Intel Skylake / Kaby Lake

[Shivansps]

Intel still calling 2H 17 for CoffeLake... i wonder if they will deliver.

 

[DrMrLordX]

Intel still calling 2H 17 for CoffeLake... i wonder if they will deliver.

Where? Everything I have seen says CFL-H and S will show in in Q1 2018 at the earliest.

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...in-h2-2017-icelake-in-2018-digitimes.2485480/

Unless that shows Dec. 2017? And some have put them as late as Q3 2018 for Back to School:

Though honestly I would be surprised if they waited THAT late.

 

[cbn]

For cheap and affordable horsepower necessary to drive a GTX 1050 Ti 4GB, I think the following 27.5L Pre-built mATX machine (with a drive layout similar to a Fractal design Core 1000) makes a good baseline for other comparisons, both DIY and pre-built:

Acer with Core i5 6400, 1 x 8GB DDR3L 1600, 2TB HD and Windows 10 Home for $399.99 free shipping:

http://www.newegg.com/Product/Product.aspx?Item=N82E16883101452

Pretty amazing value IMO. These OEMs must get Core i5 6400 on the cheap.

The Kabylake version of the above PC arrived at Newegg:

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/the-pre-built-atx-tower-hot-deals-thread.2492098/#post-38820090

(DDR3L 1600 is now gone and replaced by DDR4 2400)

 

[jpiniero]

Intel still calling 2H 17 for CoffeLake... i wonder if they will deliver.

Again, Intel has not officially confirmed Coffee Lake. What Intel is calling 8th Gen is the Kaby Lake Refresh which is the QC 15 (and maybe 28) W models.

 

[Shivansps]

Where? Everything I have seen says CFL-H and S will show in in Q1 2018 at the earliest.

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...in-h2-2017-icelake-in-2018-digitimes.2485480/

Unless that shows Dec. 2017? And some have put them as late as Q3 2018 for Back to School:

Though honestly I would be surprised if they waited THAT late.

https://newsroom.intel.com/newsroom...3/Murthy-Renduchintala-2017-Manufacturing.pdf

oh i see, they are talking about mobile U CPUs.

 

[raghu78]

mikk 14nm+ clocks very well but you have to realize it took 2 years to get to 14nm+ from 14nm. 14 nm++ will take 3 years from 14nm launch. 10nm in in a worse shape than 14nm at the same point in its lifecycle. Intel completely avoided talking about 10nm yields even for SRAM. Right now there is not much confidence in Intel's 10nm process execution.

 

[Arachnotronic]

Arachnotronic Kabylake is already clocking 5 Ghz . Coffeelake should do better. Do you think Coffeelake on 14nm++ can be beaten by Icelake on 10nm+ for max single thread performance. First of all we have to wait and see how close to 5 Ghz can Icelake get to. If they can't do 4.7-4.8 Ghz consistently on Icelake then I think their chances against a 5.2 Ghz Coffeelake (I am assuming only 200 Mhz higher OC for Coffeelake which could be higher) are not so great. 14nm++ has a 26% higher transistor performance over 14nm and roughly 10-12% higher performance than 14nm+. The first gen Skylake maxed out at 4.7-4.8 Ghz.

http://www.anandtech.com/show/9533/intel-i7-6700k-overclocking-4-8-ghz
https://www.hardocp.com/article/2015/08/05/intel_skylake_core_i76700k_ipc_overclocking_review/7

I think adding 8-10% IPC over Coffeelake is not going to be easy. These large cores are already damn good so every extra % of IPC increase comes at a much higher and non linear power increase. Coffeelake could turn out extremely difficult to beat for single thread performance even by Icelake.

Absolutely amazing. You spend the last several months telling us how Intel has gotten lazy, isn't innovating, is poised to be crushed by AMD, etc. because they aren't pushing IPC quickly and they're going to get lapped in process technology.

And now, you are praising 14nm++'s performance gains and saying that they're so good that an all new architecture on a process (10nm+) that sees a slight regression in drive currents (but lower dynamic capacitance) simply won't be able to compete on per core performance.

I am confused.

 

[raghu78]

Arachnotronic i stand by my statement that Intel has to stop this tick tock process optimization crap. Intel should be able to bring both IPC and clock speed increases going forward every year. If Apple and AMD can why can't Intel ? Having said that Intel's 14nm+ and 14nm++ process are showing that bleeding edge processe node shrinks are going to be more and more difficult going forward. So the process needs to be improved every year as process maturity and yield improves.

btw I never said Icelake on 10nm+ cannot compete on per core performance with Coffeelake on 14nm ++. I said its difficult to see Icelake beating Coffeelake convincingly for max single thread performance. Intel could get IPC up by 5-10% but the problem is I think Coffeelake will clock atleast 8-10% higher in a max clock comparison vs Icelake. eg: 4.7 Ghz for Icelake vs 5.2 Ghz for Coffeelake.

As for Intel being lazy and complacent well lets just say I will let time tell the truth .

 

[Azuma Hazuki]

Arac...look closer. What he's saying there isn't "oh wow 14nm+++ is so awesome it's as good as 10nm." It's "Uh...WTF happened to 10nm? At this point it's such a disaster they may as well just skip it."

 

[Arachnotronic]

Arac...look closer. What he's saying there isn't "oh wow 14nm+++ is so awesome it's as good as 10nm." It's "Uh...WTF happened to 10nm? At this point it's such a disaster they may as well just skip it."

10nm is pretty much being skipped -- the only product(s) that will use it are CNL-Y and maybe a dual core CNL-U (though I bet you that's been cancelled, too). Cannon Lake-H/S were killed and replaced by CFL-H/S.

10nm+ is worth using, though. Although 14nm++ performance is obviously better, 10nm+ has an advantage in both dynamic capacitance (lower means lower power) and obviously density (which allows Intel to cram more stuff into a given area, which can help chip level performance and obviously if you want to shove more crap in like dedicated accelerators, more CPU/GPU cores, etc. the density improvement certainly helps as long as your yields are good -- if they're not by 2H 2018 then Intel really screwed up).

10nm, given the yields right now (they're clearly bad) and the way lower performance compared to 14nm++ is obviously a dud for anything that's not a teensy low power Y-series part (and TBH this part is only going to see the light of day so Intel can claim that they went into production on 10nm in 2017, it's really pretty pointless), but I think you can make a reasonable case for migrating to 10nm+ from 14nm++ as long as you're doing interesting stuff on the architecture side to get more performance/integration/features.

 

[VirtualLarry]

I vote for a 10nm+ unlocked desktop i3, for $74. C'mon Intel, you know that you want to.

 

[mikk]

10nm is pretty much being skipped -- the only product(s) that will use it are CNL-Y and maybe a dual core CNL-U (though I bet you that's been cancelled, too). Cannon Lake-H/S were killed and replaced by CFL-H/S.

CNL-U should come, there is a Sisoft rank entry from last month: http://ranker.sisoftware.net/show_s...f3ceffd9b18cb99fe7daebcda8cdf0c0e695a899&l=en

 

[Sweepr]

Good thing is, 10nm+ comes into HVM in 2018, shortly after 10nm.

Can we talk abouth EMIB though? Intel's future CPUs (maybe as soon as Ice Lake) will mix and match different nodes in the same product: CPU and graphics portion on the leading edge process, comms and I/O on (N-1) process. Should make it easier to create different SKUs (the perfect one for each segment) and help with yields. I bet Iris Pro is coming back.

 

[raghu78]

Intel has confirmed they will go server first at 10nm++. Have they confirmed if EMIB is introduced at 10nm++ or 7nm. The problem for Intel is EMIB products are atleast 3 years away. In the meanwhile AMD is already pushing multi die MCM designs into servers. I think at 7nm AMD could have even better approaches than MCM like interposer based. I think AMD could make use of HBM as L4 cache as core counts rocket up. Increasing bandwidth by adding proportional number of memory channels might not be a feasible option due to increased board complexity, higher power draw and increased cooling costs.If AMD goes for a doubling of core counts at 7nm as 7nm brings a >2x increase in transistor density you are talking about a lot of cores per socket. Eventually I think a HBM based L4 cache per die will become a standard feature in a few years.

 

[coercitiv]

Eventually I think a HBM based L4 cache per die will become a standard feature in a few years.

Do we have any concrete info on how HBM latency (current or potential) measures up against DDR4? All the info I've been able to find only mentioned better latency (due to HBM layout), not significantly better. (as in a factor of X, whatever X may be)

This may be a little too OT, but hopefully most people interested in future Intel products would want to know about this as well. Any link/info would be greatly appreciated.

 

[frozentundra123456]

Arachnotronic i stand by my statement that Intel has to stop this tick tock process optimization crap. Intel should be able to bring both IPC and clock speed increases going forward every year. If Apple and AMD can why can't Intel ? Having said that Intel's 14nm+ and 14nm++ process are showing that bleeding edge processe node shrinks are going to be more and more difficult going forward. So the process needs to be improved every year as process maturity and yield improves.

btw I never said Icelake on 10nm+ cannot compete on per core performance with Coffeelake on 14nm ++. I said its difficult to see Icelake beating Coffeelake convincingly for max single thread performance. Intel could get IPC up by 5-10% but the problem is I think Coffeelake will clock atleast 8-10% higher in a max clock comparison vs Icelake. eg: 4.7 Ghz for Icelake vs 5.2 Ghz for Coffeelake.

As for Intel being lazy and complacent well lets just say I will let time tell the truth .

As for AMD, yes, they made one very impressive leap, from the construction cores to Zen. And even then, they made a huge improvement in IPC and power usage at a sacrifice in clockspeed. Otherwise, it was incremental improvements from bad to less bad. Hardly something I would tout as an accomplishment, and much easier to improve from a low level than an already very high one. Now I agree, intel is not executing well now, but with the exception of a one-time impressive improvement to Zen, I hardly think AMD is a shining example of year to year progress.

 

[raghu78]

frozen even when AMD had a horrible architecture with Bulldozer they had decent updates every 12-18 months with Piledriver, Steamroller and Excavator. I think AMD will keep pushing the Zen core at the same rate with reasonably good IPC improvements of 5-10% every 12 months. I think will see Zen 2 based Pinnacle Ridge in Q1 2018 on an improved 14LPP process and a GF 7nm based Zen 3 in H1 2019. It will be much easier to make significant improvements for the first couple of generations with a brand new architecture especially with such a unique CCX design. Improving memory support for high speed DDR4, improving fabric speeds, improving cache performance are some of the most obvious areas for Zen improvements.

 

[Arachnotronic]

CNL-U should come, there is a Sisoft rank entry from last month: http://ranker.sisoftware.net/show_s...f3ceffd9b18cb99fe7daebcda8cdf0c0e695a899&l=en

This chip looks like a dud, Kaby Lake Refresh-U is a better product. I think CNL-U will not see the light of day.

 

[jpiniero]

This chip looks like a dud, Kaby Lake Refresh-U is a better product. I think CNL-U will not see the light of day.

Is KBL-R really on 14 nm++? My guess is that it is not.

 

[Arachnotronic]

Is KBL-R really on 14 nm++? My guess is that it is not.

It's 14nm++.

 

[Arachnotronic]

I vote for a 10nm+ unlocked desktop i3, for $74. C'mon Intel, you know that you want to.

They're not going to do that, that would be a 50% price cut or so from the current unlocked i3s.

But whatever Intel does sell for $74 at the 10nm+ generation should be quite nice.

 

[Azuma Hazuki]

So...what happens from here? If 10nm on silicon is a dud, this is going to bring the entire industry to a screeching halt, aside from low-power stuff like phone SoCs. We have a problem; we need to go to III-V materials, EUV, or both, and we need to do it yesterday...

 

[Bouowmx]

What I gathered is for Intel, first outing of 10-nm, which according to Intel estimates has lower performance than 14 nm ++, will be limited, but in the time it will take to develop the second-outing, 10 nm +, which has similar performance to 14 nm ++, most Intel products can migrate to 10 nm +. So, Intel 10 nm being a dud is more about its initial products.

 

[coercitiv]

They're not going to do that, that would be a 50% price cut or so from the current unlocked i3s.

I think by now Intel owes Virtual Larry a LEDT line.

 

[Arachnotronic]

So...what happens from here? If 10nm on silicon is a dud, this is going to bring the entire industry to a screeching halt, aside from low-power stuff like phone SoCs. We have a problem; we need to go to III-V materials, EUV, or both, and we need to do it yesterday...

What's going to happen from an Intel POV, AFAICT, is that the 8th gen Core chips will be mainly on 14nm++ and then 9th gen Core will move to 10nm+ (Ice Lake), and the architectural improvements + power reductions on the process will lead to a better product family than 8th gen Core. Then when 10nm++ comes along, Intel should have the benefit of the much better architecture introduced at 10nm+ but with transistors that are unequivocally better than the 14nm++ stuff, so we get a nice frequency boost.