Intel Core i9-14900KS Draws as much as 409W at Stock Speeds with Power Limits Unlocked

[AnotherReader]

Yields are at least 80-85% for the full die, otherwise they wouldnt be releasing CPUs near 6GHz

You are conflating parametric yield loss and catastrophic yield loss (link to PDF). However, Intel also sells an enormous 770 mm^2 die: the monolithic variant of Sapphire Rapids. Given that this is sold for servers, they wouldn't need very high yields for it to make sense. Still, defect density can't be too bad or they would hardly get any functional dies out. A defect density of 0.4 per cm^2 would result in just 9.6% of the Sapphire Rapids-MCC dies being fully functional.

 

[wNotyarD]

It also says "stock".
Judging by comments in the thread this is what most people seem to have as the main takeaway.

It also says "speeds". Unlocking power while keeping speeds at stock isn't impossible, you know?

 

[Super XP]

Wow 400?
No thanks

 

[marios15]

You are conflating parametric yield loss and catastrophic yield loss (link to PDF). However, Intel also sells an enormous 770 mm^2 die: the monolithic variant of Sapphire Rapids. Given that this is sold for servers, they wouldn't need very high yields for it to make sense. Still, defect density can't be too bad or they would hardly get any functional dies out. A defect density of 0.4 per cm^2 would result in just 9.6% of the Sapphire Rapids-MCC dies being fully functional.

Not really, I am talking about the old-school metric of yields, that Intel does not disclose, I would expect defect rate to be 0.1 - 0.15 max

Their foundries might be behind TSMC, but you don't start high volume production if yields are 40%, you need 60% at least and they've been going on this node for years, it's at least 75% now even with 260mm2 dies.
If yields were very low and they had to cut-down 40-50% of their chips to make market, they would release much lower stock clocks and you would see high OC margins.

12900K was 5.2 then 13900K was 5.8 and 14900K/S are 6/6.2, meaning yields are high and you get the highest bins, unless you consider 14xxx a new design from 13xxx

Their worst 24core binned chip has 5.2 turbo

 

[londiste]

It also says "speeds". Unlocking power while keeping speeds at stock isn't impossible, you know?

It is impossible in this case. Speeds are directly related to power.

 

[wNotyarD]

It is impossible in this case. Speeds are directly related to power.

Look from another angle. With power unlocked, these Intels can go full throttle with their advertised speeds. Power limited? Yeah they can reach them, but only sparsely. In both cases clocks are at stock.

 

[InVasMani]

The 14900´s VID is trash. Same goes for the trash bin called "14700k".

And you have to deal with the VIDs beeing not the same. KS chips VIDs are aligned.

No disabling waste-cores and hypertreading would help to get that _hitty silicone to run 6 Ghz all-core.

But a 13900ks could. Without any need to tweak PLL voltages i might add.

Now imagine a 14900ks running 6 Ghz all-core without the need to turn off hyperthreading
or
wanting to use the least amount of wattage possible during gaming at 14900k speeds.

If you can, the KS is for you.

in conclusion:
The KS is for people hunting for the best p-cores possbile.
Either to overclock or to use less voltage at any given Mhz if compared to the rest of intels offerings.

And yes, if compared to the red team a 14900ks looks rather pathetic.
But that´s because intel has no imagination on how to sell it.
They easily could have disabled the waste cores completely and sell a 8x p-core only gaming beast that would loose to 60% but would win 40% by quiet a margin during gaming benchmarks.

Those 40% leads beeing almost all competetive online-games mind you. So intels marketing could have brandished that as a win over amd.

All core is a utter waste of power consumption however overall and that remains true of both AMD/Intel. It's best to linearly scale performance across per core ratio's period. There is a bit a off a larger offset between ratio's of P cores and E cores, but otherwise scale the P cores higher ratio than E core ratio's and basically linearly scale them. The only real question is if you want higher MT or ST emphasis in which case slope the E core's and P core ratio's more aggressively or less aggressively with that slight offset in between where they meet in the middle.

I found disabling HT all, none, or some a utter waste in practice on my 14700K and just dragged down MT performance. Sure the ST could bit a bit opened up with OC on P cores, but disabling or dropping E core cluster ratio will do that just as well overall and w/o sacrificing MT as heavy handed in the process. Disabling all the E cores is easily dumb though disabling 4 if you don't need or want the MT can free up power a good bit and allow you to easily push for higher ST. There is actually some merit to that, but I don't do it personally since it's a bit silly and not overly practical. Still if you've got software that won't benefit from the additional MT and can save power it's fine enough to consider doing so. Hopefully software can passively better train itself and adapt over time to recognize scenario's like that as well to save power and/or eek out higher performance more readily. I'm confident future software and hardware absolutely will at the same time with better inference capabilities baked in.

 

[AnotherReader]

Not really, I am talking about the old-school metric of yields, that Intel does not disclose, I would expect defect rate to be 0.1 - 0.15 max

Their foundries might be behind TSMC, but you don't start high volume production if yields are 40%, you need 60% at least and they've been going on this node for years, it's at least 75% now even with 260mm2 dies.
If yields were very low and they had to cut-down 40-50% of their chips to make market, they would release much lower stock clocks and you would see high OC margins.

12900K was 5.2 then 13900K was 5.8 and 14900K/S are 6/6.2, meaning yields are high and you get the highest bins, unless you consider 14xxx a new design from 13xxx

Their worst 24core binned chip has 5.2 turbo

Hmm we don't really know their defect rate; 0.1 per square cm would be pretty good. However, I agree that the trend of larger dies indicates that yields have improved significantly.

 

[absoluteUnit]

what?

if you'd ever played an e-sports game at even a middling level of competency, you'd know about the AM-DIP.

competitive players shoot for a stable 240, 360, or even 500 FPS in the 1% and .1% lows, depending on the game.

but if you only play single player titles at 4K 60-120 FPS, and don't mind inconsistent frametimes, then AMD is fine.

 

[Deleted member 237813]

Sure kid are you 16 at least?

 

[absoluteUnit]

Sure kid are you 16 at least?

i’m sending out prayers for all those who are too senile to notice frame drops

 

[AusWolf]

It also says "stock".
Judging by comments in the thread this is what most people seem to have as the main takeaway.

You have a point. We really have to define what "stock" is. Intel recommended power limits enforced? Power limits disabled? What about turbo clocks? What about different voltages supplied by different motherboards? How about cooling? Intel blower or custom loop? It's confusing.

 

[ir_cow]

While it can pull 400watts. Mosy likely it will be in the 280w range due to lack of cooling. Once it hits that thermal cap, it will just downclock.

 

[InVasMani]

I wonder if it'll be similar to 13900KS to 14900K where they dropped base power TDP 25w and offset it with a more aggressive turbo boost. Basically lower power PL1, but higher turbo ratio's slightly possibly some tweaks with speedstep as well. I don't know if perhaps PL1 will be reduced alongside maybe small boost to the PL2, but less than the reduction to PL2 alongside more aggressive turbo ratio's. Can only speculate though based off some of the past differences between similar SKU's with subtle refresh bumps to performance.

It'll be tougher to cool, but a 360 AIO in push/pull ought to do the trick pretty well I would hope or a 420 AIO push/pull at worst.

 

[londiste]

You have a point. We really have to define what "stock" is. Intel recommended power limits enforced? Power limits disabled? What about turbo clocks? What about different voltages supplied by different motherboards? How about cooling? Intel blower or custom loop? It's confusing.

It is confusing. On one hand due to the dynamic nature of clock speeds today "stock" is everything you listed and more. Clocks, turbo clocks, power limits, voltages, temperatures. This applies in the same way to Intel, AMD and other manufacturers of CPUs, SoCs and GPUs. Once you remove some limit the behaviour and result will be different.

And then there are all the shenanigans about how things are defined in spec and/or marketed.

I wonder if it'll be similar to 13900KS to 14900K where they dropped base power TDP 25w and offset it with a more aggressive turbo boost. Basically lower power PL1, but higher turbo ratio's slightly possibly some tweaks with speedstep as well. I don't know if perhaps PL1 will be reduced alongside maybe small boost to the PL2, but less than the reduction to PL2 alongside more aggressive turbo ratio's. Can only speculate though based off some of the past differences between similar SKU's with subtle refresh bumps to performance.

Read up on this trying to understand this news bit, this is what it looks like:
- 13900K has official PL1 125W and PL2 253W. Then it also has the extreme power profile of 253W/253W.
- 13900KS has official PL1 150W and PL2 253W. The power profile really is 253W/253W. Then it also has the extreme power profile of 320W/320W.
14900K/KS are likely going to follow the exact same pattern. And guess which one motherboards tend to default to.

 

[iameatingjam]

5.9ghz all pcore is crazy! I honestly didn't think intel would release a 14900ks, just because of how far the 14900k and 13900ks have pushed the boundaries of the architecture already, but I guess some people want those top of the line bins and hey no judgement there.

I am starting to get a little irritated with all the constant complaining about power draw when power limits are removed ( not necessarily anyone here). I mean until intel gets arrow lake out or whatever it ends up being, Raptor lake is what we got and we all know it uses a crazy amount of power when clocks are to the moon and power limits are off. No shit a 14900ks at full steam is going to use a lot of power. How could it have been any other way?

If intel figured out a way to make the 14900ks be just as powerful while not using much power - it wouldn't be the 14900ks, it would be a whole new generation of processor.

So this is right in line with expectations, is it not?

 

[Chrispy_]

5.9ghz all pcore is crazy! I honestly didn't think intel would release a 14900ks, just because of how far the 14900k and 13900ks have pushed the boundaries of the architecture already, but I guess some people want those top of the line bins and hey no judgement there.

I am starting to get a little irritated with all the constant complaining about power draw when power limits are removed ( not necessarily anyone here). I mean until intel gets arrow lake out or whatever it ends up being, Raptor lake is what we got and we all know it uses a crazy amount of power when clocks are to the moon and power limits are off. No shit a 14900ks at full steam is going to use a lot of power. How could it have been any other way?

If intel figured out a way to make the 14900ks be just as powerful while not using much power - it wouldn't be the 14900ks, it would be a whole new generation of processor.

So this is right in line with expectations, is it not?

What Intel is doing with 14th gen reminds me very much of the darkest days of AMD with the end of the old Bulldozer-series architecture that couldn't compete with Intel; The AMD FX-9590, a 5GHz Bulldozer (Vishera) CPU with an almost-uncoolable 220W back in the day. If you can't compete on architecture, just overclock the absolute living shit out of it and don't worry if no motherboards exist yet that can deliver 220W to the socket, Asus and Asrock will make special models for your turd of a CPU because there's always a market for fanboys who buy the company flagship no matter how bad it is!

At least competition is closer this time around; 14th gen isn't as bad as Bulldozer ever was, and if you ignore the flagship willy-waving contest, there are plenty of i5s and i7s that compete well with AM5 at more sensible power levels.

 

[AusWolf]

It is confusing. On one hand due to the dynamic nature of clock speeds today "stock" is everything you listed and more. Clocks, turbo clocks, power limits, voltages, temperatures. This applies in the same way to Intel, AMD and other manufacturers of CPUs, SoCs and GPUs. Once you remove some limit the behaviour and result will be different.

And then there are all the shenanigans about how things are defined in spec and/or marketed.

That's a real problem if you think about reviews. How do you review a CPU? You could say that it's super efficient and super cool because you run it in X motherboard, then I'll buy it, put it into Y motherboard, and conclude that it's a steaming garbage pile only because that board overvolts it like crazy, or has power limits disabled by default (which is something that reviews claim, but I've never seen it happen).

What Intel is doing with 14th gen reminds me very much of the darkest days of AMD with the end of the old Bulldozer-series architecture that couldn't compete with Intel; The AMD FX-9590, a 5GHz Bulldozer (Vishera) CPU with an almost-uncoolable 220W back in the day. If you can't compete on architecture, just overclock the absolute living shit out of it and don't worry if no motherboards exist yet that can deliver 220W to the socket, Asus and Asrock will make special models for your turd of a CPU because there's always a market for fanboys who buy the company flagship no matter how bad it is!

At least competition is closer this time around; 14th gen isn't as bad as Bulldozer ever was, and if you ignore the flagship willy-waving contest, there are plenty of i5s and i7s that compete well with AM5 at more sensible power levels.

Well said, sir!

I'm sick of all this flagship mania, honestly. Why doesn't everybody just buy the part they actually need instead of overspending on marketing crap?

 

[Chrispy_]

That's a real problem if you think about reviews. How do you review a CPU? You could say that it's super efficient and super cool because you run it in X motherboard, then I'll buy it, put it into Y motherboard, and conclude that it's a steaming garbage pile only because that board overvolts it like crazy, or has power limits disabled by default (which is something that reviews claim, but I've never seen it happen).


Well said, sir!

I'm sick of all this flagship mania, honestly. Why doesn't everybody just buy the part they actually need instead of overspending on marketing crap?

Flagship mania is a problem because manufacturers know they can sell them for absolutely bonkers profit magins, and push their flagship agenda to the press by supplying a disproportionate amount of flagship review samples.

The midrange stuff that 95% of the market will end up purchasing is barely covered, and half of the channels have to buy their own samples because they don't get given them to review. I enjoy watching Formula1 but I don't want to buy an F1 car for my commute, so as much as enjoy reading/watching some articles on the F1 cars and their development, what I really care about for myself is reviews of sensible cars that I can drive on normal roads in normal traffic.

 

[chrcoluk]

Pretty much this, flagships exist for two reasons, the rep of the products kind of moves down the stack, and those who want the best will pay the extreme prices.

For reviews, I think the whole package needs to be considered.

So if e.g. a CPU throttles on a review, because mainstream coolers cant cool it, then the reviewer should be posting the throttled results as the performance of the CPU, not doing things like raising tjmax out of spec to get a better bench score. I had a disagreement with w1zzard on this when he posted his 14900k review.

I also think bios's that ship with out of spec defaults should either be omitted from a review as a form of punishment for shenanigans or reviewed with manual settings that comply with intel/amd's spec sheet. But this requires potentially wrecking a PR relationship with the vendor which a reviewer might not want to do.

The problem with assisted (out of spec) reviews, it only further encourages the vendor to keep shipping these type of products and its misleading.

 

[Chrispy_]

So if e.g. a CPU throttles on a review, because mainstream coolers cant cool it, then the reviewer should be posting the throttled results as the performance of the CPU, not doing things like raising tjmax out of spec to get a better bench score. I had a disagreement with w1zzard on this when he posted his 14900k review.

On this point I think I side with W1zzard. You'd just buy an H-series board and i9-14900 if you intend to operate at the stock power limits - why spend extra on the high-end z-series board, and the K-series SKU tax if you're not going to overclock? They're overclocking-focused parts by design so not overclocking them is a failure to review them properly.

It's just an opinion and there's no right or wrong answer, but that's what I'd do.

 

[BoggledBeagle]

I am not sure if you guys are not overthinking it, if Intel can make say 300 14900KS CPUs a month due to the very high silicon quality needed, they will price it so that they are not getting a lot of people frustrated because they would like to buy the product but they cannot find it. They price it so that the demand and supply roughly matches.

This product is not just a branding thing, it needs to have REAL PHYSICAL PARAMETERS to qualify to be able to be sold.

 

[Deleted member 237813]

i’m sending out prayers for all those who are too senile to notice frame drops

you are an idiot and a child.

 

[chrcoluk]

On this point I think I side with W1zzard. You'd just buy an H-series board and i9-14900 if you intend to operate at the stock power limits - why spend extra on the high-end z-series board, and the K-series SKU tax if you're not going to overclock? They're overclocking-focused parts by design so not overclocking them is a failure to review them properly.

It's just an opinion and there's no right or wrong answer, but that's what I'd do.

Ok I am not saying there should be no overclocking

But if you consider undervolting is its own article, then the review would probably be three things.

1 - The core product sold as is, this runs at spec, and if things dont quite work at spec, then tough luck intel, thats reflected in the review results such as e.g. if the CPU cant be kept under tjmax and is throttling. (you might only buy this chip to o/c but it doesnt mean everyone else will, everyone I know who used to overclock never does now as an example but they still buy overclocking SKUs).
2 - Overclocked. Spec out of the window.
3 - Undervolting. Spec out of the window.

So yeah keep all the content including overclocking there, its just that the core part of reviews could be tweaked. tjmax also is not just a spec'd performance or efficiency parameter, its a safety parameter.

 

[my5xhni]

Intel's equivalent to the GTX 480, also known as the "room heater"

You can save on heating costs and have a warm winter.