Intel Comet Lake Thread

Page 9 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

EXCellR8

Diamond Member
Sep 1, 2010
3,982
839
136
...the 3950X doesn't even touch half that wattage at full tilt... and that's 6 more cores. (and not really a gaming CPU)

I wonder if Intel doesn't even care at this point, so long as the performance numbers are there. It's like AMD when showing off Vega lol
 

DrMrLordX

Lifer
Apr 27, 2000
21,629
10,841
136
...the 3950X doesn't even touch half that wattage at full tilt... and that's 6 more cores. (and not really a gaming CPU)

Well let's be fair: with enough capacity, big water can cool loads up to 400-500W on 14nm CPUs. It requires ridiculous cooling capacity, but it can be done. On 7nm AMD CPUs, ambient cooling reaches its limits somewhere in the 180W range, maybe less. At least that's been my experience. You get diminishing returns more quickly out of water or air cooling on those CPUs. That being said, who really wants to spec 300W cooling solutions for 14nm CPUs? Much less the board VRMs necessary to deliver that kind of power reliably?

I wonder if Intel doesn't even care at this point

They don't really have a choice.
 

NTMBK

Lifer
Nov 14, 2011
10,236
5,018
136
To that end, Intel confirmed that the reason no LPDDR4-based Comet Lake devices have been launched is that there will be a new stepping of CPUs, labeled K1 (succeeding the initial B0 stepping), that will enable support for LPDDR4. The new stepping should be available to OEMs soon if not already, for devices to come in early 2020.


So LPDDR4 was broken on the initial stepping of Comet Lake.
 

mikk

Diamond Member
May 15, 2012
4,140
2,154
136

So LPDDR4 was broken on the initial stepping of Comet Lake.

It was known for a very long time that the initial stepping of Comet lake won't support LPDDR4, there is nothing new in there. The initial CML-U is just a rebrand of WHL-U stepping 12 which was a security fix stepping over stepping 11. In this article there is nothing about "LPDDR4 was broken"
 

jpiniero

Lifer
Oct 1, 2010
14,591
5,214
136
It was known for a very long time that the initial stepping of Comet lake won't support LPDDR4, there is nothing new in there. The initial CML-U is just a rebrand of WHL-U stepping 12 which was a security fix stepping over stepping 11. In this article there is nothing about "LPDDR4 was broken"

AT's article does give that impression. They might have known about LPDDR4 being broken for some time and just said screw it, lets just release it anyway and we'll fix it later.
 
  • Like
Reactions: spursindonesia

mikk

Diamond Member
May 15, 2012
4,140
2,154
136
AT's article does give that impression. They might have known about LPDDR4 being broken for some time and just said screw it, lets just release it anyway and we'll fix it later.

Anandtech is confused about the missing LPDDR4 at launch, he is not aware of the steppings.
 

uzzi38

Platinum Member
Oct 16, 2019
2,627
5,929
146
Anandtech is confused about the missing LPDDR4 at launch, he is not aware of the steppings.
It's more likely he's reporting on what exactly Intel told him.

Intel can't go out and say "Oh yeah, out initial CML-U silicon stepping couldn't support LPDDR4 at all", can they? It'd be a pretty shoddy look given they touted LPDDR4 on all of the CML-U slides at launch.
 
  • Like
Reactions: lobz and NTMBK

Gideon

Golden Member
Nov 27, 2007
1,628
3,658
136
10-core Comet lake ES Geekbench:


Obvioulsy a WIP. Frequencies are pretty around 4400 MHz on average (add .gb5 to the end of the url to see them in the JSON output) and scores way worse than a 9900K.

This only scores about the same MT score as my stock 3700X on a B350 board (not even custom timings and never exceeding 88W even during turbo) and ~50 points worse ST score
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Apparently the UHD 730 graphics have some minor media related updates. The mobile version of Cometlake uses UHD 620/630 which doesn't change anything.
 

RetroZombie

Senior member
Nov 5, 2019
464
386
96
Comet Lake is obviously not a rebrand since it includes the additional two cores.
First let me say that you are correct.

But what's the correct name for such product. Refresh? Remake?
Since it's a new product made of an old product, hum....
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Comet Lake is obviously not a rebrand since it includes the additional two cores.

Wider back-end, 30%+ improvement on the same process - Goldmont Plus
Minor process changes, zero uarch changes - Kabylake/Coffeelake/Cometlake/Whiskeylake

It should have been Kabylake-R, Kabylake-6, Kabylake-8, and Kabylake-10. At least Kabylake gets media engine changes.

Code name changes are marketing for AT forum goers.
 
  • Like
Reactions: spursindonesia

moinmoin

Diamond Member
Jun 1, 2017
4,949
7,658
136
Wider back-end, 30%+ improvement on the same process - Goldmont Plus
Minor process changes, zero uarch changes - Kabylake/Coffeelake/Cometlake/Whiskeylake
Any idea why Intel's execution with the Atom family is so much better than with the Core family? Sure, the former may not have been moved to 10nm yet. But that doesn't really explain other deficits by Core that Atom already solved like lack of chipset-less SoC, much more flexible core scalability etc.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Any idea why Intel's execution with the Atom family is so much better than with the Core family? Sure, the former may not have been moved to 10nm yet. But that doesn't really explain other deficits by Core that Atom already solved like lack of chipset-less SoC, much more flexible core scalability etc.

By "core scalability" I'm guessing you mean improving the architecture? Because scalability normally means being able to make multiple variations of the said CPU by adding or removing cores/caches/graphics.

I can't explain exactly why they don't move to a full SoC with the chipset on-die. My guess is because they're also a manufacturing house, they can easily use older fabrication facilities using chipsets.

While there are benefits to on-die integration of the PCH for ultrabook class laptops, not so much for higher end productivity or gaming oriented ones and definitely not for desktops.

The difference is volume of Core chips are far higher than Atom based ones, and they are usually on the older and/or lower performance processes already, which I guess justifies their decisions?

Performance-

As for why performance increases at a higher rate, there are many reasons why. The lower performing ones have the benefit of hindsight. Cutting edge architectures rely on ingenuity of the engineers to find solutions and it comes at a slower rate. The lower performing chips benefit from that since they can simply adopt best approaches of the higher performing architectures and avoid the bad ones. Most R&D projects don't come to fruition simply because it doesn't work(or work better).

Raising the bar also comes with diminishing returns. It's simply nature and physics. Top CPUs run into density, power, and frequency limits.

For Intel specifically there are issues with process. The design was gated by not having 10nm. Atom-based cores being a process generation behind meant that it avoided that issue.
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
Any idea why Intel's execution with the Atom family is so much better than with the Core family? Sure, the former may not have been moved to 10nm yet. But that doesn't really explain other deficits by Core that Atom already solved like lack of chipset-less SoC, much more flexible core scalability etc.
I think it's a bit like Ryzen vs Bulldozer. I don't wanna sound like Richie Rich, but it wasn't a revolutionary jump to 2017's current CPU tech (only in core counts, but that wasn't a technological deficit from intel's side), but it sure was revolutionary for AMD, especially compared to their FX lineup.
I know you're not necessarily talking just about the performance of the atom chips, but honestly I always thought that Atoms were just intel saying: are you people content with 2-4% better performance per new CPU launch and pay $300+ for a new quad core forever? Alright then, we'll show you what you actually get when you're not buying our expensive products: an insufferable CPU that made me physically break 2 netbooks out of pure rage.
I still think they are truly horrible, but they are of course much more improved products compared to what they used to be, than the intel desktop CPUs.
Core deficits were not solved because I think 99% of the employees themselves also believed the management the same thing for years, what the investors and the stock market also believed: everything's fine, on track, in high volume production with yields better than expected, and shipping for revenue either right now (Cannon Lake) or the next quarter (Ice Lake in every quarter for 2 years till the end of 2019). I don't think many of them were prepared with a speech like: guys, we do what we can to survive till we get our manufacturing in order, get as much upgrades into our Core (where viable) and we'll survive this with some scratches. Instead they all got a many pages long marketing material, teaching them how to say AMD is still bad, when it's not bad then ours is better, and when ours is a bit worse in some workloads, then those workloads don't count as much as MS Word or Excel, and that AMD is an unreliable supplier anyway (oh the irony).
I think the only employees who initially got the truth were the process node engineers, who got the task of improving the 14nm as much as possible to not look as bad as they really are and to be able to go at least to 10 cores with Comet Lake. That's the biggest bullet they shot themselves with till now... ICL is an utter embarrassment in a laptop to laptop comparison against 9th gen and Whiskey being on a very-very mature and impressive node, and the much praised ICL iGPU loses to Picasso in games. All this with ICL itself being a very strong architecture with the typical intel advantages being even better and shortcomings greatly improved.
All this with foveros being a brilliant innovation, which alone could have meant ICL laptops being the best ever. I know TGL is supposed to fix many of these problems, but the fact alone that they must backport the desktop part to 14nm tells us what to expect: at best, it won't be the embarrassment ICL is right now.

Poor AMD, with intel being in the shape they are, AMD couldn't position their 7nm GPUs to be disruptive even if they wanted to (only the most expensive product has a really good price/performance ratio, but I wouldn't call even that disruptive), but I understand now why they don't want to do that.

OK that was long.
 

Hitman928

Diamond Member
Apr 15, 2012
5,262
7,890
136
From the Anandtech home page today:


Anandtech basically took Intel's own numbers from separate comparisons against AMD's 3700u and then used them to compare Comet Lake against Ice Lake and Comet Lake came out on top in most tests. This is for the 15 W CPU category. Basically it seems to come down to a combination of higher turbo in burst work loads as well as having 50% higher cores/threads keeps the Comet Lake performing higher than Ice Lake in most cases despite the higher IPC of Ice Lake.
 

wahdangun

Golden Member
Feb 3, 2011
1,007
148
106
Well let's be fair: with enough capacity, big water can cool loads up to 400-500W on 14nm CPUs. It requires ridiculous cooling capacity, but it can be done. On 7nm AMD CPUs, ambient cooling reaches its limits somewhere in the 180W range, maybe less. At least that's been my experience. You get diminishing returns more quickly out of water or air cooling on those CPUs. That being said, who really wants to spec 300W cooling solutions for 14nm CPUs? Much less the board VRMs necessary to deliver that kind of power reliably?



They don't really have a choice.

with need more advancement in cooling,if just that liquid metal cooling can take off, and not so expensive.
 

DrMrLordX

Lifer
Apr 27, 2000
21,629
10,841
136
with need more advancement in cooling,if just that liquid metal cooling can take off, and not so expensive.

Liquid metal is nice, but it isn't really enough. 60-70 W/mK is not revolutionary enough to make a significant dent in the formation of hotspots on modern CPU dice.
 

phillyman36

Golden Member
Jun 28, 2004
1,762
160
106
Havent kept up with the Intel vulnerabilities and dont understand some of the info (LPDDR4 ) I want to build a second pc. I am using a Ryzen 3900x. Wonder if I should jump on a i9 9900k or wait for the 10900k. Thoughts?
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
Havent kept up with the Intel vulnerabilities and dont understand some of the info (LPDDR4 ) I want to build a second pc. I am using a Ryzen 3900x. Wonder if I should jump on a i9 9900k or wait for the 10900k. Thoughts?
I'm wondering, which aspect of the 3900X is bothering you so much that you want to get an intel CPU that is better in practically nothing - aside from benchmarking games in 720p and 1080p.

If it's just the desire of a collector, I can totally understand that and it actually sounds cool to have the fastest CPU from both vendors :) In that case, I'd definitely wait for the Comet Lake part, as it's not a dead platform at least - Z390 ends with Coffee Lake refresh (9th Gen).
 
  • Like
Reactions: guachi

phillyman36

Golden Member
Jun 28, 2004
1,762
160
106
I'm wondering, which aspect of the 3900X is bothering you so much that you want to get an intel CPU that is better in practically nothing - aside from benchmarking games in 720p and 1080p.

If it's just the desire of a collector, I can totally understand that and it actually sounds cool to have the fastest CPU from both vendors :) In that case, I'd definitely wait for the Comet Lake part, as it's not a dead platform at least - Z390 ends with Coffee Lake refresh (9th Gen).
Nothing is bothering me per say. Just want a second pc to use. I would use 1 pc for gaming while the other pc is encoding video(nothing professional just my movie collection). I want 1 AMD pc and 1 Intel pc.
 
  • Like
Reactions: spursindonesia