Discussion Comet Lake Intel's new Core i9-10900K runs at over 90C, even with liquid cooling TweakTown

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

dmens

Platinum Member
Mar 18, 2005
2,271
917
136
No. That could be called "product dominance" (if you really need to use this word).
Basically, you've just confirmed what he said: Intel and AMD make similarly good architectures and actual advantage (chip size and power draw) is provided by the node.

In other words:
Intel vs AMD: pretty much a draw
Intel vs TSMC: currently solid lead by the latter.

You don't have the process details to make that claim, so stop trying. Getting more performance with less area and power is an absolute metric of technical excellence.
 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
?
Why compare to FX? Why not to Palomino?

Take Zen+, add 10% IPC. How would that compete with current Intel 14nm?

It wouldn't be bad, but not exactly drastically leading (especially in laptops).

They had a massive crisis for few years. They got back to the expected curve. I'm not going to praise them for that. Sorry.

They're making good CPUs. It's their business. They're making money (or should be). I'm not going to build shrines and write poems. :eek:

The way you wrote it came across as AMD is doing well largely because if TSMC. AMD still had to develop the Zen cores. Had TSMC 7nm not been available, then Zen 2 would just look different. Maybe they get enough out of Samsung? Worst case they could still make Zen 2 on 12nm, just without doubling the L3 cache most likely. Probably no 16 core chips with. TSMC is one part of AMD's recent success, but not the largest reason.

Yes they were in serious trouble for a long time. During that time Intel milked that market. You don't need to build shrines or write poems. But you should be grateful there is competition again. The new i3's will be what i7's were just a few short years ago. You know, the same core/thread count an i7 had from 2011-2017? That is no coincidence.

This seriously reminds me of the days intel has prescott and amd had athlon.
Everyone was raining on prescott even calling it boycott or the ever warming space heater, and the Athlon X2 was so hot people thought AMD had a chance to over take intel.

Then out of nowhere a completely different node archtech was being experimented on and toss'd on laptop processors.
That laptop processor happend to be what was called Banias, which later evolved to be called Dothan.
Dothan was tucked under everyone's eyes because everyone was looking at Athlon X2, and the all mighty Opteron 180.
Then Dothan evolved to what was Penryn and completely smashed AMD so bad it wasn't even funny.
Penryn other name is the allmighty Core 2 Duo.

What i do not understand is Intel had this ellegant road map all the way to 4nm with new litho designs so taught to us by IDontCare.
What i do not also understand is how intel decided to scrap those roadmaps and say it was not profitable.
Was was not profitable? Was it because the FAB's were expensive or was it not possible because Intel saw the End of the Road for Penryn.

Until Intel starts selling there FAB's and decide not to FAB inhouse anymore, i think that is when we can probably say Intel has gone down a black hole.

Otherwise im willing to bet there is another Penryn hiding behind closed doors, they are waiting to pop out on us, but now is just not the right moment as its not profitable, nor is the public ready for it, and they are feeding us the end scraps and fat off penryn until this new chip comes out.

Because these new chips feel exactly like that... the leftover scraps and EOL improvements on a dying arch.

Who know how things could've worked out back then if Intel didn't blatantly cheat in benchmarks and buy off OEM's. That's not to hold AMD blameless. They sat on K8 for too long when anyone could see that Intel's mobile line was a sleeping giant. They even made socket 478 to 479 conversions for desktops. This time around we have to wait and see how the Cove cores do.
 

Panino Manino

Senior member
Jan 28, 2017
813
1,010
136
Otherwise im willing to bet there is another Penryn hiding behind closed doors, they are waiting to pop out on us, but now is just not the right moment as its not profitable, nor is the public ready for it, and they are feeding us the end scraps and fat off penryn until this new chip comes out.

Intel isn't hiding anything, they are just unable to play for the moment.
This story will not repeat. Of couse Intel has a lot of good architectures waiting their time to shine, but AMD isn't repeating the same mistakes. What made that turnaround so brutal was that AMD stumbled in a lot of problems with unexpected faults, delays in the fabs, and finally regression of IPC. Nothing of this is happening or will happening.

Intel may still take the lead again, but the situation will end with AMD very close and as a true alternative.

This is how this market should be. No massacres, parity, with each one offering it's advantages.
 

geegee83

Junior Member
Jul 5, 2006
23
13
66
The way you wrote it came across as AMD is doing well largely because if TSMC. AMD still had to develop the Zen cores. Had TSMC 7nm not been available, then Zen 2 would just look different. Maybe they get enough out of Samsung? Worst case they could still make Zen 2 on 12nm, just without doubling the L3 cache most likely. Probably no 16 core chips with. TSMC is one part of AMD's recent success, but not the largest reason.

Yes they were in serious trouble for a long time. During that time Intel milked that market. You don't need to build shrines or write poems. But you should be grateful there is competition again. The new i3's will be what i7's were just a few short years ago. You know, the same core/thread count an i7 had from 2011-2017? That is no coincidence.

Who know how things could've worked out back then if Intel didn't blatantly cheat in benchmarks and buy off OEM's. That's not to hold AMD blameless. They sat on K8 for too long when anyone could see that Intel's mobile line was a sleeping giant. They even made socket 478 to 479 conversions for desktops. This time around we have to wait and see how the Cove cores do.

Is there any new “sleeping giant” this time round?
 
  • Like
Reactions: spursindonesia

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
@Rigg

Seems like someone's already on it then.



If AMD can offer an R5 4600 @ $199 with an all-core turbo of 4.3 GHz and a 65W TDP, Intel really is in trouble.

Not that important at all, much cheeper R5 3600 will continue to sell like mad in big numbers.Zen 2 was a decisive blow for Intel on Desktop or HEDT.

Intel si lost Desktop/HEDT CPU market, and it will not get back in dominant position.

Zen 3 APU will give Intel decisive blow on Mobile, where is very big market or hardware sales.Renoir APU is just a start, so yes on Mobile Intel will be in very big trouble.

 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
Is there any new “sleeping giant” this time round?

Nothing as obvious as last time around. The Pentium M was in use for awhile and made continual advancements before becoming Core 2 Duo. At this point we can't see how Willow or Golden Cove will work out. Willow Cove could be interesting.

Of course, if you ask some other people, ARM is the sleeping giant. It's already a giant and fulfilling plenty of needs though.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,708
3,554
136
The normalized for frequency metric is flawed. So, Intel wins overall. What a surprise
The normalized frequency metric isn't flawed. The only thing that is flawed is you constantly harping on it how it is flawed. Fact of the matter is, Zen 2 is a superior core than Skylake(5-7% higher PPC).
Intel dominates gaming across the board, not just high fps gaming.
How many people have an RTX 2080 and above and drive 1080p displays at 240Hz?
 

piokos

Senior member
Nov 2, 2018
554
206
86
You don't have the process details to make that claim, so stop trying. Getting more performance with less area and power is an absolute metric of technical excellence.
But it's TSMC's "technical excellence", not AMD's. That's what I said.
The final product is admittedly better, but without 7N it would be a Zen+ with some IPC improvement (losing core count and power consumption gains).

Do people now get offended when someone says AMD and Intel have equally good architectures? Seriously?
How did we get here? :/
 

piokos

Senior member
Nov 2, 2018
554
206
86
TSMC is one part of AMD's recent success, but not the largest reason.
Making CPUs on a lot more dense node is essential to the leadership they have now.
Discussing if AMD's work is worth 37% or 81% is totally pointless. If they stayed on 14nm, they would have a Zen++.
Server chips would stop at 32-cores and miss all the recent cloud and HPC wins AMD had with 64-core models.
3900x and 3950X would not exist. 3700X would pull 200W. APUs would stay at 4 cores and mobile SoCs would probably leave Zen of a more purpose-built architecture.

Would it be worse than what Intel makes? Absolutely not.
Yes they were in serious trouble for a long time. During that time Intel milked that market. You don't need to build shrines or write poems. But you should be grateful there is competition again.
That's just weird. Why would I be grateful for a company to do it's job properly? They're making money!

AMD should be this active all the time. They've given us the finger for 10 years and now people are grateful for comeback. That is just bonkers.
It's as if you paid someone to clean your house every few weeks and he didn't turn up for half a year. And when he finally did come, you're suddenly grateful because he helped you find car keys that sunk in dirt.
The new i3's will be what i7's were just a few short years ago. You know, the same core/thread count an i7 had from 2011-2017? That is no coincidence.
And how did AMD improve their CPUs on offer between 2011 and 2017? :D

Because you're accepting AMD selling the same stuff for 5+ years and at the same time criticizing Intel for doing the exact same thing.
The only difference: Intel went into this hiatus with very good CPUs so they kept selling, while AMD got lazy much earlier - so they already had something outdated and it didn't sell.
 
  • Love
Reactions: Zucker2k

itsmydamnation

Platinum Member
Feb 6, 2011
2,743
3,071
136
But it's TSMC's "technical excellence", not AMD's. That's what I said.
The final product is admittedly better, but without 7N it would be a Zen+ with some IPC improvement (losing core count and power consumption gains).

Do people now get offended when someone says AMD and Intel have equally good architectures? Seriously?
How did we get here? :/
because you said something factually wrong.
Zen2 on 14nm LP would offer more performance per clock at less power per clock. that is a fact stated by AMD themselves. So put up and prove them wrong!

this isn't the slide i was looking for where they make a direct reference to the amount of power they lowered per clock process independent
but its a start
 
Last edited:

piokos

Senior member
Nov 2, 2018
554
206
86
I can name so many tech companies that by that logic should be massively successful right now but aren't.

SGI?
Nortel?
compaq ?
Nokia

i could just keep going.
These companies were competing in markets with multiple entities and no strict IP entry lock.

So in case of Nokia: anyone with enough money could say "let's make phones". And as we moved from the boring traditional headset (that you used for calling and SMS) to smartphones (that suddenly were a part of ecosystems), market was dominated by companies with more general consumer electronics experience.
It's not just Nokia either. The move from old phones to smartphones purged the whole market. It's a different product, so it's being made by different companies.

It's a lot different with x86 CPUs. Intel not only is by far the largest manufacturer (none of the companies you mentioned had this kind of lead). It's also one of just 2 significant players.

Once again: if Intel was making much worse CPUs for a long time but AMD could fulfill just 20% of demand, we would still buy mostly from Intel - because we need CPUs and no one else can join this market or rapidly expand from a lower place (like Apple or Samsung did).
Zen2 on 14nm LP would offer more performance per clock at less power per clock. that is a fact stated by AMD themselves. So put up and prove them wrong!
That's a completely different situation. And yes, Zen2 is more efficient than Skylake even without the node advantage.

If both companies stayed on 14nm, Intel would launch their modern architectures to compete with a 14nm Zen2.
Both Sunny Cover and Willow Cove are more than capable of competing with AMD's stuff.

They didn't give us 14nm "Ice Lake" because it makes little sense at this point. They have to focus on the 10nm lineup to tackle what AMD does on 7N. Backporting the architecture and launching new product lineup is an expense they decided to delay until Tiger Lake / Rocket Lake. They still sell all the Skylake stuff they are able to make. There would be absolutely no gain from updating it.
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,743
3,071
136
These companies were competing in markets with multiple entities and no strict IP entry lock.

So in case of Nokia: anyone with enough money could say "let's make phones". And as we moved from the boring traditional headset (that you used for calling and SMS) to smartphones (that suddenly were a part of ecosystems), market was dominated by companies with more general consumer electronics experience.
It's not just Nokia either. The move from old phones to smartphones purged the whole market. It's a different product, so it's being made by different companies.

It's a lot different with x86 CPUs. Intel not only is by far the largest manufacturer (none of the companies you mentioned had this kind of lead). It's also one of just 2 significant players.

No offence you just shown your complete ignorance, im just going to pick one because i can be bothered doing them all.
So at peak ATM who competed with Nortel? Ever been inside a carrier core network from about 1995 to 2005, who owned all the patient on advanced optical networks? Cisco was the toy play thing in the corner at that stage.

Also did Symbian never exist in your universe, what about the N-Gage, E90 or N95? I worked in a phone store during the initial 3G to iphone 1 days, you are so wrong its not funny no one came close to nokia we easily solid 10:1 Nokia to everything else combined across all market price points.

That's a completely different situation. And yes, Zen2 is more efficient than Skylake even without the node advantage.

If both companies stayed on 14nm, Intel would launch their modern architectures to compete with a 14nm Zen2.
Both Sunny Cover and Willow Cove are more than capable of competing with AMD's stuff.

They didn't give us 14nm "Ice Lake" because it makes little sense at this point. They have to focus on the 10nm lineup to tackle what AMD does on 7N. Backporting the architecture and launching new product lineup is an expense they decided to delay until Tiger Lake / Rocket Lake. They still sell all the Skylake stuff they are able to make. There would be absolutely no gain from updating it.
This has nothing to do with what i said or what i quoted, what you said was flat out wrong! Simple AMD got 50% of there performance per watt uplift themselves, process independent.
 

naukkis

Senior member
Jun 5, 2002
701
568
136
Zen2 on 14nm LP would offer more performance per clock at less power per clock. that is a fact stated by AMD themselves. So put up and prove them wrong!

Zen2 would not have been possible with 14nm process. Even with 7nm process they had to cut L1i cache to half to be able to double op-cache, and AMD clearly stated that 256 bit fma units would have used too much power on 14nm process.

It's a process development which drives cpu archs forward. Intel did have advantage before which cpu engineers used to make good cpu's - now they are pretty much stalled on development.
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,743
3,071
136
Zen2 would not have been possible with 14nm process. Even with 7nm process they had to cut L1i cache to half to be able to double op-cache, and AMD clearly stated that 256 bit fma units would have used too much power on 14nm process.

It's a process development which drives cpu archs forward. Intel did have advantage before which cpu engineers used to make good cpu's - now they are pretty much stalled on development.
AMD cut the L1i because of two reasons, one they changed the design they didnt go 32k 4 way, they went 32k 8 way, big difference. 2. floor plan, you have to fit the things you want in the spaces you have. Zen2 wasn't a redesign , lets wait and see what they do with Zen3 when they can do a complete new floor plan from scratch if they wish.

Where did AMD clearly state that 256bit fma would use to much power? just have an AVX offset like intel.
 

piokos

Senior member
Nov 2, 2018
554
206
86
No offence you just shown your complete ignorance, im just going to pick one because i can be bothered doing them all.
So at peak ATM who competed with Nortel? Ever been inside a carrier core network from about 1995 to 2005, who owned all the patient on advanced optical networks? Cisco was the toy play thing in the corner at that stage.
I really don't know much about Nortel, so I'm afraid I can't comment on that (that's why I focused on Nokia).
I live in Europe and other companies were active here - like Diebold, who is AFAIK leading this market right now globally.
Also did Symbian never exist in your universe, what about the N-Gage, E90 or N95? I worked in a phone store during the initial 3G to iphone 1 days, you are so wrong its not funny no one came close to nokia we easily solid 10:1 Nokia to everything else combined across all market price points.
Nokia didn't lose because they were making worse phones. They lost because - precisely - they were making phones.
And we moved from phones to Pip-Boys. Android and iOS are attractive because of the ecosystem they bring, not because of the OS itself.
Symbian was good, but it was designed for phones - it was still focused on calling, messaging and (if you could afford the data plan) basic web browsing.

iOS and Android were developed as universal, internet-connected systems from the ground up.
If you couldn't afford mobile internet, they lost most of the appeal. That's why many people delayed getting their first smartphone. I got mine in late 2011.
Simple AMD got 50% of there performance per watt uplift themselves, process independent.
Zen2 vs Zen+? Nope.
 
Last edited:
  • Like
Reactions: controlflow

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
The normalized frequency metric isn't flawed. The only thing that is flawed is you constantly harping on it how it is flawed. Fact of the matter is, Zen 2 is a superior core than Skylake(5-7% higher PPC).
Then the tests should be performed at 5GHz, and let's see which architecture will need ln2, and if the scores will remain the same across the board.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
That's a completely different situation. And yes, Zen2 is more efficient than Skylake even without the node advantage.
It wouldn't. Even on 12nm, the 2700x was pulling 200W+. Zen 2 cores on 12nm would be huge and won't clock anywhere near 4GHz. People can undervalue TSMCs impact here all they want but without their progress in node development, and Intel's stagnation, crucially, things would have been far different.
 

piokos

Senior member
Nov 2, 2018
554
206
86
It wouldn't. Even on 12nm, the 2700x was pulling 200W+. Zen 2 cores on 12nm would be huge and won't clock anywhere near 4GHz. People can undervalue TSMCs impact here all they want but without their progress in node development, and Intel's stagnation, crucially, things would have been far different.
Well yes, AMD would probably have to make a larger socket for a 12/14nm Zen2. It might have been challenging to fit 8 cores (since even on 7N the 8-core chiplet is 80mm2 large).

The biggest issue I have with this AMD/TSMC confusion is that people started to treat these companies as one. Or even worse: to think that AMD is a leader of this partnership.
So they automatically assume that TSMC's 5N or 3N will make future AMD CPUs even better.

The reality is that of course TSMC is dominating this deal and it currently rips AMD on 7N pricing - taking all the profits.
So instead of cheering this situation, AMD fans should probably hope for some diversification - even at the cost of making AMD's lineup slightly less efficient on average.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
They were both tested at the maximum frequency each of them were designed to run at under normal conditions. That should be enough.
Exactly. I thought this debate had already been settled? IPC alone means jack. What matters is this: IPC x FREQUENCY. This is the performance the average person gets at stock speeds. You negate Intel's frequency advantage and reward AMD's fatter/wider, more resourced cores by normalizing to frequency. That's a flaw that disregards design strategies.
 

Zucker2k

Golden Member
Feb 15, 2006
1,810
1,159
136
This has nothing to do with what i said or what i quoted, what you said was flat out wrong! Simple AMD got 50% of there performance per watt uplift themselves, process independent.
You were calling someone ignorant and you typed this? I suppose they tested that 50% ppw uplift on their own process? Or, was that on paper? Intel has about three more advanced architectures waiting for the right process to put them on, but alas!
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
This is the same issue that makes some people struggle to understand why Intel continues to grow despite AMD offering better value.

AMD offers more than a "better value". That's why Intel is in trouble right now. When AMD comes out with Zen3 and all Intel has to counter them is Comet Lake, that just makes things worse for Intel.

But at this moment AMD can only make enough chips to cover ~5% of server and ~20% of consumer market.

According to whom? AMD isn't running their own fabs anymore. You think TSMC can't keep up with market demand? The absolutely can.

They plan to match AMD this year in mobile and next year in servers.

Can we just be honest with ourselves here and admit that Intel is not regaining technical dominance in the server market this year? IceLake-SP was intended to compete with Rome, not Milan. As for planning "to match AMD this year in mobile", I think you are ignoring the significance of Intel needing to actually do that. Mobile is one of AMD's lowest-priority development targets. Why is Intel struggling there?

High-end gaming desktops, being a low priority segment, has to wait until 2023 or something like that.

Yes, we know Comet Lake-S isn't going to be the high-end gaming CPU of choice this year.

Not too long ago, I suggested Intel will cut prices, IF THEY HAVE TO, in order to remain competitive.

Price cuts will only take you so far, and they can do irreparable harm to the brand. You just can't keep cutting prices over and over while you're getting whipped in performance and power draw. Intel has better business relationships that allow them to sell decidedly inferior hardware in bulk, so they won't go into a tailspin the way AMD did in 2011, but eventually price-cutting turns you into the also-ran discount brand. As AMD learned, that does not get you a ton of sales.

Otherwise im willing to bet there is another Penryn hiding behind closed doors,

I'm not. They can have all the great designs they want behind closed doors. Until they get their process crap figured out, they're cooked.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,708
3,554
136
Exactly. I thought this debate had already been settled? IPC alone means jack. What matters is this: IPC x FREQUENCY. This is the performance the average person gets at stock speeds. You negate Intel's frequency advantage and reward AMD's fatter/wider, more resourced cores by normalizing to frequency. That's a flaw that disregards design strategies.
If IPC alone means jack then why does it feature prominently in Intel's marketing slides for Ice Lake?
 

coercitiv

Diamond Member
Jan 24, 2014
6,151
11,676
136
You were calling someone ignorant and you typed this? I suppose they tested that 50% ppw uplift on their own process? Or, was that on paper?
You might want to take a step back and think twice about this: comparing product vs. product ppw does give the compounded value (process + arch), but since pure node ppw uplift is a known quantity for the CPU maker, isolating pure arch benefit to ppw is somewhat trivial.

Other than that I have nothing to add to these arch vs. arch discussions, I just come to this thread to read about CML-S.