13700K low frequencies / Cinebench score

[DrySaucisse]

Hello all!

I have concerns about my 13700K frequencies and Cinebench score.

First of all, here is my build:

• CPU: Intel Core i7 13700K
• Motherboard: MSI z790 Tomahawk WiFi DDR5
• RAM: G.Skill 32 Go DDR5-6000
• CPU Cooler: Noctua NH-D15
• Thermal Paste: Arctic MX-4
• Case: Fractal Torrent

(The rest is left out since it's not relevant for the topic).

I recently noticed high temperatures on my 13700K (they may have been there for a while honestly, I did not run any Cinebench R23 before with this build).
When I start the benchmark, temperatures go to 100°C almost instantly.

So I stumbled upon many threads about undervolting. This is what I changed in the BIOS:

• PL1, PL2 to 253W
• ICCMax to 307A
• Turbo Enhance Disabled
• IA CEP Disabled*
• CPU Lite Load Mode 6
• CPU VR Voltage Limit 1.400

*IA CEP Disabled since it tanks performance in Cinebench if left on with Lite Load mode 6.

My temperatures dropped to about ~85°C while running the benchmark, and my CB R23 Score is around 29k+, which I find a bit low compared to the ~30-31k I see online.
Also, while running at full load, my frequencies are lower than they should be according to Intel specs:

• P-Core: 5.087 Ghz (all of them, but they should go to 5.4Ghz); sometimes while out of the benchmark I have two cores at 5.387 Ghz.
• E-Core: 3.990 Ghz (should be at 4.190 Ghz, they are at 4.090 out of benchmark).

I've also noticed that RING: Max VR Voltage, ICCmax, PL4 is flagged at Yes and IA: Max Turbo Limit to Yes.

I've ordered TG Kryonaut thermal paste and a Thermalright contact frame to see if it can help with the temperatures... I should get them by tomorrow.

Here is my HWiNFO while running Cinebench R23:

Do you think I have a hardware issue here?

Thanks in advance!

 

[FlyingScot]

My initial reaction would be to say that your system is behaving normally given your settings and hardware. It‘s possible that you can optimize further, but I don’t see any cause to suspect that you have a degraded CPU, if that is a concern of yours. The mere fact that you can undervolt quite aggressively is a good sign of CPU health, although Cinebench R23 is not a true stability test.

You have a great PC case, so you have a good foundation. You also have one of the best air coolers.
You should probably list your graphics card because it can be an impediment to airflow. You should also tell us your complete BIOS update history so we can understand the role that your Intel microcode is playing.

In summary, you are already doing the right thing when it comes to trying to reduce voltage and temperatures. The lower they go, the higher your CPU frequencies will boost.

BTW, did you use the Alder Lake/Raptor Lake adapter plate with the NH-D15? In the case of older NH-D15’s from 2020, etc., you had to request the free adapter plate kit from Noctua.

 

[DrySaucisse]

Hello FlyingScot and thanks for your reply!
I'm relieved for my CPU!

Here are the info you asked:

GPU: Nvidia RTX 3080 Founders Edition
BIOS update history: I can't recall exactly but I think I did something like: 7D91vHC (Microcode 0x125), 7D91vHD (Microcode 0x129), 7D91vHE (Microcode 0x12B) and finally 7D91vHF a few days ago.
I did order a NM-i17xx-MP83 mounting kit back when I bought the motherboard because I could not find my invoice >.>.

Sadly repasting with Kryonaut and installing a contact frame (the Thermalright one) did not do much, maybe a few degrees C (3, 4) at max at idle...
I checked and the frame is not wobbling or anything, it's in place and the PC POST normally...

After looking at Citay's guide, I lowered PL1 and PL2 to 210, and removed the CPU VR voltage limit.
I tried to run at Lite Load 5 but OCCT crashed (the windows just closes) during a Linpack test.

With Lite Load to 6, without changing the process priority and HWInfo opened, I got a 29586 in Cinebench for a 10 minute run.
You'll find a screenshot of the sensors mid run attached.

Also, I did not touch the LLC yet, it's still at default (I think it's 8).

I tried pushing PL1 and PL2 to 220 but the temperature rise too much.

 

[FlyingScot]

Your experiences sound about right to me. Shaving off 2 - 3C is hard work. You did good. In regards to your LL setting, it's not unusual to find that LL6 is about as low as you can go and remain stable with real stability tests like OCCT. Again, I think your CPU is quiet average. If you did experience any degradation early on (i.e. before you did all the Intel mitigation microcode updates) then the effect is relatively minor.

If you see my signature below, you will see a link to my Raptor Lake Settings Survey. It's pinned to the top of the Intel motherboard forum page. It's easy to download and datamine, including highlighting the entire table of values and then sorting. I have attached a sorted version based upon the CPU. I highlighted relevant information in yellow. Settings-wise, you could take a look at what other users did to get above 30K in R23. All of the most important BIOS settings are included in my survey, which then dumps into the spreadsheet.

Raptor Lake Survey Database (sorted version)

 

[DrySaucisse]

Thanks for the survey link!

I managed to get a 29920 points in CB without changing my settings.
But when comparing with the others results from the survey, I find my VCore way higher.

I tried with an adaptive VCore offset of -0.135, LLC5 with the same limit, but my score lowers (29455).
However my VCore is also much lower.

I did not test stability either, I just wanted to see the effects on the CB score.

 

[FlyingScot]

Thanks for the survey link!

I managed to get a 29920 points in CB without changing my settings.
But when comparing with the others results from the survey, I find my VCore way higher.

I tried with an adaptive VCore offset of -0.135, LLC5 with the same limit, but my score lowers (29455).
However my VCore is also much lower.

I did not test stability either, I just wanted to see the effects on the CB score.

The LLC5 settings you tried would likely benefit gaming Vcore at the expense of full multicore workloads like R23. You can now try mixing and matching LL, LLC and adaptive offset to benefit your intended usage case. The differences can be minor, but sometimes meaningful - assuming you can stabilize your test settings across a broad range of activities.

Typically, gaming workloads are below 150W - outside of momentary shader compilations - so PL1 and PL2 don’t need to be that high for games. Definitely think about how you actually intend to use your PC and don’t get too hung up on CPU-only benchmark apps. Unless, of course, video rendering is your thing.

 

[DrySaucisse]

Hey FlyingScot,

Thanks again for the reply!
I changed the settings to this:

PL1, PL2: 200W
LLC: 5
AC/DC: 0.40/0.40
Vcore offset: -0.135
CEP: Enabled

I did a shader compilation, 5 CB15 Extreme runs, 2 OCCT 20 minutes runs (Linpack and CPU stability test), with no errors.

I managed to get 29989 in CB (this morning, this afternoon it's fluctuating around 29600 - 29700, so it might just be a lucky one).

I'm only doing gaming on this PC, so I ran the same benchmark (Monster Hunter Wilds, one of the least optimized game known to man). Here are the results:

With LL5, no offset:
Average FPS: 123.98
CPU Package temp: average: 64°C, max: 79°C
Vcore: average 1.341, max: 1.346
Power draw: average 97W

With the settings above:
Average FPS: 123.53
CPU Package temp: average: 61°C, max: 74°C
Vcore: average 1.292, max: 1.304
Power draw: average 87W

I think I'll stick to those settings from now, as the difference in FPS is not noticeable.

 

[FlyingScot]

Well, it certainly looks like you’re doing a bang up job at undervolting and bringing down temperatures! Although, you might find that you need to adjust that Adaptive Offset at some point if/when you run into stability issues. Also, keep an eye out for WHEA errors. They can be a good indicator that you’re running on the ragged edge of stability. I like to use the CALC.exe to pop up on the desktop any time a WHEA error occurs. That way, I know exactly what activity was running. In my case, I still get 1 - 2 WHEA errors while playing one specific game for an hour or more. I haven’t narrowed it down yet; it could just be a bug. Anyway, the system has been rock solid for two years. So had I not set up my WHEA alert method, I would probably never know what I had been doing when the error occurred. I mean, who routinely checks for WHEA errors on a seemingly stable system, right?

If you're interested, I have attached my WHEA error alert guide. And I also attached my Custom Event Viewer script (Windows 10) to quickly show WHEA errors in the logfile (see attached xml). Hopefully, the xml will work fine in Windows 11, too.

 

[DrySaucisse]

Oh thanks a lot for the guide, since I won't have HWInfo opened 24/7 from now .

I can confirm that the view can be imported in Windows 11 (I just imported it .

Again, thanks so much for your time and explanations, really appreciated!

 

[citay]

Good job all around in this thread. Just wanted to quickly pop in for one small hint: If you only ran Cinebench with HWinfo open at the same time so far, try running it without HWinfo, because that being open can cost you a few hundred points alone sometimes. Without it you might crack the 30K.