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

Page 22 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
20,242
7,887
136
Some time ago I asked if it would be possible for you to write units properly (well, it clearly is).
But the idea of putting temperatures in different units in the same sentence is just bizarre.

And yes, your 3900X is very hot. Live with it.
70c is hot ? And different units in the same sentence. In the US all temps for outside and houses are normally in Fahrenheit. CPU temps in Ryzen master are Celsius.

So what is your problem ?
 

RetroZombie

Senior member
Nov 5, 2019
464
381
96
And yes, your 3900X is very hot. Live with it.
70C is hot at full load, are you sure?
What cooler is it Markfw ?

I think the problem here is that you guys are interpreting intel reviews results badly, but i understand when the reviewer do graphs like this (and remember data obtained with a very good cooler):


Looking at this graph it seams that the intel i9 10900K measured temperatures ranging from 36ºC at idle and 70ºC at full load, but the author in the review says the following:

«When cooling the Core i9 10900K with a Noctua NH-U9S with dual fans, the core temperature was 29 degrees at idle, 56 degree average under load, and peaked at 96 degrees. The i9-10900K hitting in the 90s was only encountered during the most demanding multi-threaded workloads, but if you will be hitting the CPU often, it may be wise with water cooling.»
 
  • Like
Reactions: Gideon and Tlh97

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
20,242
7,887
136
70C is hot at full load, are you sure?
What cooler is it Markfw ?

I think the problem here is that you guys are interpreting intel reviews results badly, but i understand when the reviewer do graphs like this (and remember data obtained with a very good cooler):


Looking at this graph it seams that the intel i9 10900K measured temperatures ranging from 36ºC at idle and 70ºC at full load, but the author in the review says the following:

«When cooling the Core i9 10900K with a Noctua NH-U9S with dual fans, the core temperature was 29 degrees at idle, 56 degree average under load, and peaked at 96 degrees. The i9-10900K hitting in the 90s was only encountered during the most demanding multi-threaded workloads, but if you will be hitting the CPU often, it may be wise with water cooling.»
They are not giving that NEAR the load I am. They were hitting 90's during the type of load I do 70c. And I have 2 more cores. 240 AIO is my cooling.
 
  • Like
Reactions: Tlh97

lobz

Golden Member
Feb 10, 2017
1,194
1,156
106
5.1 requires a lot more power and heat. If you ran it the 10900K at 4 ghz, it wouldn't be drawing 250 W.
It also wouldn't get anywhere near to the 3900X, I'm not sure what's the suggestion here, while the golden sample 232W is obviously better than a higher number (duh), it's still not great.
 
  • Like
Reactions: Tlh97

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
20,242
7,887
136
It also wouldn't get anywhere near to the 3900X, I'm not sure what's the suggestion here, while the golden sample 232W is obviously better than a higher number (duh), it's still not great.
The 10900k does not even beat the 3900x when its at 5.1 on all benchmarks, just a couple. So at 4 ghz, it would lose everything.. I am still trying to buy one, but nothing at Amazon, newegg or bestbuy..

I am not paying more than $530 for this.
 
  • Like
Reactions: Tlh97

piokos

Senior member
Nov 2, 2018
322
106
76
I think the problem here is that you guys are interpreting intel reviews results badly
I think the actual problem here is that there's a group of users who are extremely eager to criticize anything Intel does and attack any forum member that isn't doing the same.
Instead, they could spend some time reading reviews more carefully or learning basic statistics.

This whole thread is about mocking a CPU for hitting 90*C. But much like with all the power consumption drama - we're talking about a peak measurement that only happens during max boost. And on a CPU that's obviously pushed to the limit and not as efficient as mainstream non-K models. But it's not made for mainstream, is it?

Seriously, it's not black magic. You can go to a review that does proper efficiency test. TPU already has results for most Comet Lake-S.

They are measuring the whole system and use Cinebench for the MT part.
System with 10900K used 36% more energy than one with 3900X. But since 3900X completes this benchmark quicker, the difference between CPUs alone is lower - probably in the 25-30% range.
So yeah, 14nm is more hungry - no miracle here. But it's nowhere near the peak consumption disparity.

And going back to temperatures, once again TPU's results:
Run at max turbo, 10900K heated up to 85*C. But left at stock it was just 54*C. All that under a $100 air cooler.
Performance gain: ~15%.

In other words: if you're petrified by 85*C but find 70*C perfectly fine, you could still run a 10900K maybe 5% slower than that Max Turbo.
but i understand when the reviewer do graphs like this (and remember data obtained with a very good cooler):

It's a box plot. If you've never had an opportunity to learn them, Wikipedia can help you.
 

DrMrLordX

Lifer
Apr 27, 2000
15,742
4,706
136
@coercitiv

As much as I respect Der8auer, that testing doesn't really show us anything other than that the 10900k has some serious bin variability. He didn't test the CPUs running stock (not motherboard OEM stock, but Intel standard Intel PL1/PL2 values + standard tau) to show us how they performed under those circumstances to show us what would differer between those samples to end users who are not going to go for static overclocks.

I suspect the "good" samples would have been able to hit slightly higher clocks within the allowed power limits, though I can't be sure since I don't have the testing resources available to him.
 

SKORPI0

Lifer
Jan 18, 2000
16,685
984
126
The 10900k does not even beat the 3900x when its at 5.1 on all benchmarks, just a couple. So at 4 ghz, it would lose everything.. I am still trying to buy one, but nothing at Amazon, newegg or bestbuy..

I am not paying more than $530 for this.
Micro Center (Chicago) showed 1 available for $529+tax in store only, it was sold out fast obviously. :confused: Best Buy also showed $529 but SOLD OUT.
What's with this "artificial" shortage on new released products?
 

SKORPI0

Lifer
Jan 18, 2000
16,685
984
126
Intel still has to service most of their market with 14nm wafers. The 10900k has a large die. Do the math. It may not be as artificial as you think.

Take this opportunity to not buy what is (at best) a mediocre product that rehashes old core design for the fourth time.
Yes, was hoping for a 10nm die , but my 8 year old current system (x79, i7 3930k, 32GB DDR3) needs a upgrade ASAP.
Maybe I'll wait a little more ...... :confused:
 

coercitiv

Diamond Member
Jan 24, 2014
3,725
3,500
136
As much as I respect Der8auer, that testing doesn't really show us anything other than that the 10900k has some serious bin variability. He didn't test the CPUs running stock (not motherboard OEM stock, but Intel standard Intel PL1/PL2 values + standard tau) to show us how they performed under those circumstances to show us what would differer between those samples to end users who are not going to go for static overclocks.
He gave us a fixed point [ 5.1Ghz, 1.25V, ISO cooling, R20 run ] on 30 different top binned CPUs from the same batch. If you don't see the value of that towards understanding the impact of variability for the most expensive consumer bin, then that's fine, we can all go back to fighting over which CPU runs hottest, likely looking for reviews where variability skews results towards desired message.
 

DrMrLordX

Lifer
Apr 27, 2000
15,742
4,706
136
He gave us a fixed point [ 5.1Ghz, 1.25V, ISO cooling, R20 run ] on 30 different top binned CPUs from the same batch. If you don't see the value of that towards understanding the impact of variability for the most expensive consumer bin, then that's fine, we can all go back to fighting over which CPU runs hottest, likely looking for reviews where variability skews results towards desired message.
The problem here is that it adds nothing to the debates about how much power this CPU can use at stock.
 
  • Like
Reactions: Mk pt

coercitiv

Diamond Member
Jan 24, 2014
3,725
3,500
136
The problem here is that it adds nothing to the debates about how much power this CPU can use at stock.
At stock this CPU uses 125-130W long-term, with an initial burst of 1 minute @ 200-230W for heavy loads. This is not something that needs further debate or measurement, it's a fact based on stock PL1 & PL2 limits combined with what we already know about power consumption of AVX loads @ stock clocks.

It's getting pretty annoying at this point to still pretend there's a "debate" over how much power 10900K uses at Intel stock settings.
 

lobz

Golden Member
Feb 10, 2017
1,194
1,156
106
Yes, was hoping for a 10nm die , but my 8 year old current system (x79, i7 3930k, 32GB DDR3) needs a upgrade ASAP.
Maybe I'll wait a little more ...... :confused:
If you're waiting for a 10nm desktop product, stockp up on groceries. It's going to be at least 1,5 years.
 
  • Like
Reactions: DarthKyrie

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
20,242
7,887
136
At stock this CPU uses 125-130W long-term, with an initial burst of 1 minute @ 200-230W for heavy loads. This is not something that needs further debate or measurement, it's a fact based on stock PL1 & PL2 limits combined with what we already know about power consumption of AVX loads @ stock clocks.

It's getting pretty annoying at this point to still pretend there's a "debate" over how much power 10900K uses at Intel stock settings.
And when its down at 125 watt, the performance is about half (or substantially lower) than at the 260 watt that all the benchmarks are based off of, correct ? This is the part I hate, as you are duped into thinking its faster than it really is, longterm.
 

Hitman928

Platinum Member
Apr 15, 2012
2,505
1,638
136
And when its down at 125 watt, the performance is about half (or substantially lower) than at the 260 watt that all the benchmarks are based off of, correct ? This is the part I hate, as you are duped into thinking its faster than it really is, longterm.
No, it's more like ~15% lower. It depends on the workload, extremely heavy workloads might be a little more than 15%, moderate workloads closer to 10%, light workloads marginal difference. Intel's 14++ nm hits a very steep power consumption curve once you get above ~4.2 GHz and especially as you approach 5 GHz.
 

DrMrLordX

Lifer
Apr 27, 2000
15,742
4,706
136
This is not something that needs further debate or measurement, it's a fact based on stock PL1 & PL2 limits combined with what we already know about power consumption of AVX loads @ stock clocks.
So you're telling me performance characteristics are going to be identical for all 30 of those CPUs running stock? When there's a delta of 70W between samples at fixed clocks? I don't buy it. Clocks will be different at the very least. Either that or Intel is using a primitive boost algo by modern standards.
 
  • Like
Reactions: Mk pt

coercitiv

Diamond Member
Jan 24, 2014
3,725
3,500
136
So you're telling me performance characteristics are going to be identical for all 30 of those CPUs running stock? When there's a delta of 70W between samples at fixed clocks? I don't buy it. Clocks will be different at the very least. Either that or Intel is using a primitive boost algo by modern standards.
Of course performance will be different, that was the whole ******* point! That's why I said this type of insight should be very interesting for all of us, but too many decided to die on this hill of 10900K hot/cold so I think I'll just see myself out.

And when its down at 125 watt, the performance is about half (or substantially lower) than at the 260 watt that all the benchmarks are based off of, correct?
Yes, substantially lower performance, anywhere between 10-20% depending on workload (frequency scaling, power usage). Definitely nowhere near 50% though.

I look forward to your tests on 10900K if they materialize, although I suspect that by 16th of June we'll get another very juicy batch of power/performance/thermals for both Zen and Comet.
 
  • Like
Reactions: Rigg

IntelUser2000

Elite Member
Oct 14, 2003
6,825
1,387
136
Intel's 14++ nm hits a very steep power consumption curve once you get above ~4.2 GHz and especially as you approach 5 GHz.
The only previous CPU that reached 5GHz was server Power chips. They also needed 250W+ to work.

Of course performance will be different, that was the whole ******* point! That's why I said this type of insight should be very interesting for all of us, but too many decided to die on this hill of 10900K hot/cold so I think I'll just see myself out.
I was surprised that they could get it that high on air cooling. That explains the variability since they are on bleeding edge of the cutting edge in clocks.

But,

is it really air cooling?
 

RetroZombie

Senior member
Nov 5, 2019
464
381
96
It's getting pretty annoying at this point to still pretend there's a "debate" over how much power 10900K uses at Intel stock settings.
Already asked, what is intel stock settings:
- Potato Dell/HP OEM pc with pathetic cooler paired with some pathetic motherboard with some low power PSU?​
- Z490 expensive motherboard, with expensive cooler and expensive PSU with default bios and memory settings?​
- Z490 expensive motherboard, with expensive cooler and expensive PSU with MCE enabled and XMP memory?​

Which results are the correct ones from the performance, power, temperatures, .... perspective?
 

RetroZombie

Senior member
Nov 5, 2019
464
381
96
This whole thread is about mocking a CPU for hitting 90*C. But much like with all the power consumption drama - we're talking about a peak measurement that only happens during max boost.
Well that's because todays software isn't pushing it into the limits yet, in 5 years what do you think it will happen, power and temperatures are going to rise up or go down?

If the software, and i meant all software from games into apps start using the cpu resources more properly it will heat up, it's better be prepared don't you think?

It's a box plot. If you've never had an opportunity to learn them, Wikipedia can help you.
Well maybe you could explain it to me, but now can you?

Here is the graph from the R9 3950X temperatures:



Now the intel version:


Why did he change the graph type?
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
20,242
7,887
136
Well that's because todays software isn't pushing it into the limits yet, in 5 years what do you think it will happen, power and temperatures are going to rise up or go down?

If the software, and i meant all software from games into apps start using the cpu resources more properly it will heat up, it's better be prepared don't you think?


Well maybe you could explain it to me, but now can you?

Here is the graph from the R9 3950X temperatures:



Now the intel version:


Why did he change the graph type?
First I am offended by piokos posting type, derogatory. Also, its its anything that puts Intel in a bad light, he seems to get offensive, and try to defend what they have done. Very biased.
 

DrMrLordX

Lifer
Apr 27, 2000
15,742
4,706
136
Of course performance will be different, that was the whole ******* point!
Now maybe you understand where I'm coming from? I want to see those 30 samples running stock, both on workloads <56s and >56s to see how the sample variance affects performance under those conditions. That would add a lot to our understanding of how this chip really works.
 

epsilon84

Senior member
Aug 29, 2010
981
688
136
My apologies if this video has already been posted, I don't have the time to check through 22 pages!


Summary for those for can't be bothered watching the whole video:

(Kraken Z63 280mm AIO, CB R20 loops for stress testing)
10900K stock - 46C
10900K 4.9GHz - 61C
10900K 5.2GHz - 91C !!
('auto' voltage)
10900K 5.2GHz - 65C
('tuned' voltage)

Moral of the story? Don't let the mobo 'auto tune' your voltage when you overclock. I can attest to this personally.

The 10900K doesn't run 'hot', but it still consumes a ton of power nonetheless, especially overclocked.

I also agree with jayz2cents comment about it taking Intel getting spanked by AMD before they even implemented these improvements to the die thickness / IHS. Too little, too late, but hey, you still have the worlds fastest 'gaming' CPU ;)
 

ASK THE COMMUNITY