Advice Request: 14700k + NH-D15 + Contact Frame + Kryosheet, Thermal Throttling in R23 Multi-Core @ stock

Associate
Joined
23 Oct 2002
Posts
1,525
Location
That London, née Brighton
All running on an MSI Z790 Carbon (original, not the new MAX), with 64GB G.skill 6400 @ XMP, Windows 11; I'm throttling in multi-thread R23 and only averaging 5.3-5.4GHz with both BIOS left at stock (i.e. no effective power limit) and after setting a 288w power limit (for both 1 and 2).

It's idling at ~30c, which seems ok I think (albeit I don't know what my ambient is, but I'm central London and it's pretty chilly).

There's no crashing, but does this sound right, or does this sound like I maybe didn't do either the contract frame or NH-D15 mounts quite right? Maybe not even pressure? Maybe not enough pressure? Or, maybe my CPU is just bad? But surely it should be able to hit its advertised 5.6GHz supposed max clocks, right?

Secondly, I followed this guide and forced 5.5GHz P-core and 4.3GHZ E-cores, with manual 1.35v on the core, and this did crash in R23.
 
Last edited:
What is your fan profile? it isn't easy to run a 14700K on air cooling and hold boost clocks in R23 - I have to have my fans set to max in a case known for good cooling performance and even then it is close to the temperature limit (you can up the temperature limit to 115 but I wouldn't recommend it - better to improve cooling performance).

To run a 14700 or 14900 to their max potential in heavy multi-threading workloads needs a good AIO.
 
Last edited:
I haven't checked RPMs but the fans do spin up to real "whooshing" levels during R23. I'll check to make sure they're maxed.

Right now it's still just on the floor on top of the motherboard box, but eventually will be in a Be Quiet Dark Base 900.
 
Last edited:
Sorry yes, Multi-threaded!
Then your clocks sound more reasonable, the advertised boost clock is single core/thread. I think the stock all-core max is 5.5 Ghz, but it may drop depending on the workload. There is a chart here:


This test uses a custom-coded application that mimics real-life performance—it is not a stress test like Prime95. Modern processors change their clocking behavior depending on the type of load, which is why we provide three plots with classic floating point math, SSE SIMD code, and modern AVX vector instructions.
 
Thanks for the link!

Did read many reviews in the lead up to buying all this, but I guess I happened across more lucky/optimistic ones.

I've just now rebooted after dropping both power limits to 253 instead of 288 and, while R23 is still running, provisionally the bar is a few pixels longer still than the 288w run and it's not quite hitting 100c any more, ~95 tops. Curiouser and curiouser.

Edit: spake too soon - the still in-flight 253w R23 run is now looking to land just under the 288w one.
 
Last edited:
Edit: spake too soon - the still in-flight 253w R23 run is now looking to land just under the 288w one.
If you have air cooling, trying to hit reasonable clocks/temps with these CPUs in multithreaded benchmarks is like trying to knock a nail with a sponge hammer. Just my opinion, but eh :D
 
If you have air cooling, trying to hit reasonable clocks/temps with these CPUs in multithreaded benchmarks is like trying to knock a nail with a sponge hammer. Just my opinion, but eh :D
:D

Yeah I found a mix of "don't even bother" and "it'll be fine with a contact frame and NH-D15" when doing my research, so I guess the latter batch had got my expectations a little out of alignment.
 
Right now it's still just on the floor on top of the motherboard box, but eventually will be in a Be Quiet Dark Base 900.

That might not be helping - I have to max case and CPU fans with my 14700K to tame it in R23 and as above that is with a good case, the 540 Air, as well.

I don't see too big a hit with fans on silent profile though - only around 0.5-1% drop in multi scores compared to stock on AIO.
 
If you have air cooling, trying to hit reasonable clocks/temps with these CPUs in multithreaded benchmarks is like trying to knock a nail with a sponge hammer. Just my opinion, but eh :D

For most gaming workloads I can hold 6GHz boost on silent fan profile even with an older Dark Rock 4 Pro cooler, unless a very thread heavy game. Stuff like R23 though not a chance.

Personally bought knowing the limitations - most real world applications aren't like Cine Bench, etc.
 
Jesus :/

I'm trying all-core OC of 5.5 and even with 1.38v it's still crashing a few seconds into an R23 run, with power limit set to "off". And it's crashing before it thermal caps?! Kinda frustrating and I'm not sure what's going wrong at this point. HWinfo doesn't show any WHEA errors.

You'll probably find 6GHz boost works for 2-6 cores even on air, just won't be able to run something which maxes out all P and E cores on air and hold clocks.
 
Synthetic tests will push unrealistic load on CPU. For a short test, fine, but I wouldn’t bother too much, as I only use it to compare coolers or cases, but my usual use doesn’t go anywhere near the load of a synthetic test.
But if your use is similar to it, then air or anything less than a 280/360mm AIO will struggle. Even a custom loop shared with a high end GPU may struggle.
The NH-D15 is a good cooler, but there’s a lot of newer coolers that may perform better, and not always as good on Intel as on AMD and vice-versa.
280+ W is a lot. Using HWinfo, what is the bottleneck listed? Voltage, temperature? If you manage to dial back the voltage a bit it may help with higher clocks, but it is a long shot.
 
Have you tried resetting bios back to factory and the just enabling xmp, and then applying a negative v-core offset on the CPU?

You may find it runs cooler that way and better net performance.

Also, you've not mentioned what temps your CPU is hitting in cinebench... That's kind of important information.
 
Last edited:
I have set my 14700k with a negative offset of -0.086v in the BIOS and all is well. Could be that my 420mm AF AIO is helping.
Very quiet and decently cooled.
Point reasonably made tho, there seems to be too much vcore typically added by default.
 
Last edited:
I have set my 14700k with a negative offset of -0.086v in the BIOS and all is well. Could be that my 420mm AF AIO is helping.
Very quiet and decently cooled.
Point reasonably made tho, there seems to be too much vcore typically added by default.

Mine seems the opposite - a little high vcore by default but doesn't like decreasing it but I can get a reasonable frequency increase with only a tiny bit extra voltage. The limitation for me is cooling under heavy multi-threading.
 
Using HWinfo, what is the bottleneck listed? Voltage, temperature? If you manage to dial back the voltage a bit it may help with higher clocks, but it is a long shot.
Didn't realise HWinfo showed such info, but after resetting and doing more tests just now, it's saying "thermal" and "max turbo" for IA PerfCap, and "thermal" and "voltage" for Ring PerfCap. I did a fresh run with a -0.05v adaptive offset too which at least cleared "thermal" from Ring, so that's a good sign I think!
Have you tried resetting bios back to factory and the just enabling xmp, and then applying a negative v-core offset on the CPU?

Also, you've not mentioned what temps your CPU is hitting in cinebench... That's kind of important information.
Hadn't yet tried this but just did and, with a -0.05v adaptive things got a bit 1960s R&B and knocked 3 degrees off the temps, and it stuck at 5.5GHz for seemingly longer.

Sorry, by thermal throttling I meant it was hitting the built-in 100c limit and bouncing off that. It was doing that stock, but not quite by a degree or two with the -0.05v offset.
I have set my 14700k with a negative offset of -0.086v in the BIOS and all is well. Could be that my 420mm AF AIO is helping.
Mine just crashed in R23 at -0.08, fingers crossed my -0.07 run completes...

But overall, I think I'm getting the impression that my chip isn't terrible, it's just behaving as 14700ks tend to?
 
Mine seems the opposite - a little high vcore by default but doesn't like decreasing it but I can get a reasonable frequency increase with only a tiny bit extra voltage. The limitation for me is cooling under heavy multi-threading.

I reckon it has as much to do with the board, including LLC etc, as it says about the chip. So it depends where the starting vcore is with each chip on each board, too many variables I think to compare one persons undervolting, whose board might set a higher value, to another by just using that measure.
Just a thought :)

Mine just crashed in R23 at -0.08, fingers crossed my -0.07 run completes...

But overall, I think I'm getting the impression that my chip isn't terrible, it's just behaving as 14700ks tend to?

As above, each board, probably the same model would vary from board to board, will supply slightly different voltages to the same CPU. When you add that also as a variable it is more of a matter of what is working for you than gauge someone else's is "better" for a seeming lower vcore offset. :)
Synthetics are ok, and a guide. Just like OCCT (12.1.13) using small data sets and extreme mode to test. When running at 100% full load for the CPU the vcore voltage is at 1.18v, that's using OCCT, when I had it installed.
Some prefer to undervolt / overvolt using VF points, to ensure stability throughout the range of frequencies, the offset can be considered the legacy option. But, it has worked for me with what I'm going to do. And that was just to lower the vcore a little.
I think, IIRC, the max temp is max 71c when running Cinebench R23 multicore. AIO fans are at very low, kinda inaudible.
Perhaps you will be fine with an air cooler in actual use, different to running stability testing etc. I'm just happy enough with the AIO I use, hard to think of this right now but when its summer I'm kinda grateful for the AIO and an AL, now a RL, CPU..!
 
Last edited:
These chips are excessively overclocked out of the box/defaults. As the 13900k and 14900k are basically the same I think the same applies to both.

The 13900k I have scores 34k in r23 temps max out in the 60's and stay the same in long multi day render's (water-cooled). All I did was set the PPT to 150w max and a 0.05 offset. No fiddling no tweaking, set and forget. Asus Z690 prime, 128gb 3600 corsair ram.
Single / few core stuff hits 5.8/5.9

Letting it run wild with power does increase performance to around 39k in cinebench for double the power consumption ><300w, which is around 20% more multi core performance for a 100% increase in power, more noise and heat. Not worth it in my opinion for long term use.

To conclude as you appear to be struggling with cooling, use the power target's in the bios (PPT) and let the chip/mobo figure out what performance it can get out of that, start at 120 and work your way up (we are already in in diminishing returns after around 100w with these chips.
 
Back
Top Bottom