Intel Skylake / Kaby Lake

[DerpEinstein]

Kaby Lake was not supposed to have any ipc gain, though. All this complaining about it seems odd.

I am not sure I follow. It wasn't supposed to be an ipc gain!? Why? Kaby lake shouldn't even exist, if it's not really an improvement. They could have just made a 6790k or something. This is just straight up milking the consumers!

 

[DerpEinstein]

Don't expect those 40% gains every year. AMD might come close to Intel, like they're doing now, but they won't surpass them. (By more than say 10% higher perf/clock.)

Well obviously not. The thing is AMD doesn't need to do it again, because intel has basically hit a dead end. Coffee lake will be the 4th generation on 14nm!

AMD merely has to slightly improve each generation, and they will eventually be on par with intel. Zen has a lot of room for improvement too. Skylake was already the best intel had, and will be for the next 3 years.

 

[LTC8K6]

I am not sure I follow. It wasn't supposed to be an ipc gain!? Why? Kaby lake shouldn't even exist, if it's not really an improvement. They could have just made a 6790k or something. This is just straight up milking the consumers!

You must not have been following KL developments very closely if you expected a new core with better ipc?

 

[DerpEinstein]

You must not have been following KL developments very closely if you expected a new core with better ipc?

Well, it's intel, so I didn't really expect them to do anything, except milk their monopoly.


Trolling, flamebait, and OT are not allowed.
Markfw
Anandtech Moderator

 

[Arachnotronic]

AMD merely has to slightly improve each generation, and they will eventually be on par with intel. Zen has a lot of room for improvement too. Skylake was already the best intel had, and will be for the next 3 years.

Do you expect that the best CPU that Intel will have at the end of 2019 will be a Skylake core based processor?

 

[DerpEinstein]

Do you expect that the best CPU that Intel will have at the end of 2019 will be a Skylake core based processor?

That is exactly what they will do. They will re-re-release Skylake with very minor improvements, and off a 6 core to the mainstream consumers.

 

[Arachnotronic]

Turbo ratios and cache ratio at stock.

 

[Arachnotronic]

That is exactly what they will do. They will re-re-release Skylake with very minor improvements, and off a 6 core to the mainstream consumers.

I think you will be proven wrong. Ice Lake will probably come in late 2018/early 2019 and that will be a new microarchitecture.

 

[AtenRa]

Do you expect that the best CPU that Intel will have at the end of 2019 will be a Skylake core based processor?

I think you will be proven wrong. Ice Lake will probably come in late 2018/early 2019 and that will be a new microarchitecture.

For the Desktop there is only Coffeelake in 2018-2019 and that is Skylake Arch.

 

[coercitiv]

Turbo ratios and cache ratio at stock.

What's the stock voltage?

 

[Sweepr]

Core i7-7700K @ HWBOT

http://hwbot.org/hardware/processor/core_i7_7700k

For the Desktop there is only Coffeelake in 2018-2019 and that is Skylake Arch.

Nonsense.

According to Intel's roadmap, the chipmaker is expected to roll out its 10nm notebook-use Cannon Lake and 14nm desktop-use Coffee Lake CPUs in the second half of 2017 and to bring out Ice Lake CPUs in 2018.

www.digitimes.com/news/a20160904PD201.html

Also Skylake-X might pack the same cores as the server variants, with 4x the amount of L2/core compared to regular Skylake-S/Kaby Lake-S, which could improve performance per clock - Q3-2017 launch.

 

[AtenRa]

According to Intel's roadmap, the chipmaker is expected to roll out its 10nm notebook-use Cannon Lake and 14nm desktop-use Coffee Lake CPUs in the second half of 2017 and to bring out Ice Lake CPUs in 2018.

Ehmm Kabylake-S is Q1 2017 and you actually expect Coffeelake-S in 2H 2017 ?? not happening.

For 2017 we will get Kabylkae in Q1 for mainstream and then for the new 2066 socket the 4C8T Kabylake-X and 6C and up with Skylake-X in Q3 2017. CoffeeLake is 2018 and latest leaks point around Q3 to Q4.

If Intel will release 4C8T Kabylake-X HEDT in H2 2017 they will not release a 6-Core mainstream part in the same time frame (H2 2017).

 

[Sweepr]

Ehmm Kabylake-S is Q1 2017 and you actually expect Coffeelake-S in 2H 2017 ?? not happening.

For 2017 we will get Kabylkae in Q1 for mainstream and then for the new 2066 socket the 4C8T Kabylake-X and 6C and up with Skylake-X in Q3 2017. CoffeeLake is 2018 and latest leaks point around Q3 to Q4.

There's a DigiTimes leak from the same time as the BenchLife article you keep quoting - which says 300-series motherboards in late 2017 (implying Coffee Lake-S a year after Kaby Lake-S). DigiTimes doesn't wait to post any rumor/news regardings these chips, meanwhile your roadmap could be very old and not reflective of Intel's current plans. I read 'roll out' in the DT article quoted above as 'shipping' - so actual Coffee Lake-S launch should happen in early 2018. Even if Ice Lake-S gets pushed back, it's definitely not from 2018 to 2020+ like you're saying, H1-2019 would be more likely.

If Intel will release 4C8T Kabylake-X HEDT in Q3 2017 they will not release a 6-Core mainstream part in the same time frame (H2 2017).

Using your logic they wouldn't release 6C/12T mainstream parts at all, because HEDT already exists? We've seen similar mainstream/HEDT quad-core parts coexisting many times in the past. Also Kaby Lake-X should be a niche product for those who need the absolute best ST performance, while 6C/8C/10C Skylake-X and 6C Coffee Lake-S cover the rest of the enthusiast segment.

 

[LTC8K6]

These are some of the weirdest threads ever.

AMD is apparently in a hopeless no-win scenario against Intel despite Ryzen, yet Intel is apparently incredibly stupid and stuck without any options for the foreseeable future.

 

[AtenRa]

Using your logic they wouldn't release 6C/12T mainstream parts at all, because HEDT already exists? We've seen similar mainstream/HEDT quad-core parts coexisting many times in the past. Also Kaby Lake-X should be a niche product for those who need the absolute best ST performance, while 6C/8C/10C Skylake-X and 6C Coffee Lake-S cover the rest of the enthusiast segment.

You dont release a Quad Core 4C 8T HEDT Kabylake-X in Q3 2017 with a new socket 2066 and at the same time you release a mainstream 6-Core. Coffeelake is definitely 2018, now if its Q1 or Q3 is another thing but it sure isnt 2017.

 

[Arachnotronic]

You dont release a Quad Core 4C 8T HEDT Kabylake-X in Q3 2017 with a new socket 2066 and at the same time you release a mainstream 6-Core. Coffeelake is definitely 2018, now if its Q1 or Q3 is another thing but it sure isnt 2017.

Coffee Lake won't launch in 2017, but I suspect it will ship at the end of 2017 so that it can launch in January/February of 2018.

Look, this thing taped out in November-ish according to a forum member here, and it usually takes about a year from tape-out to qualification for sale. So assuming no "oopsies" Coffee Lake should qualify for sale in November 2017-ish and make it onto shelves in early 2018.

Q3-Q4 2018 launch would mean that somebody dun messed up. And given how "low risk" this project seems to be (Kaby Lake with a few more cores and maybe some more L3$), Intel would have to be real bad at its job to see such a delay.

 

[witeken]

Coffee Lake won't launch in 2017, but I suspect it will ship at the end of 2017 so that it can launch in January/February of 2018.

Look, this thing taped out in November-ish according to a forum member here, and it usually takes about a year from tape-out to qualification for sale. So assuming no "oopsies" Coffee Lake should qualify for sale in November 2017-ish and make it onto shelves in early 2018.

Q3-Q4 2018 launch would mean that somebody dun messed up. And given how "low risk" this project seems to be (Kaby Lake with a few more cores and maybe some more L3$), Intel would have to be real bad at its job to see such a delay.

Exactly that last paragraph. This isn't rocket science anymore. It will be a 2 year old architecture on a 3 year old process technology. It's basically just making a new SKU. The 4-cores will just be rebranded Kaby Lake i5 and i7 as i3 and i5. The 6-core will be new. The GT3e one also.

I doubt the process from tape-out to QFS takes a year. At this stage I wouldn't be surprised if it was quite a bit less. At least if Intel has learned anything from the mobile guys Qualcomm, MediaTek.

 

[jpiniero]

Q3-Q4 2018 launch would mean that somebody dun messed up. And given how "low risk" this project seems to be (Kaby Lake with a few more cores and maybe some more L3$), Intel would have to be real bad at its job to see such a delay.

They did just cut 10% of their workforce, maybe they simply don't have the resources to complete it in a reasonable fashion. Maybe they would push it out because Icelake is showing signs of delays.

I still think it's more Q1-Q2 but you never know.

 

[frozentundra123456]

Kaby Lake was not supposed to have any ipc gain, though. All this complaining about it seems odd.

Yea, some people just cant resist sticking it to intel. It does seem off topic in this thread, since there is another whole thread about it based on some bogus clickbait article that ranted on about how overclocked SK and KL (to the same clocks) found no difference (duh, you think). The whole thing is absurd. They are acting like the only gains that mean anything are IPC improvements. Clockspeed improvements, and the ability to maintain turbo longer dont seem to count somehow. KL is actually a pretty decent gain in mobile, and stock clocks are faster on the desktop as well. One could take issue with calling it "Gen 7", since it doesnt really fit the traditional definition of a new generation (die shrink or new architecture), but ranting on and on about how there were no IPC gains is, for whatever motivation, clearly telling only one side of the story.

 

[mikk]

Ehmm Kabylake-S is Q1 2017 and you actually expect Coffeelake-S in 2H 2017 ?? not happening.

For 2017 we will get Kabylkae in Q1 for mainstream and then for the new 2066 socket the 4C8T Kabylake-X and 6C and up with Skylake-X in Q3 2017. CoffeeLake is 2018 and latest leaks point around Q3 to Q4.

Based on the leaked Roadmaps Coffe Lake-S is scheduled for Q2 2018.


Some news about Kaby Lake:

Intel has decided to not productize Kaby Lake-H 18W processor due to interest shifting to KBL-R. This allows Intel to better focus its engineering efforts there. KBL-R is targeting HR'17 TTM and will be 8th Gen.

https://benchlife.info/intel-cancel-18w-kaby-lake-h-and-use-kaby-lake-r-replace-12262016/


Makes sense. 4+2 H 18W is obsolete considering that Intel brings a more efficient KBL-R U 4+2 15W.

 

[Arachnotronic]

They did just cut 10% of their workforce, maybe they simply don't have the resources to complete it in a reasonable fashion. Maybe they would push it out because Icelake is showing signs of delays.

I still think it's more Q1-Q2 but you never know.

http://files.shareholder.com/downlo...A99-CB302142EB77/Earnings_Release_Q3_2016.pdf

Employee count in Q3 2016 was 105.6 thousand. Employee count the year before was 106.5 thousand.

10% workforce reduction not found

 

[witeken]

http://files.shareholder.com/downlo...A99-CB302142EB77/Earnings_Release_Q3_2016.pdf

Employee count in Q3 2016 was 105.6 thousand. Employee count the year before was 106.5 thousand.

10% workforce reduction not found

Troll.

The number is 1k lower even though it includes all the Altera people and all the acquisitions they've done this year.

 

[krumme]

Exactly that last paragraph. This isn't rocket science anymore. It will be a 2 year old architecture on a 3 year old process technology. It's basically just making a new SKU. The 4-cores will just be rebranded Kaby Lake i5 and i7 as i3 and i5. The 6-core will be new. The GT3e one also.

I doubt the process from tape-out to QFS takes a year. At this stage I wouldn't be surprised if it was quite a bit less. At least if Intel has learned anything from the mobile guys Qualcomm, MediaTek.

Its very limited how much you can reduce time. But i also guess 2 month should be possible.
Zen is here. 10nm is years out and on a new process. Its not a solution.
They simply need the new sku now. Look at i7 vs 8c zen. Serious. Its a mess.
Zen will be a burning platform for Intel and i think we will see radical changes from top management to faster execution. I am not sure Brian K is there end 2017. He looks like a COO to me. Not a CEO. It was just a step to far. His messed up fundraising also showed he is where its difficult for him.
Q4 on market is my bet. There is much more potential in Intel than this sleeping behavior we see now.

 

[frozentundra123456]

Its very limited how much you can reduce time. But i also guess 2 month should be possible.
Zen is here. 10nm is years out and on a new process. Its not a solution.
They simply need the new sku now. Look at i7 vs 8c zen. Serious. Its a mess.
Zen will be a burning platform for Intel and i think we will see radical changes from top management to faster execution. I am not sure Brian K is there end 2017. He looks like a COO to me. Not a CEO. It was just a step to far. His messed up fundraising also showed he is where its difficult for him.
Q4 on market is my bet. There is much more potential in Intel than this sleeping behavior we see now.

When you say i7, I assume you mean quad core. That is not really the competition for Zen, the HEDT platform is. If Zen gives performance between 6800 and 6900k I expect it to be priced at 500.00 or above.

 

[Ajay]

When you say i7, I assume you mean quad core. That is not really the competition for Zen, the HEDT platform is. If Zen gives performance between 6800 and 6900k I expect it to be priced at 500.00 or above.

Seems like Zen (Ryzen) will be a standard PC motherboard with an HEDT processor (What Intel is going with with hexacore Coffee Lake, likely in response to prior knowledge of Zen).
So total platform costs should be somewhere between 8C/16 i7s systems and HEDT systems (approx).