Intel Skylake / Kaby Lake

[Maxima1]

Looks like Coffee Lake-S is coming sooner than an user here expected (Q3-Q4/2018). One year after Kaby Lake-S looks like a real possibility.

LOL That came out of your ***. It was already suggested many times it would be Q1/Q2.

 

[Sweepr]

It was already suggested many times it would be Q1/Q2.

Watch your language. Someone insisted it would be Q3/Q4. But that was based on an old BenchLife slide, their latest info is positive news and confirms what DigiTimes has been saying for months now.

 

[rchunter]

I'm sure there will be Skylake-X Xeons if that is what you are asking. Skylake-EP will likely be 2S and above though.

Yes thank you. I'm curious to see what the Skylake-X server boards will be like (Intel onboard 10GbE?....) I guess skylake-x only goes up to 10 core.

 

[Arachnotronic]

LOL That came out of your ***. It was already suggested many times it would be Q1/Q2.

There is a poster on these forums who was ADAMANT that CFL was Q3/Q4.

 

[Dave2150]

In that case, Coffee is the one I'm most interested in. But I have the sinking feeling that it won't be able to match (and at least not exceed) single thread performance of Kaby. Still on 14nm, and with two additional cores, it's going to be very hard to outdo Kaby at 4.5GHz... it might even be behind like HEDT is now..

Probably incorrect - Coffeelake might be soldered (larger die after all), which will allow it to run 15-30C cooler than Kabylake.

 

[Arachnotronic]

Probably incorrect - Coffeelake might be soldered (larger die after all), which will allow it to run 15-30C cooler than Kabylake.

Unlikely, IMO. CFL-X on the other hand, yes, soldered for sure.

 

[Dave2150]

Unlikely, IMO. CFL-X on the other hand, yes, soldered for sure.

We can hope

I wonder how much larger the 6C CFL die will be, compared to the 4C SKL/KBL die? Maybe only a few more mm, so you're most likely right.

Perhaps if Zen's 4C/6C products are soldered, it might encourage Intel to solder future mainstream parts also?

 

[CakeMonster]

I don't know what they're planning but, I would assume that they would try to match the single thread performance of Kaby with Coffee. Because a performance regression for most users would be bad, unless there is a magic multi threading revolution in the mean time. 14nm has been polished for quite a while, but will they be able to do that with 6 cores? I certainly hope so, and if it takes soldering or whatever, I hope they do whatever it takes...

 

[Arachnotronic]

We can hope

I wonder how much larger the 6C CFL die will be, compared to the 4C SKL/KBL die? Maybe only a few more mm, so you're most likely right.

Perhaps if Zen's 4C/6C products are soldered, it might encourage Intel to solder future mainstream parts also?

6C CFL die is 149mm^2,

As far as Zen goes, these are derived from server/workstation parts to begin with so no doubt they will be soldered. I also expect the mainstream Intel dies when used in the HEDT platform (KBL-X/CFL-X) to be soldered as well, since they will most likely be more expensive and so Intel won't mind ponying up the extra for the solder in that case (and it could make it a selling point too..."if you buy 4C KBL-X on the X299 platform, you get better overclocking" kind of deal).

 

[Arachnotronic]

I don't know what they're planning but, I would assume that they would try to match the single thread performance of Kaby with Coffee. Because a performance regression for most users would be bad, unless there is a magic multi threading revolution in the mean time. 14nm has been polished for quite a while, but will they be able to do that with 6 cores? I certainly hope so, and if it takes soldering or whatever, I hope they do whatever it takes...

They would be silly to regress on per-core performance in 1-4 core workloads. There is little value in a "new" CPU that performs worse than an "old" CPU in most workloads, and Intel knows that.

 

[Arachnotronic]

Yes thank you. I'm curious to see what the Skylake-X server boards will be like (Intel onboard 10GbE?....) I guess skylake-x only goes up to 10 core.

Skylake-W for workstations will have more cores than 10, leaks show that this platform will support chips using both the LCC and HCC (do they mean MCC?) dies, so expect >=15 cores.

 

[DrMrLordX]

Hmm. Nice of Intel to be so forward with their launch plans. Sincei it's all 14nm of some kind through Coffeelake we can safely assume that process delays won't alter their release schedule.

Now the question is whether or not the news of 6c Coffeelake in Q1 2018 will tamp down sales of LGA2066 6c chips. I seriously can't see them selling many 4c Kabylake-X chips.

 

[jpiniero]

Now the question is whether or not the news of 6c Coffeelake in Q1 2018 will tamp down sales of LGA2066 6c chips.

It shouldn't since 6C Skylake-X should be faster than Coffee Lake, unless Coffee Lake can clock much higher (i'd say unlikely but possible).

 

[Arachnotronic]

It shouldn't since 6C Skylake-X should be faster than Coffee Lake, unless Coffee Lake can clock much higher (i'd say unlikely but possible).

I think Coffee Lake will be faster because it will clock better out of the box. OC to OC, who knows, but I think CFL will be built on a 14nm++, so it could have superior frequency potential.

 

[Sweepr]

Some Skylake-X, Kaby-Lake X and Coffee Lake-S information, thanks to BenchLife!

With the X299 chipset motherboard, Intel plans to release Skylake-X and Kaby Lake-X in August

- Intel will not announce new chips at Computex 2016
- Motherboard manufacturers will showcase some of the new X299 motherboards (Computex 2016)
- The Basin Falls platform, including LGA 2066 and the new HEDT CPUs Skylake-X & Kaby Lake-X will be available in August - official launch at Gamescom between August 22-26
- Four models initially: 140W TDP 6C, 8C and 10C (Skylake-X) + 112W TDP 4C (Kaby Lake-X)
- Skylake-X and Kaby Lake-X cache structure has been adjusted to make it more balanced and improve performance
- These are K processors, not X - does this mean there's a faster model coming later?
- Memory support: Up to 4-channel DDR4-2667, Intel Optane Technology support
- Skylake-X and Kaby Lake-X will be part of the Core i7-7000 family
- 6C Coffee Lake-S will be announced Q1-2018
- Coffee Lake-S will have 35W, 65 and 95W products

https://benchlife.info/intel-will-not-announce-skylake-x-kaby-lake-x-with-x299-in-computex-01222017

Looks like Coffee Lake-S is coming sooner than an user here expected (Q3-Q4/2018). One year after Kaby Lake-S looks like a real possibility.

Here's a thought. If these are K processors I wonder if they will be cheaper than their Broadwell-E counterparts. It's hard to imagine a non-X 10C/20T CPU costing >$1700. Would be great to have $600 8C/16T and $1000 10C/20T SKUs, with a new 10C/20T or 12C/24T model replacing Core i7-6950X.

 

[Arachnotronic]

Here's a thought. If these are K processors I wonder if they will be cheaper than their Broadwell-E counterparts. It's hard to imagine a non-X 10C/20T CPU costing >$1700. Would be great to have $600 8C/16T and $1000 10C/20T SKUs, with a new 10C/20T or 12C/24T model replacing Core i7-6950X.

That'd be nice.

I'd like to see:

6 core/28 lanes @ $379
6 core/44 lanes @ $499
8 core/44 lanes @ $799
10 core/44 lanes @ $1199

Then Intel could to a true "extreme edition" which would be a 12 core with all 48 lanes enabled @ $1649.

 

[krumme]

Sound like a viable approach. If you cant compete on performance, efficiency or cost market a new metric as important though it matters for like 0.01% of the consumers. Classic strategy.

 

[Arachnotronic]

Sound like a viable approach. If you cant compete on performance, efficiency or cost market a new metric as important though it matters for like 0.01% of the consumers. Classic strategy.

Lol what? Do you really think Skylake-X won't be competitive in terms of performance, power, or cost?

 

[krumme]

Lol what? Do you really think Skylake-X won't be competitive in terms of performance, power, or cost?

We dont know as it depends on pricing vs perf/eff.
But we know for sure next to nobody needs all those lanes. So its just sidetracking what matters and its grasping for straws. Perf and efficiency for total cost is whats matter. Its years since # lanes, sata and usb was an issue for the midrange. Its similar nonsense argument as the marketing of phenom beeing true quad core. It didnt matter. Marketing fluff.

 

[Arachnotronic]

We dont know as it depends on pricing vs perf/eff.
But we know for sure next to nobody needs all those lanes. So its just sidetracking what matters and its grasping for straws. Perf and efficiency for total cost is whats matter. Its years since # lanes, sata and usb was an issue for the midrange. Its similar nonsense argument as the marketing of phenom beeing true quad core. It didnt matter. Marketing fluff.

Dude, the chips are basically server dies being offered to enthusiasts who want more cores. That's why there are a bunch of lanes.

 

[dahorns]

We dont know as it depends on pricing vs perf/eff.
But we know for sure next to nobody needs all those lanes. So its just sidetracking what matters and its grasping for straws. Perf and efficiency for total cost is whats matter. Its years since # lanes, sata and usb was an issue for the midrange. Its similar nonsense argument as the marketing of phenom beeing true quad core. It didnt matter. Marketing fluff.

Midrange? These aren't i3s we are talking about here. And with the growth of NVMe SSDs I'm not sure how you can claim that nobody needs the extra lanes.

 

[krumme]

Dude, the chips are basically server dies being offered to enthusiasts who want more cores. That's why there are a bunch of lanes.

Yep its a way to segment the market using existing products. You have to do something when you have shrunk a sandy bridge from 32nm to now 14nm skl and your exsisting i3 to i7 line is using the same number of core. Then you grab the serverline and use it to create new pricing segments. Thats how it is to get most profit.

Still its a functionality that benefits next to no one and that includes enthusiast. I would have preferred a 8c i7 at 350 usd to a handfull of lanes i can not use. I dont take it as a generous "offer" just segmentation possible because there is no compettition.

So why the distinction between the two 6 cores in your wishlist?
And why mention the lanes at all?

"
That'd be nice.

I'd like to see:

6 core/28 lanes @ $379
6 core/44 lanes @ $499
8 core/44 lanes @ $799
10 core/44 lanes @ $1199

Then Intel could to a true "extreme edition" which would be a 12 core with all 48 lanes enabled @ $1649.
"

 

[krumme]

Midrange? These aren't i3s we are talking about here. And with the growth of NVMe SSDs I'm not sure how you can claim that nobody needs the extra lanes.

Next to nobody needs it in consumer market.
Show me the consumer segment that uses 44 lanes?
How do they use the 44 lanes and point me to the actual persons doing so?

 

[Phynaz]

Next to nobody needs it in consumer market.

640k....

 

[dahorns]

Next to nobody needs it in consumer market.
Show me the consumer segment that uses 44 lanes?
How do they use the 44 lanes and point me to the actual persons doing so?

Next to nobody in the consumer market today needs 8 high performance cores. But the argument for 8 cores is seldom about today, but the future, and I don't see why the same shouldn't be true for PCIe lanes.

And there are certainly people today running 1080s in sli at 16x. Throw in a PCIE ssd and/or even a third 1080 for physx and you'll be at 40-44. Is the net benefit huge? Probably not. But we are talking about a segment of the market that gets giddy pushing every bit of performance out of a CPU.