Intel Skylake / Kaby Lake

[Edrick]

Competition affects the SKUs that come to market and the prices that they're offered at. The lead times in the semi industry for product development are too long for the launch of a product to impact fundamental core arch decisions or even the SoC design decisions.

I have always believed that Intel "sits" on skus that are ready for release until market conditions dictate. And with no competition, it may be months/year until they see the light of day. AMD has changed the market conditions. At least that is what I believe.

There is no doubt in my mind that the 12-18C products would have been released in 2018 if not for TR.

 

[Glo.]

Are there any words on 35W Quad Core i3's incoming?

 

[Christopher Bohling]

I have to admit I have been reluctant to even post I bought one. Every where I look people are coming down on these chips. God forbid I didn't buy a Ryzen 1700X. Anyways I have avoided threads on this but you are right. It is my money and so far I am very happy with the chip. Easiest overclocking i have done in a good long time too.

Yarly god forbid you buy a product and are then happy because it does what you want it to do!

 

[Shivansps]

I have always believed that Intel "sits" on skus that are ready for release until market conditions dictate. And with no competition, it may be months/year until they see the light of day. AMD has changed the market conditions. At least that is what I believe.

There is no doubt in my mind that the 12-18C products would have been released in 2018 if not for TR.

Well thats a different matter enterely, timing DID change because of AMD. CFL-S was 2018, and SKL-X around September.

 

[DrMrLordX]

IMHO, a lot of it has to do with Samsung. AMD has always had great chip designers. However, try to design anything competitive when your process tech sucks. Then GF got Samsung's process and their new shiny finfet transistors on 14nm and voila. Lisa Su better not screw it up. She can't blame process or design teams. AMD will sink and swim purely on the ability of her management team to execute.

Don't forget the IBM acquisition. If you don't see its effects by now, you will by the time AMD is on 7nm HP.

Bulldozer & lineage didn't suck because of the process they were built on, they sucked because they were crappy chip architectures.

They kind of did. AMD struggled with process at 32nm, took forever to get it right, and then struggled again at 28nm which took a loooooong time to get right (all the way to GV-A1 arguably). They never got 20nm to work, and had to license tech from Samsung AND enlist the aid of IBM's old engineers to get anything working beyond 28nm (and to respon 32nm and 28nm to not suck).

AMD had no legit answer to Intel's 22nm process where Intel did the most damage to AMD.

I'm really digging my 7800X. Even if that means I will get lots of hate for buying one. lol

Nobody hates you. Most people around here will neither know nor care.

 

[Topweasel]

Competition affects the SKUs that come to market and the prices that they're offered at. The lead times in the semi industry for product development are too long for the launch of a product to impact fundamental core arch decisions or even the SoC design decisions.

But it's not only a new product. For example Intel didn't look at Threadripper and make a new CPU for the 12+ lineup. They had a CPU that worked in the socket that had more cores. Then they announced it and the whole platform early, while actually holding up the shipping of these CPU's because they didn't actually validate it for this platform. Or back when 1GHz was broken. Intel didn't make a new CPU for 1GHz to fight AMD, they rushed through verification their already manufactured CPU to run at 1GHz. They tried to to do it with the 1.13 and that blew up in their faces.

Intel has a very deversified lineup of actual CPU options. So while you are right an Intel can't pop out a new CPU to compete with AMD, Intel can still make new SKUs that they never planned on can be made to battle a product in the short term.

 

[Shivansps]

But it's not only a new product. For example Intel didn't look at Threadripper and make a new CPU for the 12+ lineup. They had a CPU that worked in the socket that had more cores. Then they announced it and the whole platform early, while actually holding up the shipping of these CPU's because they didn't actually validate it for this platform. Or back when 1GHz was broken. Intel didn't make a new CPU for 1GHz to fight AMD, they rushed through verification their already manufactured CPU to run at 1GHz. They tried to to do it with the 1.13 and that blew up in their faces.

Intel has a very deversified lineup of actual CPU options. So while you are right an Intel can't pop out a new CPU to compete with AMD, Intel can still make new SKUs that they never planned on can be made to battle a product in the short term.

Again thats still taking the Xeons (SKL-X are already Xeons, they are all mesh, the original rumor was 10C SKL-X was ring bus LCC and 12C+ Xeons MCC or HCC, thus an emergency addition, wrong they are all server mesh cpus, they are all xeons), validating a 12 to 18C cpu in a socket designed for 10C max, moving the production schedules of SKL-X, CFL-S and Xeon to fit the SKL-X +12C, in the time-schedule that was originally intended for SKL-X in the first place. If they managed to do that is at minimum amazing.

And for what? is not needed, they had the Xeons to fight TR in high end workstations, like it was with BDW-E and before.

Just face it, this rumor died the day "September launch" was know, everything leading up to this rumor turned out to be fake. The only thing you have are the old pre-release leaked slides and they may not be accurate to start with. How many pre-release slides had errors or were compeltely fake before the launch of X product?

 

[Zucker2k]

"Typically a lot of these users are doing prosumer type work; this is time to money," Srivatsa said. "It was clearly indicated to us that if we offered more, people would come back in and refresh faster. That's really where we think the money is."

Edit: Intel has added two more cores to the flagship HEDT lineup on the last few lineups. We were at 10 cores on Broadwell-E so logically, the least we would have gotten from Intel this time around would have been 12 cores. Looking at the statement above though, who knows? How many cores would it take to get prosumers to "refresh faster"?

Spoiler

https://www.engadget.com/2017/08/07/intel-18-core-x-series/

 

[swilli89]

Its pretty obvious to everyone with at least couple of brain cells that TR forced Intel to bring HCC dies to HEDT platform. They have NEVER done that before. Heck sometimes they haven't even enabled the whole LCC die on HEDT platform like SB-E, which had 8 cores but even the 3970X refresh only had 6 core enabled.

And all the leaks + slides were saying up to 10 cores before TR rumours started emerging.

Right exactly. Why are people so adamantly defending the fact that Intel pulled in HCC dies to combat ThreadRipper? What's the problem with admitting AMD is giving Intel competition and thus they are reacting accordingly, just like any good business should? Posters here seem to be sticking their heads in the sand and refusing to acknowledge any of the strides AMD has made. Again.. why? For this 8th gen..maybe Intel had always plotted to bring out a 6-core mainstream SKU.. even if it has just been 6 months they launched Kaby Lake on LGA 1151. In fact I'd say they probably had planned this.. but pulling in 18C dies when they had very predictably for years increased HEDT core count linearly? Its pure mental gymnastics to try to argue that ThreadRipper didn't have some effect on this.

However.. none of that really matters as its just forum banter.. the "why" doesn't really matter to people actually forking over cash for PC parts. The "what" in this case.. Intel cutting of Z270 LGA1151.. what the hell is that? I thought I read here many times that LGA 1151 was supposed to have an upgrade path? There hasn't been any major new platform technology that comes close to requiring this, it just seems like pure artificial segmentation to promote chipset sales.

Edit: Been reading about cannonlake as I've been lost in the whirlwind of "new" Intel cpu families. Is cannonlake basically more Skylake or is it a Tock?

 

[dullard]

Are there any words on 35W Quad Core i3's incoming?

I haven't heard anything yet. With the 8100 and 8350k rumor (if true) bumping up the TDP it doesn't look like 35 W chips are happening at launch. They may happen later, depending on the 10 nm progress. There are some rumors about 15W quad core versions for mobile:
https://en.wikichip.org/wiki/intel/cores/coffee_lake_u

 

[dullard]

Right exactly. Why are people so adamantly defending the fact that Intel pulled in HCC dies to combat ThreadRipper? What's the problem with admitting AMD is giving Intel competition and thus they are reacting accordingly, just like any good business should?

The part that people fight over most is the idea that Intel suddenly rushed out new processors just in response to AMD. No, that did not happen. Yes, Intel is responding to AMD. But, just about all the details that people claim that Intel is doing are wrong. What did happen was that Intel is repackaging processors that they already made (Xeon) and putting a more consumer friendly price tag on them. In exchange for the lower cost, features are taken away from the original Xeon product (such as multiple socket compatibility).

It was not a new design, it was not rushed, it is not a panic, it is not being forced, it is not new turbos, etc. It is just taking an existing product and selling it two ways.

 

[Topweasel]

Having ring bus vs. not having ring bus completely depended on whether Intel planned on having a 6c-10c Xeon SP.

Xeon, Xeon SP, SL-X are and were always going to be the same products. The only reason LCC might have been ring bus is that at 6-10c ring bus is quicker and if they weren't going to offer and SP unit in that size then the 6-10c Ring bus Xeon would make a great SL-X.

But that doesn't really matter in regards to x299 options. I would challenge that Intel has always stayed on the smaller X299 option. But even then we only need to look at availability. If Intel was going to launch SL-X with MCC, then why didn't it launch with MCC? In fact when has Intel ever launched a product and left all the data and the release date as TBD?

 

[Glo.]

I haven't heard anything yet. With the 8100 and 8350k rumor (if true) bumping up the TDP it doesn't look like 35 W chips are happening at launch. They may happen later, depending on the 10 nm progress. There are some rumors about 15W quad core versions for mobile:
https://en.wikichip.org/wiki/intel/cores/coffee_lake_u

Well we have seen always that Intel released T versions of the CPUs, immediately, at launch.

That is what I am hoping for. I cannot buy low-power desktop CPU from AMD, so I hope I can buy low-power CPU from Intel, that has 4 cores and 8 threads.

 

[swilli89]

The part that people fight over most is the idea that Intel suddenly rushed out new processors just in response to AMD. No, that did not happen. Yes, Intel is responding to AMD. But, just about all the details that people claim that Intel is doing are wrong. What did happen was that Intel is repackaging processors that they already made (Xeon) and putting a more consumer friendly price tag on them. In exchange for the lower cost, features are taken away from the original Xeon product (such as multiple socket compatibility).

It was not a new design, it was not rushed, it is not a panic, it is not being forced, it is not new turbos, etc. It is just taking an existing product and selling it two ways.

Glad we can agree on that. My question is, what exactly is Cannonlake though? If its just more Skylake(Kaby lake).. then those TOO have existed in 6 core versions for almost two years now! Its very hard to answer if Intel was ever planning on moving mainstream to 6 if you stare at this chart.

I mean for over 9 years, Intel was content to offer increasingly smaller die chips as 4-core SKUs for mainstream. Who is to say Cannonlake isn't just a 7800X for a (randomly) new mainstream platform? If IPC and clockspeed are nearly the same, then I don't see the argument that Intel spent years developing this silicon to bring to market when its possibly the same exact story as Xeon silicon being re-purposed for HEDT. If cannonlake brings tangible IPC or clockspeed improvements over existing Skylake then I will gladly admit it must have been in development before Ryzen was released.

 

[Timmah!]

Again thats still taking the Xeons (SKL-X are already Xeons, they are all mesh, the original rumor was 10C SKL-X was ring bus LCC and 12C+ Xeons MCC or HCC, thus an emergency addition, wrong they are all server mesh cpus, they are all xeons), validating a 12 to 18C cpu in a socket designed for 10C max, moving the production schedules of SKL-X, CFL-S and Xeon to fit the SKL-X +12C, in the time-schedule that was originally intended for SKL-X in the first place. If they managed to do that is at minimum amazing.

And for what? is not needed, they had the Xeons to fight TR in high end workstations, like it was with BDW-E and before.

Just face it, this rumor died the day "September launch" was know, everything leading up to this rumor turned out to be fake. The only thing you have are the old pre-release leaked slides and they may not be accurate to start with. How many pre-release slides had errors or were compeltely fake before the launch of X product?

There was never any rumor like that. The rumors or better said actual slides like the one i linked dealt strictly with core counts. There was no mention of MESH until maybe 2 months ago:

http://wccftech.com/intel-skylake-x-skylake-sp-mesh-architecture-interconnect/

Just for the recapitulation:

Intel originally released info, which pointed toward the possibility of 10C being the top Skylake-X consumer part. Its on that previously linked slide.
Later, rumors came up, that the top HEDT chip will actually be 12 cores, not just 10C. This was reinforced by this photo:

This is the first mention of 12 core Skylake-X part, although as you can see, its properties were unknown yet, which kinda makes it look like afterthought, not surprising given the original Intel slides.
At this point of time, there was no talk about LCC and HCC and whatnot, this stuff came into light later when the whole line-up up to 18C was revealed . There was a bit of confusion, whether 12C will be still LCC, or already HCC, which was later definitely put to rest with the Mesh info.

Finally the Computex slide:

Threadripper was officially mentioned by AMD for the first time on 16th May, 2 weeks before Computex.

http://www.amd.com/en-us/press-releases/Pages/amd-unveils-expanding-2017may16.aspx


this is how it all happened to my knowledge. You are free to dismiss it as fake.

 

[dullard]

Well we have seen always that Intel released T versions of the CPUs, immediately, at launch.

That is what I am hoping for. I cannot buy low-power desktop CPU from AMD, so I hope I can buy low-power CPU from Intel, that has 4 cores and 8 threads.

See the top graph in Sweepr's post on page 559:
https://forums.anandtech.com/thread...s-out-page-554.2428363/page-559#post-39014373

Only the 95W and 65W versions are listed in 2017. 35W versions aren't listed until 2018. That jives with the rumors of a limited Coffee Lake launch soon and a second launch in early 2018.

 

[Glo.]

See the top graph in Sweepr's post on page 559:
https://forums.anandtech.com/thread...s-out-page-554.2428363/page-559#post-39014373

Only the 95W and 65W versions are listed in 2017. 35W versions aren't listed until 2018. That jives with the rumors of a limited Coffee Lake launch soon and a second launch in early 2018.

That is fine by me. All I wanted to know whether I will be able to buy 4C/8T, 35W TDP CPU that has around 150$ price tag.

 

[Bouowmx]

what exactly is Cannonlake though

Cannon Lake is Skylake architecture in 10 nm (also a new GPU architecture, Gen 10). Process shrinks of past Intel architectures after Sandy Bridge brought small increases in IPC.
10-nm Cannon Lake in any form outside of 2-core for mobile has been replaced by 14-nm++ Skylake/Kaby Lake optimized.

That is what I understand. Correct me, if wrong.

 

[swilli89]

Cannon Lake is Skylake architecture in 10 nm (also a new GPU architecture, Gen 10). Process shrinks of past Intel architectures after Sandy Bridge brought small increases in IPC.
10-nm Cannon Lake in any form outside of 2-core for mobile has been replaced by 14-nm++ Skylake/Kaby Lake optimized.

That is what I understand. Correct me, if wrong.

Ok wait so what is this 8th generation then if Cannon lake comes after Kaby Lake.. I am genuinely confused. And will these new 8th gen's have soldered IHS? Base clocks of 3.6ghz + non soldered IHS may actually bring a regression compared to 7700k.. how will they fair against 1600X?

 

[TheF34RChannel]

Cannon Lake is Skylake architecture in 10 nm (also a new GPU architecture, Gen 10). Process shrinks of past Intel architectures after Sandy Bridge brought small increases in IPC.
10-nm Cannon Lake in any form outside of 2-core for mobile has been replaced by 14-nm++ Skylake/Kaby Lake optimized.

That is what I understand. Correct me, if wrong.

I.e. Coffee Lake

 

[PeterScott]

I mean for over 9 years, Intel was content to offer increasingly smaller die chips as 4-core SKUs for mainstream. Who is to say Cannonlake isn't just a 7800X for a (randomly) new mainstream platform? If IPC and clockspeed are nearly the same, then I don't see the argument that Intel spent years developing this silicon to bring to market when its possibly the same exact story as Xeon silicon being re-purposed for HEDT. If cannonlake brings tangible IPC or clockspeed improvements over existing Skylake then I will gladly admit it must have been in development before Ryzen was released.

There is no doubt Intel was working on switching to 6 core mainstream before even the first leaks of Ryzen 8 core.

So this was coming regardless.

Also there is an easy way to tell if a modern Intel CPU die is intended for Mainstream or for Xeon/HEDT:

Integrated graphics.

Cannonlake and Coffee Lake are mainstream parts with GPUs.

 

[Bouowmx]

8th generation Core, or code name Coffee Lake, are client (mainstream) names for the 14-nm++ Skylake/Kaby Lake optimized replacement for 10-nm Cannon Lake. HEDT and server will also get their Cannon Lake replacement, code named differently.

 

[coercitiv]

It is not a new design, it is not rushed, it is not panic, it is not being forced, it is not new turbos.

Can't wait to see Intel's new Turbo policy for Coffee Lake. No rush, I'll wait. (and maybe even buy)

 

[dullard]

Glad we can agree on that. My question is, what exactly is Cannonlake though? If its just more Skylake(Kaby lake).. then those TOO have existed in 6 core versions for almost two years now! Its very hard to answer if Intel was ever planning on moving mainstream to 6 if you stare at this chart.

Are you confusing Cannon Lake with Coffee Lake? Cannon Lake has been on roadmaps since at least April 2014: https://forums.anandtech.com/threads/cannon-lake-appears-on-roadmap-due-mid-2016.2422950/

Cannon Lake is a 10 nm shrink. Processor die shrinks are normal part of electronics. Is Intel doing a die shrink because of AMD? Maybe. Or maybe they are doing it because it allows Intel to make far more chips, each far less expensive, in the same amount of time.

Coffee Lake is the one that was unexpected. But, that is because Intel wanted Cannon Lake in 2016 (see the chart above). The 1.5+ year delay in Cannon Lake gave Intel a window to tweak Skylake once again.

If this is solely due to AMD, please explain for me with your chart why Broadwell went from 8 to 10 cores in 2016 before Ryzen or TR had any real released data? Why did Haswell go from 6 cores to 8 cores in 2013? Also due to AMD's strength? Why did Westmere go from 4 to 6 cores in 2010? Was this because of Ryzen or AMD's strength in 2010 as well? Why did Penryn go from 2 cores to 4 in 2007? Was this also due to Ryzen? Were these all AMD caused? Or did Intel just add 2 cores every couple of years?

 

[Arachnotronic]

Are you confusing Cannon Lake with Coffee Lake? Cannon Lake has been on roadmaps since at least April 2014: https://forums.anandtech.com/threads/cannon-lake-appears-on-roadmap-due-mid-2016.2422950/

Cannon Lake is a 10 nm shrink. Processor die shrinks are normal part of electronics. Is Intel doing a die shrink because of AMD? Maybe. Or maybe they are doing it because it allows Intel to make far more chips, each far less expensive, in the same amount of time.

Coffee Lake is the one that was unexpected. But, that is because Intel wanted Cannon Lake in 2016 (see the chart above). The 1.5+ year delay in Cannon Lake gave Intel a window to tweak Skylake once again.

If this is solely due to AMD, please explain for me with your chart why Broadwell went from 8 to 10 cores in 2016 before Ryzen or TR had any real released data? Why did Haswell go from 6 cores to 8 cores in 2013? Also due to AMD's strength? Why did Westmere go from 4 to 6 cores in 2010? Was this because of Ryzen or AMD's strength in 2010 as well? Why did Penryn go from 2 cores to 4 in 2007? Was this also due to Ryzen? Were these all AMD caused? Or did Intel just add 2 cores every couple of years?

It's really hilarious that some people think that Intel releases new products at all because of AMD and not because, you know, that's what chip companies do.