Question CPU speed dropping near the end of Cinebench R23 run

[Teddydog]

I built my current PC back in Spring 2022, then added new RAM and a new GPU in August. Been running it at stock speeds up until now, but wanting to tweak it a bit now.

CPU: Intel i7-12700K
MB: Asrock Z690 Steel Legend
Cooler: Vetroo V5 air
RAM: Neo Forza MARS DDR4-4400 2x16GB
GPU: PowerColor Red Devel RX 6700 XT (testing this currently but permanent card is an MSI RTX 3070)

The RAM is supposed to be able to run at 4400 CAS 19 under XMP, but I cannot get it to run under any XMP profile. Currently have it set to 440 and its stated timings manually and Gear 2 as it doesn't want to run anywhere above 3066 under Gear 1.

I've been doing some benchmarking before starting to change anything with the CPU config to be sure it is running around expected performance for stock. I'm currently getting the following:

Cinebench R23: ST 1890-1900, MT 22100-22500
Timespy: 13000 range
Port Royal: 5600 range

Today I ran a 10-minute cycle of Cinebench MT and noticed that at start, CPU-Z was showing all cores running at x47 but around the end of the first run the temp was around 89 deg C and then they dropped down to x43 and x44 and the overall temp stayed in the 60 deg C range. At the end of the 10 minutes, the score was just above 20000.

I was surprised to see this and was wondering if this was "normal" or expected behavior? I would have thought that it would have stayed at the higher clock speeds as long as temp were not going above 90 deg C and would have found a stable level where it could run the most speed it could at that temp rather than borking down to x43/44. Is this something that the default/basic settings on my MB is doing? It doesn't seem like this is a failure of cooling either. I know that I have only a modest air cooler, but it should be fine for base timing, I think.

Note that I am planning to upgrade the cooler before any serious overclocking, but wanted to see what could be done with this setup first. Thinking a basic AIO as I'm not planning to push the OC very hard as even the base performance should be good enough for me.

Any thoughts on this? Thanks.

rob

 

[hotaru.hino]

I would say this is a cooling issue. The cooler you have I'd argue is more suited for a Core i5 or Ryzen 5.

What makes it worse is it performed last in a lot of the tests Gamers Nexus threw at it:

 

[CelicaGT]

IIRC there is a power limit timer on Alder Lake. Sounds like it could be the culprit here as some cursory reading elsewhere indicates limiters were implemented differently on different boards and BIOS. Might be a possible explanation.

 

[geofelt]

You are testing in a multithreaded environment when your intended use is likely for gaming where only a few cores will need the performance.
Run Hwmonitor. It will give you the current, minimum and maximum temperatures per core.
I suspect that you will see a few cores with 100c. in red.
That indicates some throttling.
But, unless there is a failure, that is not all bad.
The motherboard and cpu are performing to their max capability given the conditions.

There are some better coolers out there.
Noctua maintains a list of suitable coolers for different processors.
Here is the list for the 12700K:
https://ncc.noctua.at/cpus/model/Intel-Core-i7-12700K-1579
You have a number of very good air cooling options.

What is the make/model of your case and fan arrangement?
Any cooler needs a good source of fresh air to let it do it's job.

 

[Teddydog]

I started a 10-minute MT test again while watching the temps on the cores instead of just the overall. This is the result after 2 minutes when the CPU had throttled down to x43-44.

HWInfo64

To me this looks like it started downshifting the CPU when one core hit just 86 deg C.

Running a ST test, after 5 minutes the first pass was about 2/3 complete and these are the temps:

So it seems that under ST, none of the cores are hitting temp limits and they all show as running x47-49 the whole time.

Here are the Core Power Limits exceeded stats:

Note that these are currently "No" while the ST test is running, but do show that they all hit "Yes" previously. So it does seem that the limits to CPU speed during the MT test are due to power limits, not temp limits. The "thermal throttling" all say "No".

Do my R23 scores of 22362 MT and 1865 ST seem about on par for an i7-12700K in a Z690 board?

ALSO...would this post be better in the CPU area instead of the Overclock one?

rob

 

[hotaru.hino]

Considering around 22,000 was what AnandTech, TechPowerUp, various other sites did as well, yes, that's on par.

 

[uWebb429]

@Teddydog
Use HWINFO to report the limit reason data. The CPU has the ability to tell you why it is throttling.

It appears to be power limit throttling. Look in the BIOS for the PL1 and PL2 turbo power limits and max them out to 4095W. No more power limit throttling.

 

[Teddydog]

Yes, it was power limits. Now that I understand the PL1, PL2 and tau time a bit better, I could see that the stock MB settings are using the old non-K settings and pulling the power after the 56 seconds even though the temps were not getting over 80 deg C.

When I used the Intel tool to overclock it by removing the power limits (or moving them, anyway), then it went to 100 C on one core and got thermally throttled. Still got a bit higher Cinebench R23 score that way, though.

I have an MSI AIO cooler to install and once I do, I'll play around with overclocking a bit more. Should be able to run faster longer with that setup vs. the very minimal air cooler I have now.

Rob

 

[Teddydog]

I have have new results to relate and so far I'm pleased with the improvements!

The case is a 6-year-old Rosewill Star Predator mid-tower case. It came with 2 front-mounted 120mm intake fans and one 120mm exhaust fan. The front panel is probably pretty restrictive, but does have some openings with mesh and is open at the bottom with about 3/4" standoff from the floor (but it sits on carpet, so no).

With the initial setup using the Vetroo air cooler, no overclocking and default BIOS settings, I was getting about 1900 Cinebench R23 Single Thread, single pass scores. The Multi Thread single was pretty steady in the low 22,000-22,500 range for single pass. When running a 10-mintue stress test, though, the system started slowing down after about 1 minute and the full-pass score dropped down to just over 20,000. As many stated early, it was hitting the power limits for non-K i7s and throttling down.

Before updating the system with an MSI 240mm AIO setup, I used the Intel ETU to "auto overclock" by removing the power limits, as should be for a 12700K CPU. This just caused the system to hit THERMAL throttling very early and didn't help overall performance at all.

Next I just kinda tossed the AIO into the case following the pictorial instructions, which leave a lot to be desired. This setup did help some, but with the Auto OC (50x multiplier and power limits both set to 260W), the Cin R23 single pass scores didn't change by more than 1-200 points (22,300). With a 10 minute MT stress test, the system was able to stay out of thermal throttling for about 4 minutes, but ultimately it did throttle and peaked at 22,566

But I realized that just tossing the AIO into the system was NOT optimized for cooling or air flow. With my case, there isn't room for the radiator at the top and it just barely fits in front with the tubes at the top. The default fan install also sets them to push from inside the case to outside, which meant OUT the front of the case. My rear fan was also exhausting, so the only air intake was from the top mesh and maybe a little from the bottom mesh (through the carpet) and back slots. I also realized that I had plugged the radiator fans into CHASSIS headers in the MB so they were not ramping up appropriately for the heat generated in the MT stress tests (I saw they were only running about 900 RPM). With all this in mind, it is surprising that the system did any better at all!

TODAY, I reconfigured the AIO to PULL cool room air in from the front and added an additional 120mm case fan at the top, so I now have intake from the front through the radiator and then exhaust from the back and top fans. I also changed the 2 radiator fans to the CPU1 header on the MB so they can properly adjust speed.

The NEW OC, AIO and better air path results:

Cinebench R23
ST, 1x pass: 1895 (unchanged from initial results)
MT, 1x pass: 23,328 (gain of about 1000)
MT, 10 min stress test: 23,334

• CPU max temp at 5 min: 90C
• CPU max temp at 7 min: 92C
• CPU max temp at completion: 93C

As we see, not only did mild OC gain 1000 points, but the TEMPS were kept well under control even at 10 minutes of MT stress testing. NO THROTTLING was reported by HWInfo for the entire test, leading to the same score for the 10 min test as for the single pass! The radiator fans were also able to run up to 2000 RPM during the test, greatly helping with the temperature control.

I'm now feeling like I can play a bit more with the overclocking to see what I can do with it. Just because, not because I need a little more CPU power. I also might look at modifying my case front panel to allow for more airflow. Better yet, a newer better designed case!

Thanks for the early thoughts folks!