News Intel Rocket Lake CPUs Might Max Out With Eight Cores at 125W

bit_user

Polypheme
Ambassador
It wouldn't be a horrible idea to move the new uArch to the process that allows for higher clock speeds and more cores right now.
Do you realize they're in opposition? Higher clock speeds is why Intel is keeping the high-end at 14 nm, while the low/mid-range is moving ahead to 10 nm (i.e. the process where it's most economical to add more cores).
 
Do you realize they're in opposition? Higher clock speeds is why Intel is keeping the high-end at 14 nm, while the low/mid-range is moving ahead to 10 nm (i.e. the process where it's most economical to add more cores).

Low power is pushing to 10nm currently yes. That does not mean the uArch, Sunny Cove, is tied only to 10nm. In fact Intel expressly said they felt a point of failure was tying uArchs and process tech together.

There is nothing stopping Intel from moving Sunny Cove to 14nm as a stop gap. The uArch itself does show some signs of improvement and even at lower clock and boost speeds tends to outperform Whisky Lake parts. Especially since 14nm has vastly better yields and probably margins than 10nm does at current.

I would also imagine more cores is easier since Intel has yet to push past 4 cores on 10nm yet meaning that either yields are not up to snuff or there is an issue elsewhere.

But as I said, rumors are rumors. They mean absolutely nothing until it is officially announced. Otherwise we would all have 16 core Ryzen chips clocked at 5GHz on simple air cooling.
 

joeblowsmynose

Distinguished
Something doesn't look right here in this chart ...

"S" series CPUs with 125wTDP? That makes little sense unless we should expect their "K" series to have 145~165w TDPs - which is probably too high for mainstream desktop.

EDIT: I wouldn't put too much faith in this rumour (or at least the details) If 10nm is really as broken as it appears (I think maybe it just can't do high clocks at all, or not without sucking too much power), I agree with Jimmy, not the worst idea to backport to the 14nm node in that case. If true though, that seems like an admission that 10nm has even bigger issues than ... already big issues, or, that 10nm will never work for high performance parts at least. Losing a lot of density though going from planned 10nm to 14nm.
 
Last edited:

joeblowsmynose

Distinguished
The "S" lineup is every desktop CPU for their mainstream socket, including "K" CPUs.

Of course it would be confusing as hell, this is Intel naming conventions we're talking about here ... I guess I was going by old info, where "S" series was a separate line from k/x altogether.

 
Last edited:
  • Like
Reactions: TJ Hooker

TJ Hooker

Titan
Ambassador
Of course it wouldn't be confusing as hell, this is Intel naming conventions we're talking about here ... I guess I was going by old info, where "S" series was a separate line from k/x altogether.

Yeah, it looks like it used to be that way but now they refer to their whole mainstream socketed lineup as -S.
 
Last edited:
  • Like
Reactions: joeblowsmynose
Yeah, it looks like it used to be that way but now they refer to their whole mainstream socketed lineup as -S.

It still is. Rocket Lake-S is the same as Sky Lake-S. It covers from the lowest end desktop part to the highest end K series. Has not changed for quite a while and -S is still desktops.

Of course it would be confusing as hell, this is Intel naming conventions we're talking about here ... I guess I was going by old info, where "S" series was a separate line from k/x altogether.


You are mixing up models and uArchs. -S is typically after the uArch and has for quite a while, at least since Sky Lake, meant the desktop line of chips from entry level to top end K series. S, notice no dash, after the model number is a lower power version.

Nothing has changed.
 

bit_user

Polypheme
Ambassador
I would also imagine more cores is easier since Intel has yet to push past 4 cores on 10nm yet meaning that either yields are not up to snuff or there is an issue elsewhere.
Okay, I see your point. You think that because Intel hasn't made a > 5 core (let's not forget about that weird heterogeneous Lakefield CPU, I think it's called) @ 10 nm, it's because they can't.

The way I see it, it's just market segmentation. Right now, 10 nm seems good only for mid-range laptop CPUs, and they didn't want to make a 6-core because that would compete too much with their high-end Comet Lake CPUs. I mean, if 10 nm were so area-limited, why the heck does Ice Lake have AVX-512 and (more to the point) up to 64 EU iGPUs? Such a large iGPU would make no sense, if their yields were too poor to manage even a 6-core part.

Anyway, their roadmap shows 10 nm, 26-core Ice Lake server CPUs in Q1 2020. So, we'll know soon enough if their current 10 nm (which I guess is technically 10 nm+) can deliver the goods.
 

bit_user

Polypheme
Ambassador
If 10nm is really as broken as it appears (I think maybe it just can't do high clocks at all, or not without sucking too much power), I agree with Jimmy, not the worst idea to backport to the 14nm node in that case. If true though, that seems like an admission that 10nm has even bigger issues than ... already big issues, or, that 10nm will never work for high performance parts at least.
This is not news. Leaked roadmaps (from 7 months ago) have shown Intel will skip 10 nm for the mainstream desktop segment. It'll stay on 14 nm through the end of 2021.

FmUw4g59WEwuksLMFntYHZ.png


Source: https://www.tomshardware.com/news/intel-roadmap-comet-lake-client,39185.html

Actually, that's the client commercial roadmap, but I recall it's the same story for the client consumer roadmap - just maybe not Q4 2021, since new products tend to launch first in the consumer channel.
 
Okay, I see your point. You think that because Intel hasn't made a > 5 core (let's not forget about that weird heterogeneous Lakefield CPU, I think it's called) @ 10 nm, it's because they can't.

The way I see it, it's just market segmentation. Right now, 10 nm seems good only for mid-range laptop CPUs, and they didn't want to make a 6-core because that would compete too much with their high-end Comet Lake CPUs. I mean, if 10 nm were so area-limited, why the heck does Ice Lake have AVX-512 and (more to the point) up to 64 EU iGPUs? Such a large iGPU would make no sense, if their yields were too poor to manage even a 6-core part.

Anyway, their roadmap shows 10 nm, 26-core Ice Lake server CPUs in Q1 2020. So, we'll know soon enough if their current 10 nm (which I guess is technically 10 nm+) can deliver the goods.

I would say yields are not good enough personally to focus on anything but higher margin server and HPC parts. While mobile is a higher margin part than desktop, HPC and server is much higher.

I could be wrong but I still doubt 10nms yields are high enough to take over mobile, desktop and HPC markets. Plus I would imagine Intel is working on a way to compete on a core to core level as with AMD shoving 16 cores into mainstream they will have to match them. I also imagine they don't want to lose their clock advantage which 10nm also doesn't show as good of clocks as 14NM has, makes sense as 14nm has been matured well beyond most process techs.

It may all change and Intel could turn around tomorrow and release an entire 10nm desktop lineup that matches AMD on cores and price per core. But nothing we have seen in roadmaps, leaked or released, or even rumors point to that and while Intel has stated they plan 10nm desktop parts it still feels like they might skip over it for 7nm.
 

bit_user

Polypheme
Ambassador
I would say yields are not good enough personally to focus on anything but higher margin server and HPC parts.
I still don't see how that's supposed to work. So, they can make 26-core server chips but not 6-core mobile chips? If the yield on 6-core mobile chips isn't viable, the yield on 26-core chips must be atrocious - even if you have some spare cores on there.

I could be wrong but I still doubt 10nms yields are high enough to take over mobile, desktop
No, the reason they can't take (high-end) mobile + desktop with 10 nm is down to clock speeds. Server chips and mid-range mobile already run at lower clocks, so they're ripe for 10 nm. It's really not that hard to grasp.
 
Last edited:
I still don't see how that's supposed to work. So, they can make 26-core server chips but not 6-core mobile chips? If the yield on 6-core mobile chips isn't viable, the yield on 26-core chips must be atrocious - even if you have some spare cores on there.


No, the reason they can't take (high-end) mobile + desktop with 10 nm is down to clock speeds. Server chips and mid-range mobile already run at lower clocks, so they're ripe for 10 nm. It's really not that hard to grasp.

As I said I could be wrong but there is a reason why Intel is not pushing 10nm faster and harder. Even with a clock deficit the 10th gen 10nm parts seem to perform well against Whisky Lake parts that have higher clock speeds. So while, and I did say this, losing a clock speed advantage wouldn't be the best thing it wouldn't be the end of the world.
 

joeblowsmynose

Distinguished
You are mixing up models and uArchs. -S is typically after the uArch and has for quite a while, at least since Sky Lake, meant the desktop line of chips from entry level to top end K series. S, notice no dash, after the model number is a lower power version.

Nothing has changed.

Not confusing at all ... "-s" series and "s" series -- entirely different! (and there's no dash in the chart on the article - it just says "s series" which is the same designation for the models)

In all fairness I'm starting to get used to saying "eye nine tenty nine eighty ex ee" :)
 
Not confusing at all ... "-s" series and "s" series -- entirely different! (and there's no dash in the chart on the article - it just says "s series" which is the same designation for the models)

In all fairness I'm starting to get used to saying "eye nine tenty nine eighty ex ee" :)

But the -S is not a series. Its a uArch market designation. Typically they have three, mobile, desktop (mainstream) and HEDT.

The S/K/Whatever they use is a model designation with S being low power, no letter being the normal model and K being the unlocked variant. Of course the 9900K has three of its own, K, KF and KS.
 

joeblowsmynose

Distinguished
But the -S is not a series. Its a uArch market designation. Typically they have three, mobile, desktop (mainstream) and HEDT.

The S/K/Whatever they use is a model designation with S being low power, no letter being the normal model and K being the unlocked variant. Of course the 9900K has three of its own, K, KF and KS.

Fine, but ehe article chart literally named it "S series" -- see, everyone is confused - its not just me. ;)
Don't forget the 9900T and whatever happened to the 9900 "KFC" edition?
 
  • Like
Reactions: TJ Hooker
Fine, but ehe article chart literally named it "S series" -- see, everyone is confused - its not just me. ;)
Don't forget the 9900T and whatever happened to the 9900 "KFC" edition?

I see where they did that. It is a TH mistake though.

And yea the KFC edition? Not sure what that really was. It might have been an internal only model for testing, they have tons of those, that never get released.
 

JayNor

Reputable
May 31, 2019
426
85
4,760
the 10nm laptop chips are still using a ring bus, with the gpu also on the ring. Perhaps they have a reason to stick with that ring bus for the laptop chips, even though going to 14nm. Maybe they just don't have an efficient design for GPU interconnect to the mesh yet.

So, I guess the real question is, how many willow cove cores can they hang on that ring bus before it gets noticeably slow.
 
  • Like
Reactions: bit_user