14900k - Tuned for efficiency - Gaming power draw

[Dr. Dro]

...voltage and heat to push it further, than it should be.

The 14900K, specifically, is an aberration that should never have seen the light of day. It takes the concept of the i9-13900KS and applies it haphazardly to a mass production processor that hasn't undergone a selection process, and without granting it extra TDP headroom, the result is that you have a CPU that looks nicer in bursty loads such as short benchmarks than it actually is, with little regard for long-term stability.

The method you're using is a bit meaningless pertaining to efficiency because there's no meaningful load on the processor. The utilization percentage can be quite misleading sometimes. I suggest a heavy workload like Cinebench 2024 instead. Even with your recipe (4.8 P-, 4.0 E-, no HT, no graphics), it's gonna take ~140 W to sustain that during a CB24 run, and this is considering I also applied a -0.200V offset to try and shoot the curve to the moon.

 

[BoggledBeagle]

The 14900K, specifically, is an aberration that should never have seen the light of day. It takes the concept of the i9-13900KS and applies it haphazardly to a mass production processor that hasn't undergone a selection process ... with little regard for long-term stability.

I am not sure what you mean, because Intel test the same pieces of silicon and make all kinds of CPUs from it depending on how functional the pieces are and what frequencies can they sustain.
13900KS and 14900K are possibly from the same bin of fully functional dies capable of running at those frequencies.

BTW I do not believe that these CPUs can survive many years working at those stock frequencies/temperatures.

The method you're using is a bit meaningless pertaining to efficiency because there's no meaningful load on the processor. The utilization percentage can be quite misleading sometimes. I suggest a heavy workload like Cinebench 2024 instead. Even with your recipe (4.8 P-, 4.0 E-, no HT, no graphics), it's gonna take ~140 W to sustain that during a CB24 run, and this is considering I also applied a -0.200V offset to try and shoot the curve to the moon.

I think that applying negative voltage offset is not a good practise at all, because the voltage excess is there to guarantee the stability of the CPU.

I cannot test the power draw of my CPU without an offset at this moment, but say it is 170W - that is not that bad for a 24 core CPU and the high performance. At those moderate frequencies the CPU is also acceptably energy efficient.

 

[Dr. Dro]

I am not sure what you mean, because Intel test the same pieces of silicon and make all kinds of CPUs from it depending on how functional the pieces are and what frequencies can they sustain.
13900KS and 14900K are possibly from the same bin of fully functional dies capable of running at those frequencies.

BTW I do not believe that these CPUs can survive many years working at those stock frequencies/temperatures.


I think that applying negative voltage offset is not a good practise at all, because the voltage excess is there to guarantee the stability of the CPU.

I cannot test the power draw of my CPU without an offset at this moment, but say it is 170W - that is not that bad for a 24 core CPU and the high performance. At that moderate frequencies the CPU is also acceptably energy efficient.

I'll go out on a limb and say that they aren't the same bin at all, unless Intel's yields are just bat insane on this node. It's a mature node but it'd mean that practically 90%+ of the allotment of CPUs are fully functional, low-leakage and high-clocking, which is just insanity to believe IMHO. As far as longevity, I think they'll do mostly fine.

Agreed on the excess voltage, but only to an extent. Past a point, excess voltage won't do anything positive to stability, it may actually either lower the efficiency factor or actually introduce some instability because of the mismatching range. -0.200 was perfectly fine as far as a few CB24 runs on my 13900KS at the settings you mentioned, maybe yours will crash, maybe it won't, the only way to know is try.

 

[AusWolf]

Isn't intel still on 10nm technically, just branded as 'intel 7?' Though I guess the whole measuring in NM thing is getting kind of muddy either way.

Anyway, considering how far back the node is compared to AMD, its quite impressive intel is able to get the performance that they do, I guess that gap is bridged through extra power draw. Well, while at load anyway. They're still really good at power draw during idle/low load.

Intel was good with idle power even in the 14 nm+++++ era. I suspect that even AMD would be good if not for the chiplet design. The cores consume extremely low amounts of power. It's the infinity fabric and memory controller in EXPO mode that throw idle efficiency out of the window.

 

[BoggledBeagle]

I'll go out on a limb and say that they aren't the same bin at all, unless Intel's yields are just bat insane on this node. It's a mature node but it'd mean that practically 90%+ of the allotment of CPUs are fully functional, low-leakage and high-clocking, which is just insanity to believe IMHO.

I still do not follow you - the majority of these chips is used for incomplete and lower clocking products! I guess only a fifth or less of the total number of chips produced needs to be fully functional. Their product portfolio also must reflect the distribution of quality of available chips.

Agreed on the excess voltage, but only to an extent. Past a point, excess voltage won't do anything positive to stability, it may actually either lower the efficiency factor or actually introduce some instability because of the mismatching range. -0.200 was perfectly fine as far as a few CB24 runs on my 13900KS at the settings you mentioned, maybe yours will crash, maybe it won't, the only way to know is try.

-0.2 V is a huge offset, in fact, I believe that most CPUs at stock frequencies would not function reliably with this sort of voltage offset.

 

[Dr. Dro]

I still do not follow you - the majority of these chips is used for incomplete and lower clocking products! I guess only a fifth or less of the total number of chips produced needs to be fully functional. Their product portfolio also must reflect the distribution of quality of available chips.


-0.2 V is a huge offset, in fact, I believe that most CPUs at stock frequencies would not function reliably with this sort of voltage offset.

Which is why the rarer they become the more expensive they get, and the 14900K is actually cheaper? There seems to be no shortage of them either, same volume as the 13900K, but at the same time 13900KS's weren't sprouting everywhere. Just food for thought.

I imagine it's a huge offset but, wasn't a big deal here at least for those clocks.

 

[ratirt]

Which is why the rarer they become the more expensive they get, and the 14900K is actually cheaper? There seems to be no shortage of them either, same volume as the 13900K, but at the same time 13900KS's weren't sprouting everywhere. Just food for thought.

MSRP for the 13900k and 14900k is the same $589. Maturity of the silicon comes with lowering production costs due to refinement and makes it better quality, increase yields and enables the CPUs to clock higher due to voltage increase (unless you hit the ceiling for the voltage. Everything has its limits you now) since the silicon can handle it but that is with the cost of power as always. You don't see shortages of the 14900k since it's the same silicon as 13900K so no shock there to be honest. That is mostly why people call the 14th gen a blunder due to difference in frequency only.

 

[BoggledBeagle]

Ratirt: Dr. Dro was speaking about the KS model of 13900, which was supposed to have much more limited availability and higher price than the normal K model.

I do not think that Intel has now many problems with supply on the DYI PC market, data from Mindfactory show that at least in this shop they sold just 10% of all CPUs, while AMD sold 90%, this is a worst number for Intel ever.

 

[Panther_Seraphin]

Intel at the moment are struggling to find a way to compete with AMD without a drawback coming back to bite them.

It is similar to when it was AMD Bulldozer/Piledriver vs Intel Core 2/3/4xxx. Only thing AMD had at the time was price. Intel DOESNT have that at the moment.

 

[ShrimpBrime]

Intel at the moment are struggling to find a way to compete with AMD without a drawback coming back to bite them.

It is similar to when it was AMD Bulldozer/Piledriver vs Intel Core 2/3/4xxx. Only thing AMD had at the time was price. Intel DOESNT have that at the moment.

That's because AMD FX was slower than first gen core eye processors.

Intel competes very directly to AMD these days. Visa versa.. the pricing isn't horrible. They used to release "extreme" chips well over 1000$. All the way back to socket 939 single cores, top AMD cpus where also like 1000$ IE: FX-53/55/57.

Both chip makers push their hardware to the bleeding edge, both with power usage and frequency.

So gamers can have a few more FPS? Cause people want productivity?

Oh wait. Everyone in this thread have not actually shown an energy effecient chip. We are comparing top end chips like a 14900K and think we should expect power efficiency of a notebook cpu.

 

[BoggledBeagle]

14600K and 14700K have a very good price to performance ratio. Gamers Nexus named 14600K the most balanced CPU on the market and I agree with that.

I just tried how does my ECO 14900K (set with limits of 4800 and 4000 MHz and HT off) behave under a really heavy load, and with Cinebench R23 it draws 157W, runs at 1.125V (stock, no offset) and max temp was 61°C with a small air cooler on it (17.5°C ambient). Score with HT off is 31440, with HT on it would be few thousands higher.

Once I forgot the HWinfo running for a few hours and the CPU voltage range it recorded is 0.699 - 1.214V (with above mentioned settings).

 

[ShrimpBrime]

14600K and 14700K have a very good price to performance ratio. Gamers Nexus named 14600K the most balanced CPU on the market and I agree with that.

I just tried how does my ECO 14900K (set with limits of 4800 and 4000 MHz and HT off) behave under a really heavy load, and with Cinebench R23 it draws 157W, runs at 1.125V (stock, no offset) and max temp was 61°C with a small air cooler on it (17.5°C ambient). Score with HT off is 31440, with HT on it would be few thousands higher.

Once I forgot the HWinfo running for a few hours and the CPU voltage range it recorded is 0.699 - 1.214V (with above mentioned settings).

Why not some testing at base clocks.
That's 3.2ghz P
That's 2.4ghz E
Cores.

These numbers at base in comparison to stock vs user desired settings?

 

[Dr. Dro]

Why not some testing at base clocks.
That's 3.2ghz P
That's 2.4ghz E
Cores.

These numbers at base in comparison to stock vs user desired settings?

I mean, you can use ThrottleStop to manually define the desired wattage, and the CPU will do the rest

How "stable" the clocks will be? It will depend on your specific sample's binning quality and the voltage applied to the CPU.

 

[ShrimpBrime]

I mean, you can use ThrottleStop to manually define the desired wattage, and the CPU will do the rest

How "stable" the clocks will be? It will depend on your specific sample's binning quality and the voltage applied to the CPU.

No, I don't use throttle stop. Well not typically, I have used it.

All this being discussed is completely the opposite of what my normal is. I've had my B660 board for like 2 years and had no idea it was under-volt protected until recently. Lol. What a dumb adventure. My cure would be to inject micro code to eliminate UV protection.

But I use it for overclocking.

I'm opposed to let CPU do what it wants. You guys are talking all these different wattages and settings. I'm used to seeing 300w without E-cores. To put that into perspective, 1.40v. Heavy LLC.

You know, like this. Totally not effecient. (?)

 

[BoggledBeagle]

Why not some testing at base clocks.

Base clocks are really slow and performance suffers too much, I do not want to test something I would never use...

.... I'm used to seeing 300w without E-cores. To put that into perspective, 1.40v. Heavy LLC. You know, like this. Totally not effecient. (?)

Jesus, you run all P cores at 6 GHz? You should probably start a thread "How to make Intel CPUs sizzle" or something like that.

 

[phanbuey]

Intel at the moment are struggling to find a way to compete with AMD without a drawback coming back to bite them.

It is similar to when it was AMD Bulldozer/Piledriver vs Intel Core 2/3/4xxx. Only thing AMD had at the time was price. Intel DOESNT have that at the moment.

Maybe if you're going for a 13600KF or a 13700kf used etc. They definitely don't if you're going 14th gen though.

 

[ShrimpBrime]

Base clocks are really slow and performance suffers too much, I do not want to test something I would never use...

Jesus, you run all P cores at 6 GHz? You should probably start a thread "How to make Intel CPUs sizzle" or something like that.

Ah, mate. They'll go more than 6ghz. That was left over from doing a SuperPi 32m competition on our team forum.

That was run at 2 cores 6.7ghz and a time of 4m 6s and a little change. I plan to revisit the ddr4 board. Dropped it to the floor last time I was cleaning it up.

Saving power doesn't interest me unfortunately. But I know how to make the cpu fast

 

[chrcoluk]

It's a dumb choice. Let's be real.

Guess I am dumb then as I am currently running on my desktop with HT off, as there is no performance from it just wasted power. I have a power schema which allows HT cores to run, just in case I come across a workload they are needed. I do have the opinion e-cores obsolete HT.

Do you also call overclockers, PBO, and curve optimiser dumb, If not whats the difference? all are tuning the hardware post install.

What I do call dumb is claiming HT is half of the CPU, they are effectively virtual cores that allow a second thread to be scheduled on a real core which in specific workloads (only some) can yield a small performance boost. I think the absolute best case for HT on a legacy CPU would be around 40% in CPU processing bandwidth, but realistically for most people closer to 10%. Its main benefit was actually getting rid of scheduling bottlenecks. On these hybrid chips there is now actual extra 8 real cores, which not only drops these 40% and 10% numbers down, but they are a better substitute for dealing with scheduling bottlenecks.

This is why I had stayed out of the thread for a while, almost half of it is AMD fan boys telling Intel owners they are idiots for tuning their kit and instead should just buy something else, the sad thing is a TPU reviewer I have respect for has took part in this childishness as well, and is a sad state of affairs for this forum.

Just as a reminder this is not a intel vs AMD thread, it is not a what should I buy thread either. Whats happened is plain and simple derailing. If there was a ignore thread button I would be clicking it right now.

 

[dgianstefani]

BTW I do not believe that these CPUs can survive many years working at those stock frequencies/temperatures.

Intel extensively tests their parts to run for a minimum of 20 years.

 

[Toothless]

Guess I am dumb then as I am currently running on my desktop with HT off, as there is no performance from it just wasted power. I have a power schema which allows HT cores to run, just in case I come across a workload they are needed. I do have the opinion e-cores obsolete HT.

Do you also call overclockers, PBO, and curve optimiser dumb, If not whats the difference? all are tuning the hardware post install.

I never said anyone was dumb. I said it was a dumb choice. Don't be twisting words or meanings or adding in things that aren't relevant to the thread.

You seem to control your hyper threading for when it's needed. To be fair that's actually pretty cool but for some that's just more unnecessary work over getting a lower chip and calling it good. Point of the thread is disabling part of the chip, not PBO, curve optimizer, etc.

 

[chrcoluk]

I never said anyone was dumb. I said it was a dumb choice. Don't be twisting words or meanings or adding in things that aren't relevant to the thread.

You seem to control your hyper threading for when it's needed. To be fair that's actually pretty cool but for some that's just more unnecessary work over getting a lower chip and calling it good. Point of the thread is disabling part of the chip, not PBO, curve optimizer, etc.

Yes I have done it via power schema settings and bios is on defaults. I can turn it on live on a whim by switching schema.

Obviously prior to buying the 13700k, I had no experience with e-cores, so there is performance things I didnt learn until after using the chip, things that dont really get covered in reviews. In addition, chips without HT also have less p-cores, there is no Raptor lake (refresh) that has 8 p-cores, and e-cores and no HT support, the product doesnt exist.

 

[ShrimpBrime]

Yes I have done it via power schema settings and bios is on defaults. I can turn it on live on a whim by switching schema.

Obviously prior to buying the 13700k, I had no experience with e-cores, so there is performance things I didnt learn until after using the chip, things that dont really get covered in reviews. In addition, chips without HT also have less p-cores, there is no Raptor lake (refresh) that has 8 p-cores, and e-cores and no HT support, the product doesnt exist.

Can you explain this with better detail?

Last I checked, you can prioritize usage, but shutting off HT requires a cold boot. Meaning F10, PC shuts off, turns on and shuts off HT and then restarts again. Same thing for core disablement.

 

[chrcoluk]

Can you explain this with better detail?

Last I checked, you can prioritize usage, but shutting off HT requires a cold boot. Meaning F10, PC shuts off, turns on and shuts off HT and then restarts again. Same thing for core disablement.

Windows has dozens of hidden power schema settings, you can fine tune a crap ton of stuff.

Google a tool called powersettingsexplorer (no spaces). It will widen your eyes when you see whats in there.

There is settings under the category "processor power management".

Setting:
  Processor performance core parking min. cores for Processor Power Efficiency Class 1

Description:
  Specify the minimum number of unparked cores/packages allowed for Processor Power Efficiency Class 1 (in percentage).

It defaults to 0 for all schemas.

Also the same for max cores. Efficiency class 1 means p-cores.

Setting it to 50 means the first core on each physical core, so basically all physical cores but no HT.

Windows seems to prioritise core unparking in the following order, preferred cores (the two high clocking cores), rest of physical cores, HT cores for preferred cores, then finally rest of HT cores.

The ones in bold are included in a setting of 50.

The bad news is the core parking settings dont work on Windows 11, there is a discussion about this elsewhere right now, the current common opinion is that the thread director exclusive to 11 overrides it. I can do it as I am using Windows 10. Its being worked on to try and find a way to disable the thread director to gain control on 11.

 

[ShrimpBrime]

Windows has dozens of hidden power schema settings, you can fine tune a crap ton of stuff.

Google a tool called powersettingsexplorer (no spaces). It will widen your eyes when you see whats in there.

There is settings under the category "processor power management".

Setting:
  Processor performance core parking min. cores for Processor Power Efficiency Class 1

Description:
  Specify the minimum number of unparked cores/packages allowed for Processor Power Efficiency Class 1 (in percentage).

It defaults to 0 for all schemas.

Also the same for max cores. Efficiency class 1 means p-cores.

Setting it to 50 means the first core on each physical core, so basically all physical cores but no HT.

Windows seems to prioritise core unparking in the following order, preferred cores (the two high clocking cores), rest of physical cores, HT cores for preferred cores, then finally rest of HT cores.

The ones in bold are included in a setting of 50.

The bad news is the core parking settings dont work on Windows 11, there is a discussion about this elsewhere right now, the current common opinion is that the thread director exclusive to 11 overrides it. I can do it as I am using Windows 10. Its being worked on to try and find a way to disable the thread director to gain control on 11.

Try quickcpu for core parking. Works fine in W11.

The rest is essentially setting priority as I had thought because windows can't literally "turn off" HT, it can be selected for non use.

I get it, thanks for the explanation!

 

[tugrul_LordOfDrinks]

Just for comparison, my ryzen 7900 gets around 18000 points in cinebench at 44Watts. At stock 65Watts its 24000 points and 30000 points at 180Watts (on 15C-20C ambient). Imo its like exponentially increasing waste of electricity so it's wiser to lower performance a bit during gaming to achieve much better energy consumption.

You can simply tune the RAM timings to regain the lost 5-10 FPS with only 1-2 Watts extra on RAM.