Coffee Lake quads - HT?

[Yuriman]

I haven't kept up with releases as well in the last few months as I have historically, and I'm trying to catch up. Here's where I'm at:

After poking around for bit on google and looking through threads on here, it looks like CFL is desktop only (for now?), and that Intel is still calling 8th gen mobile "Kaby Lake", though Intel is moving mobile i5's and i7's from 2+HT to 4+HT. From this I suppose that the 8th gen mobile i5's and i7's, are still using 14nm+ rather than CFL's ++, otherwise they'd be CFL.

Right now Coffee Lake for desktop exists as 6+HT, 6, and 4 core variants... which leads me to wonder, do the CFL quads even have the physical circuitry to allow HT? Is there any indication Intel has a 4+HT CFL CPU in the works? It seems odd to me that a feature might be included in a die, which is seen in literally no production CPUs using that die.

Additionally, I gather that Cannon Lake, Intel's first 10nm CPU, will be mobile only. What comes next? I haven't yet found any slides which show anything explicitly.

 

[LTC8K6]

I would guess that there would be little performance difference between a 4C/8T CL chip and the 6C/6T CL chip, so it's redundant.

 

[jpiniero]

It's possible the desktop T models will be 4C8T but we will have to see. Possibly some mobile models too.

Next for desktop looks like more Coffee Lake; with higher core counts and that's it. The i7 would be 8C16T.

 

[dullard]

I would guess that there would be little performance difference between a 4C/8T CL chip and the 6C/6T CL chip, so it's redundant.

Exactly. 4C/8T vs 6C/6T is a difficult marketing and consumer education difference. 4C/8T will usually perform about the same as a 5C/5T chip. But not always. Sometimes, it will be far better than that. So, which chip do you charge more for? Which chip do you suggest goes into each segment (budget, middle, premium, etc).

Having features for the sake of having features is not always good for the company, especially when there is a nearly equivalent part already for sale.

 

[Yuriman]

I understand a 4c8t and 6c/6t chip would step on each others' toes, but since it doesn't appear Intel is or will be selling any 4c8t chips based on that die, is the actual silicon there for hyperthreading? I mean, the quads are not harvested 6c dies, are they? Or are they?

 

[VirtualLarry]

I'm personally convinced that the i3-8100 and i3-8350K 4C/4T are simply Kaby Lake 4C dies, and do contain HyperThreading, even if it does seem like a waste that it's not enabled on any of the 4C CFL SKUs.

 

[Yuriman]

I'm personally convinced that the i3-8100 and i3-8350K 4C/4T are simply Kaby Lake 4C dies, and do contain HyperThreading, even if it does seem like a waste that it's not enabled on any of the 4C CFL SKUs.

While this would make sense from a cost perspective, this would mean they're not built on the same process as the 6c variants? And, the changes to the iGPU are strictly software/firmware? And the changes to pinout/socket can be done entirely on the substrate?

 

[LTC8K6]

While this would make sense from a cost perspective, this would mean they're not built on the same process as the 6c variants? And, the changes to the iGPU are strictly software/firmware? And the changes to pinout/socket can be done entirely on the substrate?

The IGP is identical to KL as far as I know. They just changed the name.

Keep in mind the 4C KL-X chips as well, as far as using a different socket with the same core.

 

[firewolfsm]

My impression was that the i3-8350k was overclocking better than kabylake quads. And that the die is the same size as the 6 cores.

It also makes sense to have a lower priced 8th gen processor for dies with defective cores.

 

[VirtualLarry]

My impression was that the i3-8350k was overclocking better than kabylake quads. And that the die is the same size as the 6 cores.

It also makes sense to have a lower priced 8th gen processor for dies with defective cores.

Maybe. But if there is only one CFL-specific die, the 6C w/HT, then it also makes sense that they would re-badge 4C w/HT Kaby Lake dies.

And don't believe the hype about 14nm, +, ++.... it's all one continuous process improvement. Benchmark recently mfg KBL quads, and I think that you'll find that's true.

There is a material difference with the 6C CFL die, they did do stuff to the fin height, etc., with that design. But not the quads, I don't think. Unless all i3-8100 CPUs are really 6C dies, which delidding would probably identify if true.

 

[VirtualLarry]

And the changes to pinout/socket can be done entirely on the substrate?

Easily. They've already adapted KBL quads to the X299 platform / socket. And AMD did the same thing with Carizzo, putting it on AM4 and calling it Bristol Ridge.

 

[PeterScott]

, do the CFL quads even have the physical circuitry to allow HT? Is there any indication Intel has a 4+HT CFL CPU in the works? It seems odd to me that a feature might be included in a die, which is seen in literally no production CPUs using that die.

CFL Quads are just Kaby Lake Quads with new pin layout, to work in CFL motherboards.

They just have HT disabled, just like all of Intel Core CPUs without HT.

 

[Jan Olšan]

My impression was that the i3-8350k was overclocking better than kabylake quads. And that the die is the same size as the 6 cores.

It also makes sense to have a lower priced 8th gen processor for dies with defective cores.

Nope, the quadcores show stepping B0 which is the same as Kaby Lake. Hexacores report stepping U0. There is also a physical similarity in the packaging underneath (i3-8350K has the same components under it as i7-7700k, while i7-8700K is different). It's almost certain that 8100/8350K are just Kaby Lake rebadges. There is also the hint that they are more readily available compared to the hexacores.

7700K
http://hw-museum.cz/data/cpu/pic/Intel_Core_i7-7700K_B.jpg

8350K
https://www.techpowerup.com/reviews/Intel/Core_i3_8350K/images/cpu2.jpg

8700K
https://www.techpowerup.com/reviews/Intel/Core_i7_8700K/images/cpu2.jpg

(from http://pctforum.tyden.cz/viewtopic.php?f=2&t=210041&p=9343871 )

Also CPU-Z shows 8350K having stepping B0, same as reported for 7700K... 8700K has stepping U0. You can say with 99% certainty that the new i3s are just Kaby Lake slightly modified to work on Z370 platform.

 

[Yuriman]

Maybe. But if there is only one CFL-specific die, the 6C w/HT, then it also makes sense that they would re-badge 4C w/HT Kaby Lake dies.

And don't believe the hype about 14nm, +, ++.... it's all one continuous process improvement. Benchmark recently mfg KBL quads, and I think that you'll find that's true.

There is a material difference with the 6C CFL die, they did do stuff to the fin height, etc., with that design. But not the quads, I don't think. Unless all i3-8100 CPUs are really 6C dies, which delidding would probably identify if true.

The major differences highlighted in the Anandtech review are HDCP 2.2 support, and that the gate pitch was increased when compared with Kaby Lake.

Nope, the quadcores show stepping B0 which is the same as Kaby Lake. Hexacores report stepping U0. There is also a physical similarity in the packaging underneath (i3-8350K has the same components under it as i7-7700k, while i7-8700K is different). It's almost certain that 8100/8350K are just Kaby Lake rebadges. There is also the hint that they are more readily available compared to the hexacores.

(from http://pctforum.tyden.cz/viewtopic.php?f=2&t=210041&p=9343871 )

This makes a lot of sense.