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

[IEC]

Ryzen 3000 series does not run hot or take a lot of power, this is what many of us are now used to, so yes, its a big deal that its hot and power hungry,.

To be correct, you should state that Ryzen 3000 series shows better efficiency rather than to post that it does not run "hot". This statement is not necessarily true in all scenarios and circumstances, as the increased thermal density of 7nm chips means there can be brief hotspots on the die that instantaneously measure "hot". Even if on average they are not. "Power-hungry" is also a subjective measurement. Efficiency is objectively testable.

It's totally anecdotal of course, but one review from launch last year I checked just now to refresh my memory ran a Ryzen 3950X 16-core first with unlocked power budget using precision boost at which point the processor pulled ~185W in a stress test, then at 4.35GHz all cores @ 1.35Vcore with an AIO watercooler, CPU hit 90C with radiator fans on full and burned ~220W.

You're comparing a max OC'd, overvolted R9 3950X to a stock i9-10900K there. That's another apples to oranges comparison. Your larger point about there being a subset of users willing to ignore efficiency for the sake of performance is correct, however.

 

[dmens]

And future chips regardless of manufacturer isn't going to get significantly better either since process shrinks aren't giving us the kind of power savings they used to.

It is entirely incorrect to assume all the companies designing CPU’s operate at the same level of competency and quality.

I worked at three of them, the differences are enormous.

 

[FaaR]

You're comparing a max OC'd, overvolted R9 3950X to a stock i9-10900K there. That's another apples to oranges comparison.

It's an anecdotal example of what happens when you push a high-end microprocessor - which is what Intel is doing themselves and calling it stock specs. Totally legit and within their purvey to do of course, I'm not disputing that!

What's the vcore of that 10900K at max all-core clock? Maybe not 1.35V, but it's gotta be higher than normal to hit such power draw figures as that chip does.

 

[Zucker2k]

You're comparing a max OC'd, overvolted R9 3950X to a stock i9-10900K there.

Isn't it generally accepted though, that the 10900K is pushing the limits of 14nm too , at those 'stock' speeds? Where's @DrMrLordX when you need him?

 

[DrMrLordX]

I think honestly as things get smaller, we need a better X axis heat transfer so more of the IHS can spread the heat out on more heat pipes.

Agreed, though I think TECs may actually be a way to fix the problem through y-axis cooling. Not your bog standard eBay TEC, but actually a set of tiny TECs composed of only a few thermocouples each set up in tile configuration and mounted on top of each die. AMD CPUs at least have a large network of temp sensors on each die which allows them to pinpoint hotspots (and accurately report those temps to monitoring software). All you would have to do is cycle current towards the tile on top of the hotspot to aid in rapid dissipating of heat from that locality. It would shift the hotspot away from the die and towards the IHS. TECs are usually hideously inefficient, but I suspect the power/mm2 will be low enough at individual hotspots that activating a few thermocouples in the same area wouldn't chew up that much power. It's something Intel should probably look at since I think their R&D capabilities are potentially much stronger than AMD's (at least from a budget perspective). They don't need it for 14nm, but we haven't seen their 7nm process yet.

You need to define those weasel words within some sort of bounds of comparison if you want me to buy what you're selling... My recollection is the high-end 3000 series does drink quite a lot of juice when running all cores pegged at top clock - as you would expect.

Unless you overclock their CPUs, the top-end Matisse chips top out at 142W no matter what you do. PBO is overclocking. And no, they don't come configured to run that way by default.

Isn't it generally accepted though, that the 10900K is pushing the limits of 14nm too , at those 'stock' speeds?

We'll find out in a few days when the embargo lifts! I can't wait to see what those launch UEFIs do with power limits. 250W here we come!

 

[FaaR]

Unless you overclock their CPUs, the top-end Matisse chips top out at 142W no matter what you do.

Yes, I know! But that means the chip downclocks a lot when running all cores, so isn't pushing the limits anywhere near as much as Intel's newest crazymobile does.

That's what I'm saying; when you actually push a modern bleeding edge CPU, it's not going to run cool, and it's not going to be modest with its power draw either. It's not true only for Intel and their ancient 14nm++++++++ process.

Look at what happened to AMD Vega64. You run it in power saver profile and it's actually a fairly modestly thirsty GPU, not bad at all really. Put it in turbo though and it's a different matter. It's 100+ watts for an additional 75-150MHz or so.

 

[Markfw]

Yes, I know! But that means the chip downclocks a lot when running all cores, so isn't pushing the limits anywhere near as much as Intel's newest crazymobile does.

That's what I'm saying; when you actually push a modern bleeding edge CPU, it's not going to run cool, and it's not going to be modest with its power draw either. It's not true only for Intel and their ancient 14nm++++++++ process.

Look at what happened to AMD Vega64. You run it in power saver profile and it's actually a fairly modestly thirsty GPU, not bad at all really. Put it in turbo though and it's a different matter. It's 100+ watts for an additional 75-150MHz or so.

We need to wait for benchmarks, but I bet they show that they are close to Ryzen in performance, with much more power utilization, and much more heat produced. Until then, all arguments are opinions.

MINE INCLUDED

 

[DrMrLordX]

Yes, I know! But that means the chip downclocks a lot when running all cores

Downclocks compared to what? What is your point-of-reference? My 3900x sustains 4175 MHz on all cores in a semi-intensive AVX2 benchmark like CBR20. In old games that don't tax the CPU that hard, it seems to sustain clocks of around 4.3 GHz. The only way to get it to boost any higher than that sustained is to run something purely ST like CBR20 ST or SuperPi.

 

[Topweasel]

We need to wait for benchmarks, but I bet they show that they are close to Ryzen in performance, with much more power utilization, and much more heat produced. Until then, all arguments are opinions.

MINE INCLUDED

That's the key. That's the point I was trying to explain before. Neither of these guys actually give us rated TDP. But to beat on MT a 3900x which at all core is roughly 4.1GHz maybe 4.0 after a decent cooler is heatsoaked. This will need to run at ~4.4-4.5GHz. Considering the power usage on 7nm and Intels 14nm. That's going to be like almost 50% more power. Figure the 3900x at about 140w usage at that level, we are talking about a minimum of 210w to sustain it. Good thing for Intel is that the silicon will hold up to that. But now you need a cooler for just the "standard usuage" can keep up with 200w+. That is just to beat a single CPU in MT load. Considering other things with its ratings just stock, no OC, no MCE, to hit TVB you need temps not just below thermal limits but below certain levels. Now Ryzen is probably more sensitive to hot spotting. But measured temp hits 50-60c regardless of non-chilled cooler under lightish loads. As core clocks go down and core usuage comes we don't see crazy increases in temps. But even in an uncapped enviroment its still capped at realitively close to 140w of power usuage. There are a million questions on how the 10900k is going to be hitting its rated clocks inregards to cooling requirements. We are starting to see a culmination of the issues that started when Intel started pushing away giving us turbo information (and power usage at multiple core level loads) and one of my biggest issues with the 9900k was that by basically having the enthusiast board manufacturers set crazy limits on PL2 and still rating the cpu at 90w that they were hiding the true requirements and creating curcumstances where people may buy the chip and a reasonable cooler and find their CPU underperforming and constantly up against a thermal wall. Fast forward to this one and I am starting to wonder if its another step farther, are we going to be seeing a circumstance where their isn't a normal solution at all that can provide the rated performance. What kind of cooler is it going to take that will allow people to get what Intel is advertising. Because its not even about max temps and keeping it under the thermal wall. You now have create a circumstance where you are generally 35-40c under that limit.

 

[tamz_msc]

Downclocks compared to what? What is your point-of-reference? My 3900x sustains 4175 MHz on all cores in a semi-intensive AVX2 benchmark like CBR20. In old games that don't tax the CPU that hard, it seems to sustain clocks of around 4.3 GHz. The only way to get it to boost any higher than that sustained is to run something purely ST like CBR20 ST or SuperPi.

Does Cinebench R20 use AVX2, or even AVX? As far as I'm aware it doesn't even trigger offsets on Intel CPUs while running, even though it's based on Embree which does use AVX.

 

[DrMrLordX]

Does Cinebench R20 use AVX2, or even AVX? As far as I'm aware it doesn't even trigger offsets on Intel CPUs while running, even though it's based on Embree which does use AVX.

It's based on Embree, so it should be able to use all the way up to AVX512.

 

[Falkentyne]

Does Cinebench R20 use AVX2, or even AVX? As far as I'm aware it doesn't even trigger offsets on Intel CPUs while running, even though it's based on Embree which does use AVX.

CB R20 triggers an offset on mine if I have one set. CB R15 does not.

 

[piokos]

Ryzen 3000 series does not run hot or take a lot of power, this is what many of us are now used to, so yes, its a big deal that its hot and power hungry,.

Of course Ryzen 3000 runs hot.
3900X (which 10900K competes with) under decent 240mm AiO will heat up to 40*C in idle and 80+*C under load.
Not only is this just ~10*C less than 10900K (and just ~5*C in distance to Tmax).
Just 3 years ago people used to call such temperatures unacceptable in overclocked 7700K.

So it seems the thermals some are willing to accept and get used to depend on the logo on the box. Weird.

Power hungry - absolutely, Intel will pull more because of the node difference. There's absolutely no way to mitigate this.

 

[Markfw]

Of course Ryzen 3000 runs hot.
3900X (which 10900K competes with) under decent 240mm AiO will heat up to 40*C in idle and 80+*C under load.
Not only is this just ~10*C less than 10900K (and just ~5*C in distance to Tmax).
Just 3 years ago people used to call such temperatures unacceptable in overclocked 7700K.

So it seems the thermals some are willing to accept and get used to depend on the logo on the box. Weird.

Power hungry - absolutely, Intel will pull more because of the node difference. There's absolutely no way to mitigate this.

I have 3 of them. They don't hit 80c, 70 c is highest. At least mine don't One is one air, 2 on 240 mm aio's

 

[mopardude87]

I have 3 of them. They don't hit 80c, 70 c is highest. At least mine don't One is one air, 2 on 240 mm aio's

Not gonna lie, running a 3900x on this stock cooler has been quite impressive with me usually hitting like 78-81cel depending on ambient temps. I had a short run in with a i7 8700 and i got temps very similar to that with a CM 212 Evo. Even the mere idea of all the e- waste Intel has contributing offering that rinky dinky useless cooler would trigger some i think.

If Intel offers the same or similar cooler again, i may start a petition and a riot. WATCH what happens if they do it again. Why bother to include such a piece of junk cooler with the 8700 and chips of that caliber? Are they being thoughtful and figured we wanted a ugly ass paper weight? I got a Morgan Silver dollar from 1884 minted in New Orleans if i want to use a NICE paper weight.

 

[DrMrLordX]

Has anyone seen the coolers that ship with Comet Lake-S yet?

 

[mopardude87]

Has anyone seen the coolers that ship with Comet Lake-S yet?

Grabs some popcorn, grabs some edibles cause YEAH this is going to be GREAT! Oh man i do love a good surprise, maybe they learned something?

 

[FaaR]

Downclocks compared to what? What is your point-of-reference?

Downclocks compared to max single core turbo clock. 3900X is "only" 12 cores, with a 3950X and all 16 cores active it might not reach as high an all-core turbo as what you get.

In old games that don't tax the CPU that hard, it seems to sustain clocks of around 4.3 GHz.

Yes, well of course it does, since a lot of PC games aren't heavily threaded (or doing interesting things on the CPU.) My point of concern was what happens in an all-core heavily taxed situation.

Has anyone seen the coolers that ship with Comet Lake-S yet?

No idea what it'll look like, but I'm looking forward to seeing it. Yes, I'm a bit of a cooling nerd, honestly...

If it's Intel's standard extruded aluminium spiral it will have to be at least a meter tall to handle the load I'm thinking.

 

[DrMrLordX]

Downclocks compared to max single core turbo clock.

I hate to tell you this, but that clockspeed almost never happens, and it can't be sustained through the normal boost algorithm unless you're running something like SuperPi.

Team Fortress 2 (a game from 2007!) uses maybe 2-3 threads, and produces sustained clockspeeds of around 4.3 GHz.

CBR20 MT sustains steady clockspeeds of 4175 MHz.

Blender Benchmark seems to oscillate between 4150 and 4200 MHz. Averages are around 4170 MHz.

4.6 GHz is there for bursts lasting a few milliseconds. I can record max clocks in games of 4575 MHz frequently. It just never stays there. And I'm running undervolted with low LLC (heavy vdroop) so I lost about 25 MHz in ST performance. I gained some MT performance, so I do not complain.

 

[jpiniero]

If this is gonna be the prices, it does make it look like you were better off buying the 9700K when it was $340 than any Comet Lake.

 

[coercitiv]

If this is gonna be the prices, it does make it look like you were better off buying the 9700K when it was $340 than any Comet Lake.

Not really.

Even if prices stay up with the same mark-up across the board (which would not happen anyway), the 10600K will still be considerably cheaper and offer equivalent performance. Meanwhile the non-K 10700 will be a throughput upgrade over 9700K for those looking for "moderate" power usage numbers. The 10600 offers slightly more performance over my 8700 while asking for $100 less.

 

[phillyman36]

Saw this video this morning no idea what cooler is used.

 

[piokos]

Has anyone seen the coolers that ship with Comet Lake-S yet?

The same model as before. These CPUs will pull roughly as much as 9th gen, so there's no need to replace.
Of course on 8-10 cores it'll be very loud and not even near all-core boost potential. But that was already true for 9th gen.

We should not expect Intel to invest in a better stock cooler, given that probably 90% of their desktop CPUs go to OEMs and most people getting 6+ cores for DIY use a 3rd party cooler anyway.

There are only 3 ways to make this cooler better: making it taller, using copper or getting a much better fan.
Fans and copper are expensive.
Making it taller would compromise the main reason to keep it in a DIY build: it's really small.
There's a handful of easily available low-profile coolers that could be seen as an upgrade. Most of them are so expensive that they don't make much sense for half of Intel's lineup.

 

[Makaveli]

Saw this video this morning no idea what cooler is used.

hmm Team blue is really going to be in trouble after Zen 3 launches if the IPC gains are true.

 

[piokos]

hmm Team blue is really going to be in trouble after Zen 3 launches if the IPC gains are true.

Not really. We're only talking about high-performance desktops, mostly DIY. It's a small market. Even if miraculously AMD takes it all, Intel has more important niches to take care of.
Tiger Lake will even out the notebook competition. It seems we'll also see Tiger Lake U in NUCs and SFF desktops.

In the end it's mostly about marketing and media coverage. Because people buy the brand that they hear about in media. And high-end gaming is the most covered segment. So naturally, even if you're looking for a $200 mid-range CPU, you'll buy form the company that wins in $500+ segment.