Intel Skylake / Kaby Lake

[Yeroon]

Intel sure likes their lakes lately. It's difficult to follow what's new in terms of CPU codenames....

We had separate but memorable codenames before, then fields, bridges, wells, and now since 2015 we have a multitude of lakes well into 2020 or so.


Gotta draw that codename vs time graph...

Yup, I'm mostly confused since the *well generation.
Done right (and I'd think as per original plan) it would be node/arch dependent (ie sandy - ivy bridge)
If you do draw a graph, throw in the HEDT/xeon name-dates, for extra confusion.

 

[LTC8K6]

Are we still talking about x299 HEDT ? @ 10nm ?

No, the question was about Cascade Lake / Skylake-SP.

 

[jpiniero]

Huh, and I thought Ice Lake was supposed to be DCG b/4 CCG - meh.

I'd have to think so. Kind of makes sense... Cascade Lake (14++) 2018, Ice Lake Server (10++) 2019; Sapphire Rapids (7) 2020.

Icelake mainstream is probably 10+ while Tigerlake is 10++.

 

[WingZero30]

Has there been any further details on Skylake-W ?Afaik it should also be releasing on 2066 platform.

 

[LTC8K6]

Has there been any further details on Skylake-W ?Afaik it should also be releasing on 2066 platform.

Those:
https://linustechtips.com/main/topic/761694-intel-x299-skylake-w/

don't seem to make any sense at all. There must be something we don't know to justify a 140W tdp i5...

 

[TheF34RChannel]

 

[WingZero30]

So it appears that Skylake-W will be on a slightly different socket R LGA 2061 but using the same underlying platform Basin Falls.

 

[jpiniero]

don't seem to make any sense at all. There must be something we don't know to justify a 140W tdp i5...

The mythical 5+ Ghz chip for the NSA and HFT? Although I would think the TDP would be higher...

 

[dullard]

The mythical 5+ Ghz chip for the NSA and HFT? Although I would think the TDP would be higher...

There may be a slight clock boost. And if TheF34RChannel's post is correct there is a big boost in features (quad channel faster ECC memory, far more PCI lanes, etc).

But I think the 140 W TDP is Intel's attempt to finally allow upgrades on the same motherboard. They are unrolling what they call a "scalable family" of workstation processors. The only way to actually be scalable is if the chips all have the same TDP. That way the motherboard, power supply, and cooling system all are fully capable to run any similarly pinned processor. Then you can buy a workstation with one line of processor and later on upgrade to any of the others.

 

[LTC8K6]

The mythical 5+ Ghz chip for the NSA and HFT? Although I would think the TDP would be higher...

Maybe with 14nm++ you could get there in reasonable fashion with just 4 cores on a bigger die?
Assuming SL-W is actually 14nm++...

 

[blue11]

Those:
https://linustechtips.com/main/topic/761694-intel-x299-skylake-w/

don't seem to make any sense at all. There must be something we don't know to justify a 140W tdp i5...

There have always been quad-core server chips with ~140 W TDP, such as E5-1630v4 and E7-8893v4. The TDP can be explained in a number of ways. Usually all the cache on the die is enabled, so the power draw from the L3 is the same as the full chip. Additionally, due to binning, these quad-core chips are probably those with very high defect levels, so they may be particularly inefficient. Lastly, the TDP is officially only for sizing heatsinks, so not all chips may actually draw that much power.

 

[LTC8K6]

There may be a slight clock boost. And if TheF34RChannel's post is correct there is a big boost in features (quad channel faster ECC memory, far more PCI lanes, etc).

But I think the 140 W TDP is Intel's attempt to finally allow upgrades on the same motherboard. They are unrolling what they call a "scalable family" of workstation processors. The only way to actually be scalable is if the chips all have the same TDP. That way the motherboard, power supply, and cooling system all are fully capable to run any similarly pinned processor. Then you can buy a workstation with one line of processor and later on upgrade to any of the others.

I thought scalable platform was Skylake-SP? The Bronze/Silver/Gold/Platinum chips.
Sklyake W is Workstation.
Or are they the same now?

https://en.wikichip.org/wiki/intel/microarchitectures/skylake#Codenames

 

[LTC8K6]

There have always been quad-core server chips with ~140 W TDP, such as E5-1630v4 and E7-8893v4. The TDP can be explained in a number of ways. Usually all the cache on the die is enabled, so the power draw from the L3 is the same as the full chip. Additionally, due to binning, these quad-core chips are probably those with very high defect levels, so they may be particularly inefficient. Lastly, the TDP is officially only for sizing heatsinks, so not all chips may actually draw that much power.

Yeah but the W 4C chip doesn't even have HT...
I think they just label all of the chips with the same TDP, regardless.
Probably for customer ease of mind that the chips are all compatible.

 

[blue11]

I thought scalable platform was Skylake-SP? The Bronze/Silver/Gold/Platinum chips.
Sklyake W is Workstation.
Or are they the same now?

https://en.wikichip.org/wiki/intel/microarchitectures/skylake#Codenames

The "SP" in Skylake-SP stands for "scalable performance." The scalable platform is Purley (LGA-3647). It refers to how the same family of chips scales from 2S to 8S (not exactly, but it's less segmented than E5/E7).

 

[LTC8K6]

The "SP" in Skylake-SP stands for "scalable performance." The scalable platform is Purley (LGA-3647). It refers to how the same family of chips scales from 2S to 8S (not exactly, but it's less segmented than E5/E7).

It doesn't appear to stand for either actually, but we are talking about the same thing overall.
The W chips are not the "scalable" chips...

 

[Ajay]

There may be a slight clock boost. And if TheF34RChannel's post is correct there is a big boost in features (quad channel faster ECC memory, far more PCI lanes, etc).

Those extra features (PCIe lane, DRAM channels, etc.) only apply to Skylake-X, not to Kaby Lake or Cannon Lake (the internal architecture of the client based CPUs just don't support these features).

 

[Edrick]

So skylake-w is just the Xeon version of skylake-x.

Looks like the real platform to get will be the skylake-sp

 

[Ajay]

Skylake-W is just the platform name, IIRC.

 

[Sweepr]

Kaby Lake-R (4C/8T 15W) @ SiSoftware

http://ranker.sisoftware.net/show_r...d5e3daead2e2dafc8eb383a5c0a598a88efdc0f8&l=en

 

[dooon]

SuperMicro Skylake-W mobo
https://www.supermicro.com.tw/ResourceApps/MB_matrix.aspx

 

[jpiniero]

Meh, they also have a Quad Xeon Gold 6152 score now too, which gets 8041. Kind of interesting that Xeon Gold 6xxx will allow 4P (?)

 

[Sweepr]

https://ecauk.com/products/asmb-825/

 

[dooon]

Platinum 8180

http://tieba.baidu.com/p/5127168939

 

[Sweepr]

Skylake-SP is already dominating SiSoftware's Multi-Media ranks.

Rumor says Xeon Platinum 8180 scores 4200+ in CB R15, ahead of AMD's Epyc 32C/64T (4000+). This is really impressive considering how well Zen performs in CB R15, Ryzen 7 1800X actually beats stock Core i7-6900X here, even if it's slower overall.

 

[Sweepr]

DigiTimes reaffirms Coffee Lake-S in late August, this is their third article mentioning the earlier than expect launch:

Because of weak demand in the PC DIY market, Intel is considering bundling its Optane memory with motherboards using 200 series chipsets such as Z270, in the PC DIY channel to stimulate demand for the Kaby Lake platform. However, some retailers are pessimistic about the idea since the next-generation Coffee Lake is set to become available in late August and consumers are waiting for the new products.

...The retailers are now placing their hope in the third quarter and believe the new platform will stimulate demand.

http://www.digitimes.com/news/a20170518PD204.html