Question Want to improve temps on my Core i9 13900K by around 10C staying with air cooling

[Wolverine2349]

*Using Core i9 13900K with e-cores disabled and all P cores clock at 5.6GHz with VCORE static set to 1.35 at LLC 6 on an MSI Z690 MEG Unify X motherboard.

*Using a Noctua NH-D15S CPU cooler with 2 140mm fans running at 1000 RPM each and Thermal Grizzly Kryonaut paste and also the Thermalright LGA 1700 BCF frame

*Using Be Quiet Silent Base 802 case with 3 140mm Silent Wings 3 intake fans all at 1000 RPM and 1 rear exhaust Silent Wings 3 140mm 1000 RPM fan (Using mesh front panel with dust filter removed for better air flow though have top with seal panel on as no fans up there and want to dampen weird noise such as mild coil whine and such)

Now when running Cinebench R23 full load, temps hit 90C and average 90C and only go as low as 89C and peak as high as 93C and sometimes 94C which is very disappointing as Cinebench R23 is supposed to be more realistic all core workload. This is in an ambient room of 19C which is around 67F.

My case is as clean as can be with only RTX 4090 Gaming OC and CPU cooler and CPU. No HDDs nor SSDs in main area and only 2 M.2 SSDs of course in the mobo and only 2 SATA SSDs on back of case so not in way at all of air flow.

I would like to lower them around 10C staying on air cooling with 1000 RPM fans during Cinebench type run at my settings

Would it be possible to change air coolers or thermal paste and lower the temps. Would the Be Quiet Dark Rock Pro 4 be a better fit for my config. I know research suggests almost no temp difference between Dark Rock Pro 4 and NH-D15S. However does the Dark Rock Pro 4 or maybe Assassin 3 have a better cold plate for better contact with LGA 1700 socket using the Thermalright LGA 1700 BFC frame?? Or no difference??

Or is there just no way to cool these things any better on air and they run hot no way around it at mid 5.X GHz speeds even with e-cores off. I would have thought with the superior binning and refinements from Raptor Lake that running those extra speeds would take same power as Alder Lake and same tempos. But no it seems to just run hot hot hot.

I can only imagine when summer comes and my AC keeps my room at 21C at best and even tougher to cool.

The P Out value goes to like 190 watts during Cinebench R23 run and CPU package power goes to like 195 to 210 watts. I would have thought with the larger heat spreader on LGA 1700 and cooling capacity of 250 watts for NH-D15S and Dark Rock Pro 4 temps would be low 80s

 

[Phaaze88]

TL;DR: Your goal might not be possible short of delidding the cpu, going to direct die cooler application, and balancing intake - exhaust a bit better.


The DRP4 has a greater focus on low noise operation, so I have doubts it would do better than D15 on performance. The old Big D(s) are still used in cooler comparisons today, due to how well balanced they are in performance to noise.
The user friendly mounting hardware and free support to a new motherboard socket is also of great value.
Assassin 3 is stronger, yes, but it does so with more aggressive fans. Once you've tuned it to your comfort zone, it won't be much better.

LLC 6? Out of what? No need to go over the medium/middle setting on any board, unless you're LN2 overclocking; that extra voltage overshoot is a necessity there.
While a flat-ish voltage curve is desired, using higher levels than necessary will yield higher power and thermals for the sake of it.

*Using Be Quiet Silent Base 802 case with 3 140mm Silent Wings 3 intake fans all at 1000 RPM and 1 rear exhaust Silent Wings 3 140mm 1000 RPM fan

'So much' intake, and not much exhaust to exchange it out with. They're both equally important.
This is with the overall stronger NF-A15 fan not being taken into account.

I would have thought with the larger heat spreader on LGA 1700 and cooling capacity of 250 watts for NH-D15S and Dark Rock Pro 4 temps would be low 80s

A 360, 280, and 240mm radiator are supposed to be adequate for up to 350, 300, and 250w loads respectively - that's all on paper. What you get in practice is different.
The current cpus have greater thermal density and more layers than gpus. You may have noticed, systems with 360mm AIO/CLCs struggle with these i9s, yet a big ass air cooled gpu handling a 3090Ti/4090 with ease.
The higher density of smaller process nodes makes the chips harder to cool/get the heat away, and adding layers reduces cooling efficiency. Cpus have two more layers than gpus: the heatspreader and the TIM between the heatspreader and die(s).

However does the Dark Rock Pro 4 or maybe Assassin 3 have a better cold plate for better contact with LGA 1700 socket using the Thermalright LGA 1700 BFC frame?? Or no difference??

You could be waiting some time, or forever, for a solid answer to this, as someone would need to have these coolers, the frame, that cpu, take the time to do such comparisons, and then post their experience here.

 

[Paperdoc]

The most obvious limit on what you can achieve to lower the CPU temperatures is your insistence that your fans must be fixed at 1000 RPM. If you simply change your CPU_FAN configuration Profile setting to allow it to do its default automatic temperature control, it will adjust the fan speed for you according to cooling needs as determined by the temperature sensor inside the CPU chip.

 

[Wolverine2349]

The most obvious limit on what you can achieve to lower the CPU temperatures is your insistence that your fans must be fixed at 1000 RPM. If you simply change your CPU_FAN configuration Profile setting to allow it to do its default automatic temperature control, it will adjust the fan speed for you according to cooling needs as determined by the temperature sensor inside the CPU chip.

Even putting CPU fans at max 1500 RPM on NH-D15S only improves the temps like 1 or 2 or maybe 3 degrees for a peak on CPU

 

[Wolverine2349]

TL;DR: Your goal might not be possible short of delidding the cpu, going to direct die cooler application, and balancing intake - exhaust a bit better.


The DRP4 has a greater focus on low noise operation, so I have doubts it would do better than D15 on performance. The old Big D(s) are still used in cooler comparisons today, due to how well balanced they are in performance to noise.
The user friendly mounting hardware and free support to a new motherboard socket is also of great value.
Assassin 3 is stronger, yes, but it does so with more aggressive fans. Once you've tuned it to your comfort zone, it won't be much better.

LLC 6? Out of what? No need to go over the medium/middle setting on any board, unless you're LN2 overclocking; that extra voltage overshoot is a necessity there.
While a flat-ish voltage curve is desired, using higher levels than necessary will yield higher power and thermals for the sake of it.


'So much' intake, and not much exhaust to exchange it out with. They're both equally important.
This is with the overall stronger NF-A15 fan not being taken into account.


A 360, 280, and 240mm radiator are supposed to be adequate for up to 350, 300, and 250w loads respectively - that's all on paper. What you get in practice is different.
The current cpus have greater thermal density and more layers than gpus. You may have noticed, systems with 360mm AIO/CLCs struggle with these i9s, yet a big ass air cooled gpu handling a 3090Ti/4090 with ease.
The higher density of smaller process nodes makes the chips harder to cool/get the heat away, and adding layers reduces cooling efficiency. Cpus have two more layers than gpus: the heatspreader and the TIM between the heatspreader and die(s).


You could be waiting some time, or forever, for a solid answer to this, as someone would need to have these coolers, the frame, that cpu, take the time to do such comparisons, and then post their experience here.

Well LLC6 as there is always VDROOP and often too much to keep t stable. Build Zoid had said 2nd or 3rd highest LLC is good to get balance between some vdroop but not too much. And there are 8 LLC levels.

It was not stable in Cruncher and spit out WHEA errors. on the 3rd test (toughest heat and thermals without upping VCORE too 1.35V at LLC6. Temps will hit 100C in Y Cruncher and that is ok and understandable, but wanted to ensure rock stability if temps were no object. Even opened the window to cool it to prevent throttling during Y Cruncher.

But CInebench is not so extreme and expected better temps.

I am picking up a Phanteks Enthoo Pro 2 later today from a Facebook seller and going to see if it improves my thermals. Its a bigger case and has a back PSU chamber so much more openness so airflow should be better I would think while keeping fans at 1000 RPM.

 

[Phaaze88]

Build Zoid had said 2nd or 3rd highest LLC is good to get balance between some vdroop but not too much.

Could you link where you got that from? Because the suggestion to use around medium setting also came from him:

Spoiler

View: https://www.youtube.com/watch?v=NMIh8dTdJwI

I am picking up a Phanteks Enthoo Pro 2 later today from a Facebook seller and going to see if it improves my thermals. Its a bigger case and has a back PSU chamber so much more openness so airflow should be better I would think while keeping fans at 1000 RPM.

That's not going to be any better, if not worse. Bigger isn't always better.
For fans, that means more time for fan air pressure to DECREASE from the extra distance. Smaller cases have an advantage because the fans are 'right there', in closer proximity to the components that need the air.

 

[Karadjgne]

Larger case = larger volume, that's all. Requires more fans to exchange that air or ambient temps go up, for air coolers.

Your problem is your cooling. It's inadequate. That cpu can hit 300w stock, and you overclocked it. A 250w+ cpu air cooler isn't going to do the job.

You need a 420mm or custom loop.

 

[TerryLaze]

Your problem is your cooling. It's inadequate. That cpu can hit 300w stock, and you overclocked it. A 250w+ cpu air cooler isn't going to do the job.

You need a 420mm or custom loop.

The P Out value goes to like 190 watts during Cinebench R23 run and CPU package power goes to like 195 to 210 watts. I would have thought with the larger heat spreader on LGA 1700 and cooling capacity of 250 watts for NH-D15S and Dark Rock Pro 4 temps would be low 80s

He overclocked it down, from 300W to 210W...

 

[Karadjgne]

No difference, coolers need to be 1.5-2x capacity of the heat output, if you want good temps. 210w on a 250w cooler is still close enough to a 1:1 ratio to be seeing temps in the low to high 90's in full core burns, which that OC has done by locking the cores. If Op had left core management open, he'd be fine except in applications like blender or Prime95 who Do use all available threads.

 

[Phaaze88]

OP disabled the e-cores...

 

[Karadjgne]

E-cores only get used during applications the scheduler feels would benefit from that kind of application. If the E-cores are disabled, just means the scheduler is useless as all applications will use P-cores instead. So for stuff like blender and/or cinebench there can possibly/probably be a loss in performance in multi-core scores.

It'll only be in P-core apps like games where the single thread master will benefit from the OC, not from the locked cores, which does nothing but add heat, but from the increased clocks.

Back in the day, i7's had 4c/8t and games were using upto those limits, so an all core locked OC was a benefit both ways, but today's games are relatively little changed in use but today's cpus are hitting upto 20 threads, so OC'ing 12-14 unused threads is absolutely useless waste of energy, creating way more heat than required or necessary.

Until games change, raising turbo limit clocks is a far better option than locked core OC. Less power consumption, less heat generated, better temps, same (if not higher) single thread clocks.