Intel 10-Core Comet Lake-S CPU Could Suck Up To 300W

[RadiatingLight]

Seems crazy-high, but actually not too unreasonable.

9900KS has a 95W TDP and realistically uses like 135W

9900KS @ 5.2 on all cores uses 165W (https://www.tomshardware.com/reviews/intel-core-i9-9900ks-special-edition-review/3)

add 25% more cores = 165 * 1.25 =~ 210W

9900KS is binned, so a normal-quality chip might use 20% more power, so 210 * 1.2 = 252W

 

Probably not 300, but still WAY too high compared to AMD.

The 3970X uses 280W, and that has over 3x the cores.

 

[SpikeSpiegel]

10 Core CPU that uses 300 watts.....really?

[spartaman64]

 i think this officially applies to Intel now

[Princess Luna]

2 minutes ago, CalintzJerevinan said:

10 Core CPU that uses 300 watts.....really?

Ring Bus is still the best performance, lowest latency solution there is for connecting processor cores and cache altogether *however* the solution has poor scaling past 6 cores... this is the main reason the last time we had a 10 cores processor using Ring Bus, the i7 6950X, that processor was a rare sight and was a power house.

 

The efficiency degrades a lot at 8 cores reason why the 9900K already is less than ideal specially at such high clocks despite the more refined 14nm++ stage but forcing two more cores in there is being extremely troublesome.

 

If they manage to get it working however the i9 10900K should in theory be the best performing 10 cores processor in the market even if it's still 14nm++... *worth it or not* is up to each individual.

[spartaman64]

inb4 540w 18 core 10980xe

[Taf the Ghost]

42 minutes ago, spartaman64 said:

It was rumored a while back also but maybe it's more concrete now? If it's true then they not only delayed it but lied about it also ... I wonder if that violates any sec regulations

First issue seems to have been about a need to respin the larger die due to voltage management. (Bug in the silicon, basically.) This one going around is just a reality. We know what the power usage of 14nm Skylake is. 5 Ghz all-core for 10 cores is going to be around 300w. Skylake-X is known as a power hog because of these exact reasons.

 

What's likely going on is Intel is working with OEMs so we don't run into the issues highlighted back with the 8700 and "TDP rating". At the ~125w TDP that Intel will give this, that'll really only do Base clocks on 10 cores. Which means the 10c, at "stock", will likely get smoked by the 3900X and have trouble beating the 8 core parts at anything. Intel really only cares about the 10900k actually being at the top of the charts, regardless of using the same power as the 64 core Threadripper will. Issue is actually getting it there in anything but an open-air test bench.

[Taf the Ghost]

2 minutes ago, spartaman64 said:

inb4 540w 18 core 10980xe

Remember when people criticized AMD for the 225w FX9590? Intel apparently looked at that and went, "you know, that's a great idea!". No, no it wasn't.

[TVwazhere]

-Threads Merged-

[dizmo]

On 1/9/2020 at 1:46 PM, Zando Bob said:

^^^ I'd care even less, my CPU already eats 300W or more lol. I don't think many enthusiasts buying the top end SKU really care all that much about power consumption.

Top 10 reasons for Intel to not have backwards compatibility with older mobos, it narrows the range of fire hazards uninformed or idiotic people have access to. 

Exactly. I mean, I can see the issue in terms motherboard compatibility, but other than that the actual running cost of having a higher wattage CPU are pretty low.

Heat management on the other hand though is a little tricker...most coolers cap out at "250".

On 1/9/2020 at 1:49 PM, spartaman64 said:

idk how i feel about having to buy a noctua nh-dh15 cooler for a consumer cpu

For the highest end performance chip? You really shouldn't be running anything less.

On 1/9/2020 at 1:59 PM, AnonymousGuy said:

Give me a 600W TDP...fuck if I care with a waterblock and car-sized radiator.

Hahaha, at first I thought you were joking, and then I looked at your bulid...

[spartaman64]

1 hour ago, dizmo said:

Exactly. I mean, I can see the issue in terms motherboard compatibility, but other than that the actual running cost of having a higher wattage CPU are pretty low.

Heat management on the other hand though is a little tricker...most coolers cap out at "250".

For the highest end performance chip? You really shouldn't be running anything less.

Hahaha, at first I thought you were joking, and then I looked at your bulid...

highest end performance chip? LUL that would be the threadripper 3990x (which is has about the same power consumption as) not some measly 10 core consumer cpu or even the 3950x (which has like half of the power consumption)

[dizmo]

16 minutes ago, spartaman64 said:

highest end performance chip? LUL that would be the threadripper 3990x (which is has about the same power consumption as) not some measly 10 core consumer cpu or even the 3950x (which has like half of the power consumption)

In Intel's non-HEDT stack. You should have been able to realize that.

 

[spartaman64]

13 minutes ago, dizmo said:

In Intel's non-HEDT stack. You should have been able to realize that.

 

yes and it shouldnt have the same power consumption as the top end HEDT processor

[dizmo]

3 minutes ago, spartaman64 said:

yes and it shouldnt have the same power consumption as the top end HEDT processor

As I've already mentioned. I don't think it matters, and I don't care.

[spartaman64]

3 minutes ago, dizmo said:

As I've already mentioned. I don't think it matters, and I don't care.

you could not care but i dont think your justification is a good one and i think a lot of people would care especially if they have a 600 watt psu

[dizmo]

12 minutes ago, spartaman64 said:

you could not care but i dont think your justification is a good one and i think a lot of people would care especially if they have a 600 watt psu

Right, because people buying a $500 CPU are really going to stress over paying another $100 for a new PSU

Not only that, if it's a good PSU, that'd still be fine. It only uses 100 more watts than an OC'd 9900k.

[spartaman64]

1 minute ago, dizmo said:

Right, because people buying a $500 CPU are really going to stress over paying another $100 for a new PSU

Not only that, if it's a good PSU, that'd still be fine. It only uses 100 more watts than an OC'd 9900k.

yeah why do people spending 60 dollars on a video game care about paying 5 dollars more for a microtransation 

[dizmo]

3 minutes ago, spartaman64 said:

yeah why do people spending 60 dollars on a video game care about paying 5 dollars more for a microtransation 

Horrible comparison.

[spartaman64]

3 hours ago, dizmo said:

Horrible comparison.

you right 100 dollars is 20% of 500 and 5 dollars is only 8% of 60 dollars so i should have said 12 dollars microtransaction my bad

[dizmo]

6 minutes ago, spartaman64 said:

you right 100 dollars is 20% of 500 and 5 dollars is only 8% of 60 dollars so i should have said 12 dollars microtransaction my bad

Sure, except you have to look at it in the scope of the entire build, which would most likely make it closer to 5%.

[RejZoR]

Intel is facing what I warned about before. They are hitting the clock limits which are around 5GHz. You can't realistically make a faster CPU that would run at 6GHz or 7GHz. And pushing it on old fab so high just means you'll have enormous power consumption. It's like with overclocking. To a certain point you're increasing clock for minor voltage bumps. But you come to a point where you need to unreasonably raise the voltage to achieve desired clocks. And that's the point where you generally need to give up as you've reached the sweet spot. Pushing it higher may gain extra 100MHz, but will make tons of extra heat and power consumption. Only way to mitigate that is to use a smaller node or redesign the CPU to raise IPC at same or lower clock. AMD did both which is why they are winning. Intel hasn't done either. They run old node and their IPC hasn't changed for years. They've only been mitigating IPC through clock increases. When they finally go 10nm and stop dicking around with same rehashed Skylake architecture, they might change things.

[mr moose]

58 minutes ago, RejZoR said:

Intel is facing what I warned about before. They are hitting the clock limits which are around 5GHz. You can't realistically make a faster CPU that would run at 6GHz or 7GHz. And pushing it on old fab so high just means you'll have enormous power consumption. It's like with overclocking. To a certain point you're increasing clock for minor voltage bumps. But you come to a point where you need to unreasonably raise the voltage to achieve desired clocks. And that's the point where you generally need to give up as you've reached the sweet spot. Pushing it higher may gain extra 100MHz, but will make tons of extra heat and power consumption. Only way to mitigate that is to use a smaller node or redesign the CPU to raise IPC at same or lower clock. AMD did both which is why they are winning. Intel hasn't done either. They run old node and their IPC hasn't changed for years. They've only been mitigating IPC through clock increases. When they finally go 10nm and stop dicking around with same rehashed Skylake architecture, they might change things.

everyone knows 5ghz is a basic limit on current silicon tech,  Intel have no choice but to push harder with refinements on 14nm. Until they get their own 7nm off the ground or fix 10 for mass production not much is going to change in the short term. 

[Bombastinator]

1 hour ago, mr moose said:

everyone knows 5ghz is a basic limit on current silicon tech,  Intel have no choice but to push harder with refinements on 14nm. Until they get their own 7nm off the ground or fix 10 for mass production not much is going to change in the short term. 

I’m not sure that things even did change.  Remember ryzen is more memory hungry than i7.  It could very well be the processor is the same speed it always was, they’re just using memory better.  If chip quality silicon wasn’t wildly expensive because it needs to be cut with a saw from bars of unreasonably pure material they could make a bigger die.  There’s a methodology for condensing silicon around a rod that is less expensive.  The problem is the silicon produced is ring shaped rather than flat. The lithography would have to be reversed drum scanned onto the inside or outside of a spinning ring.  If it could be worked through we might end up with ring shaped processors though

[porina]

15 hours ago, 0head said:

Intel at this point hasnt announced their new 10th gen main stream CPUs. having a 300W TDP next to AMDs offerings seems like a darn good reason why that might be.

I think the point needs re-iterating. TDP does not equal maximum power draw. Or even typical power draw. The 300W number being thrown around is what might be taken under some circumstances. The TDP will be a LOT lower.

 

This applies to AMD CPUs also. Their stock power limit value is higher than the TDP value e.g. 88W for 65W TDP parts.

 

14 hours ago, Taf the Ghost said:

At the ~125w TDP that Intel will give this, that'll really only do Base clocks on 10 cores.

Not necessarily so. TDP is defined as the guaranteed clock you get when running on the minimal cooling supporting that TDP (implicitly, CPU running at maximum temperature). It is possible if you have better cooling to have more performance at the same power due to increased efficiency at lower temperatures.

 

14 hours ago, Taf the Ghost said:

Which means the 10c, at "stock", will likely get smoked by the 3900X and have trouble beating the 8 core parts at anything.

Keep in mind the 3900X at stock has a 142W limit, but even if the power limit were equal it will still have two advantages, one from more cores (more cores lower clock > fewer cores higher clock for multithread workloads), and it is more generally efficient anyway. Without the power limit it should generally beat the 8 core though.

 

Intel really don't have any answers in the short term if you focus towards highly threaded workloads. 

 

2 hours ago, RejZoR said:

Only way to mitigate that is to use a smaller node or redesign the CPU to raise IPC at same or lower clock. AMD did both which is why they are winning. Intel hasn't done either.

They have done both, just not for desktop yet. I think things will get a lot more interesting now that AMD have finally bought Zen 2 to mobile.

[Bombastinator]

23 minutes ago, porina said:

I think the point needs re-iterating. TDP does not equal maximum power draw. Or even typical power draw. The 300W number being thrown around is what might be taken under some circumstances. The TDP will be a LOT lower.

 

This applies to AMD CPUs also. Their stock power limit value is higher than the TDP value e.g. 88W for 65W TDP parts.

 

Not necessarily so. TDP is defined as the guaranteed clock you get when running on the minimal cooling supporting that TDP (implicitly, CPU running at maximum temperature). It is possible if you have better cooling to have more performance at the same power due to increased efficiency at lower temperatures.

 

Keep in mind the 3900X at stock has a 142W limit, but even if the power limit were equal it will still have two advantages, one from more cores (more cores lower clock > fewer cores higher clock for multithread workloads), and it is more generally efficient anyway. Without the power limit it should generally beat the 8 core though.

 

Intel really don't have any answers in the short term if you focus towards highly threaded workloads. 

 

They have done both, just not for desktop yet. I think things will get a lot more interesting now that AMD have finally bought Zen 2 to mobile.

It sounds like another acronym is needed like “MPD” or something.  Of course all the webmasters of police departments of cities beginning with the letter M would them have to do (more) silent weeping.

[porina]

28 minutes ago, Bombastinator said:

It sounds like another acronym is needed like “MPD” or something.

Part of the problem is that like in a lot of tech, we want to try to simply represent relative performance in a single value, for what can and does vary a lot in ways that a given user may or may not care about. The max power consumption when running, for example, Prime95 small FFT is one data point, but not relevant for >99% of people.

 

Actually, I think the solution kinda goes the other way. Varying power with workload is more visibly seen if you fix voltage and clocks like we used to. Now, it is more complicated. On AMD side, we generally run off a power limiter unless you bypass it by either turning on PBO or other manual overclocking. The CPU will run as fast as it can up to that power limit. There are also thermal and current limits, but with a good cooler it'll be mainly power limited. Harder workloads means lower clock, but you get the best performance at that power. Intel can do that too, but it is only commonly found in off-the-shelf systems with cheap cooling, as enthusiast builds will tend to have the power limiter disabled. You get the max (stock) performance the CPU can give, up to the limits of cooling. I think we are moving into a world where we have to get used to running off a power limiter more often, and we get more into a performance-per-watt situation.

 

Compare the power limit policy of the two sides:

Intel have recommended PL1 (=TDP) and PL2 (short term boost) values. Values are not enforced, system builders discretion. Increasing these values are NOT considered an overclock. Performance enthusiast systems (Z chipset with k CPU) would tend to default to PL2 set unlimited for power and duration, so it would never fall back to PL1.

AMD have fixed power limit called PPT (88W for 65W TDP CPUs, 142W for 105W TDP CPUs). This is on by default. Bypassing it is considered an overclock.