uses HMC too, wouldn't that be awesome on all servers?

Hey guys, hey, not sure if the proper thread but both are in the name so figured I'd ask here.

I just passed Go and collected money, itching to upgrade my CPU/MoBo/RAM.

Currently:
i7 4970K + MSI Gaming 5 + 16GB DDR3 2133

Can't decide:
Snag a Skylake or Broadwell-E or just wait to see what Kabylake brings to the table (even if just price reductions on the former). Got the itch, so not sure I can wait for Kabylake (Google has no solid release date outside of OEM/Laptop models sometime within the next few weeks/months.

Just got the itch and I needs to scratch it.

Kabylake will be a clockspeed boost at best also it's coming January (7700K) or later, Skylake too won't be a huge bump over Broadwell and will be available next year around summer so if you need more than 4 cores just get them now.
You already have a Haswell at 4.4GHz if you don't OC so you'll need about 4.2GHz for Broadwell to reach single thread parity, the big increase is only on multi-thread.
Otherwise just wait for Kabylake-X (released some time after 7700K, probably along server Skylake parts) and if that can reach 4.8-5GHz you'll have a decent boost in single thread, think 20-30% if you add up clocks and IPC. I wouldn't hold my breath for Cannon-lake or Coffe-lake at that point, Skylake class IPC+ 5GHz will be awesome for the next 5 years just like Sandy was. ;)
 
Mar 10, 2006
11,715
2,012
126
Hey guys, hey, not sure if the proper thread but both are in the name so figured I'd ask here.

I just passed Go and collected money, itching to upgrade my CPU/MoBo/RAM.

Currently:
i7 4970K + MSI Gaming 5 + 16GB DDR3 2133

Can't decide:
Snag a Skylake or Broadwell-E or just wait to see what Kabylake brings to the table (even if just price reductions on the former). Got the itch, so not sure I can wait for Kabylake (Google has no solid release date outside of OEM/Laptop models sometime within the next few weeks/months.

Just got the itch and I needs to scratch it.

Kabylake-S is coming in January. It is a slightly higher clocked version of Skylake-S built on a slightly better process.

Broadwell-E gives you moar cores :)
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Based on the FX-9800P score in Cinebench R15, the STAMP boost is not available. FX-8800P scores 82 points with the default configuration (15W TDP, 25W STAMP boost). When STAMP is disabled (i.e true 15W) on FX-8800P it scores exactly the same, 74 points. With default settings STAMP will allow the chip with 15W TDP to boost to 25W for 200 seconds, which is almost enough for Cinebench R15 to complete (at 3400MHz). 25W should be sufficient to keep both the 9800P and 9600P at their maximum advertized boost speed in single threaded mode, but 15W is definitely not. I just did a new DAR with the default configuration (STAPM enabled) and with STAPM disabled. I will plot them to excel and upload the pictures once I'm home.

Factory config (15W TDP, STAPM = 200 seconds, 25W)
Average frequencies = 3400MHz (prior 200s mark), 3255MHz (past 200s mark), 3359MHz (combined). The STAMP boost doesn't simply switch on or off. Once it is saturated it starts to taper down, instead of switching straight back to the base limit.

wTAIFyJ.png


True 15W config (15W TDP, STAPM = Off)
Average frequency: 2976MHz.

H9jBJOz.png
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Tried to check the bios on the ENVY x360 to see if they did disable STAPM intentionally. HP uses Insyde bioses which are RSA encrypted, so no luck there. Anyway, each and every FP4 part should support and have STAPM enabled at default. The ENVY x360 looks pretty slim, so I guess HP intentionally disabled it during the AGESA build in order to improve the thermals and battery life o_O
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
Soo... any idea on why Skylake server has so little L3 per core? Skylake-X too was leaked as 13.75, so just 1.375MB per core rather than 2.5 as all the other server generations?
I hope this means L4/eDRAM onboard, or much faster caches (less probable, also cutting max capacity by half doesn't sound like a great idea). Anyway it's a further proof that Skylake cores are different from the desktop variant this time, not just for some instructions.

Wow, that's interesting. I had figured that they had cut the L3 on Skylake-X to save power to boost higher; maybe that is the same deal on Skylake Server. But very surprising since server apps tend to take advantage of the extra cache. It of course should save a ton of die space, esp since I bet the SS core is much bigger than Skylake mainstream.
 

railven

Diamond Member
Mar 25, 2010
6,604
561
126
Indeed I noticed that, yet I'm still expecting some other change... I mean Power added large memories pools long before Intel, also Xeon Phi uses HMC too, wouldn't that be awesome on all servers?



Kabylake will be a clockspeed boost at best also it's coming January (7700K) or later, Skylake too won't be a huge bump over Broadwell and will be available next year around summer so if you need more than 4 cores just get them now.
You already have a Haswell at 4.4GHz if you don't OC so you'll need about 4.2GHz for Broadwell to reach single thread parity, the big increase is only on multi-thread.
Otherwise just wait for Kabylake-X (released some time after 7700K, probably along server Skylake parts) and if that can reach 4.8-5GHz you'll have a decent boost in single thread, think 20-30% if you add up clocks and IPC. I wouldn't hold my breath for Cannon-lake or Coffe-lake at that point, Skylake class IPC+ 5GHz will be awesome for the next 5 years just like Sandy was. ;)

Kabylake-S is coming in January. It is a slightly higher clocked version of Skylake-S built on a slightly better process.

Broadwell-E gives you moar cores :)

Holy cow, I thought Kaby Lake was closer to release. Also, you don't tell a man to wait when he's got the itch. Haha. From my own research, OC'ing my Devils' Canyon will most likely cover me, but that doesn't satisfy the desire to break down my rig, open some new boxes, and savor that new toy smell.

How is Broadwell-E OCing? Maybe I just should throw a few out of pocket dollars at this and go for the moar cores route.
 

railven

Diamond Member
Mar 25, 2010
6,604
561
126
Well manager just asked me to work a few extra hours for him. I don't need the extra time to do what he's asked me, but I'll gladly come in early and get paid 1.5x for it :D

So I'm joining the HEDT club!

MOAR CORES!!!

Google-Fu says Broadwell-E OCs poorly, but even with mediocre OC it still runs with a Haswell-E at top OC. So 6800K it is.
 

CakeMonster

Golden Member
Nov 22, 2012
1,384
482
136
Holy cow, I thought Kaby Lake was closer to release. Also, you don't tell a man to wait when he's got the itch. Haha.

Hah, same.. And I'm on a 4.5GHz 6700K and I know for a fact I don't need more cores... yet (mostly gaming).

I have the money and I'm bored enough to probably spend them, yet the KL-K didn't arrive on schedule. I'm probably in a shrinking demographic though or Intel would have put something out and taken my money.

I'll only take something above 4 cores (mainstream or HEDT, whatever) when and if:
1) there's no clock speed/IPC penalty because of heat/TDP.
2) the platform/architecture is not lagging behind mainstream.

I want to give Intel more of my money, yet they keep ignoring my needs.
 
Mar 10, 2006
11,715
2,012
126
Hah, same.. And I'm on a 4.5GHz 6700K and I know for a fact I don't need more cores... yet (mostly gaming).

I have the money and I'm bored enough to probably spend them, yet the KL-K didn't arrive on schedule. I'm probably in a shrinking demographic though or Intel would have put something out and taken my money.

I'll only take something above 4 cores (mainstream or HEDT, whatever) when and if:
1) there's no clock speed/IPC penalty because of heat/TDP.
2) the platform/architecture is not lagging behind mainstream.

I want to give Intel more of my money, yet they keep ignoring my needs.

Skylake-X will be what you want, comes in up to 10 core configs and will use KBL-X PCH, a slightly more featured variant of the KBL-H PCH. SKL-X may actually have better perf/clock than vanilla SKL/KBL and overclocking should be quite robust thanks to the new architecture (SKL inherently hits better frequencies than BDW, for example) and 14nm+ process.

Bad news is you've got to wait a year.
 

mikk

Diamond Member
May 15, 2012
4,112
2,105
136
For out of the box tests Kabylake should be slighty faster (DDR4-2133 vs DDR4-2400) at the same frequency, depends on the test.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Thanks to mikk (above) we now have a more clear picture about Intel's mobile plans for 2017 and H1-2018:

97pc7s9n_zpscd4mcrsj.jpg


efww33ez_zpskq6spubu.jpg


- Apparently Skylake-H GT4e won't get a Kaby Lake successor, but this doesn't rule out a possible refresh (higher clocks) next year
- Kaby Lake-H should arrive this quarter, November/December
- Intel's 10nm debut will be in the form of Cannonlake-Y/U, starting with a 2C+GT2 SKU in Q4-2017
- Coffee Lake is real, and replaces Kaby Lake mobile parts >15W TDP
- Coffee Lake will have quad-cores with Iris Graphics at 15W/28W TDP from the start (we might see 4C/8T 15W Kaby Lake-U before)
- The one many people want to hear: Coffee Lake-H brings mainstream hexa-cores from Intel, it's almost a given that desktops will get a similar SKU
- If Skylake is any indication, Coffee Lake for DT could arrive sooner than mobile - DigiTimes says desktop Coffee Lake is H2-2017
 
Last edited:
Mar 10, 2006
11,715
2,012
126
RIP Iris Pro. Nobody but Apple was interested, and even they are probably going to just stick with dGPUs from now on.
 

Maxima1

Diamond Member
Jan 15, 2013
3,515
756
146
When does Intel release an SKU with HMC? I can't help but laugh looking at this roadmap and the predictions I've seen here.
 

mikk

Diamond Member
May 15, 2012
4,112
2,105
136
When it makes sense. The potential ROI of a product with HMC is not looking great when it's pricier than Intels edram. Intel adds two cores and you are moaning. Despite being on 14nm and two more cores Coffe lake could bring µarch improvements to CPU and GPU, that's what I'm hoping for.