CPU Core Clock Minimum drops 600 -700MHz during AIDA64 Memory Benchmark

[JayDeezy]

Hope everybody is well. Thanks for your help!

Just prior to the most recent update to HWINFO 64, and since I installed the update I have noticed a weird reading in the Minimum Core Clock column that I had never before seen. My 9900k is locked at 5000MHz in BIOS with Speedstep, SpeedShift, and all C states disabled. All of the Turbo Timing windows are set to the maximum and the Vcore is static. I do run with an AVX offset of -1. In windows, I have had the Power Plan set to Ultimate with each individual option within the Plan maxxed out. Obviously CPU Clock variation periodically drops a few Mhz although I also increase the BCLK to 100.4 within Windows using the Asus program TurboV. So the Minimum reading should never be less than 4908MHz or so when Windows detects AVX running. I was seeing readings as low as 4208MHz and didn't see how that was possible. Then I realized that this aberrant behavior is the very definition of a software bug lol. I then narrowed down the possible causes to AIDA 64. Specifically the Memory Read and Copy Benchmarks. It is repeatable and never happens I I don't run those two tests specifically.

So I'd really love a fix for this, but I am cool with waiting for it if you want to fold it into your next update. Now that I know my CPU isn't actually losing clocks for no good reason its not a big deal. Just thought I should inform you and maybe provide an answer to any other users searching the interwebs in near panic for reassurance about their precious overclock... not that I was ever like that, you understand but other people.

 

[Martin]

I don't see a reason why HWiNFO would lie about the clock here. What you see is what should be really happening. Such down-clocking might be due to throttling if some of the limits is reached. Check other things like temperatures, power limits and performance limiting reasons.

 

[JayDeezy]

Happy Turkey Day Martin! I appreciate your reply. Much as I hate it when my machine gets buggy, I love gettin into the weeds, to look for fixes lol! I question the lowered clock reading because I have all of the pertinent BIOS and Windows clock and power behavior set to disallow clock frequency reduction other than an AVX Offset of -100MHz. For the last year or so this has achieved the desired CPU Clock behavior and I never once saw a Minimum Clock Frequency reading of less than the predicted AVX Offset result while running in this configuration. Then last week I'm performing the Benchmarks and system config checks I do periodically to make sure that all of my hardware is happy and Windows is behaving, and I see that for the first time since I established my maximum stable overclock of 5.0GHZ all core suddenly my CPU Clock Frequency is mysteriously dropping 800 MHz without obvious cause!? And every single benchmark test that I use to gauge the wellness of my rig produces the same scores as they always have for this overclock. Obviously there is a small amount of variation, but Cinebench R15 and R20 are showing results within 50 to 100 points. In AIDA64 I have 15 or more results for the Memory Read and Copy benchmarks that represent the only actions yet determined to result in dropped frequency saved from over the course of the last year. When my 9900k is running at 5GHz, it scores between 52000 and 53134 in the Read test and the results table displays the Clock speed as 5000MHz. Same goes for the Mem Copy bench with the results ranging between 50000 and 51500. So you can see how I figured it was most likely a software bug. If you are confident that HWINFO64 is accurate about the frequency drop then how can the CPU still achieve the same bench results? HWINFO also shows the same Vcore, amp draw and wattage behavior that it always has for my 9900k. So if the clocks are suddenly 800 MHz lower at heavy workload, the power draw footprint should look different right? How am I still performing tasks at the same level of performance if the 9900k has been kneecapped somehow? I mean, I really hope its just a software bug cuz I do not relish the idea of replacing a 9900k that is degrading. This aint the year for expensive tech refurbishment, you know what I mean lol!

 

[JayDeezy]

Just saw your Post about EFFECTIVE Clock sampling. Very interested in the new approach, especially if it covers my little mystery. Wish I had managed to recognise it as a possible resolution to my Q lol. Anyway, wanted to make clear that I wasn't looking to troll.

 

[Martin]

Well, you're right - if the clock would be dropping then you should see different performance data. What does the Effective clock show then during the benchmark? This will be certainly lower at idle as the averaging interval takes idle states into account.

 

[JayDeezy]

The minimum clock reading that shows drops by either 700 or 800 MHz, depending on which of the two effected tests is running, stays at 4900 MHz until just before the benchmark finishes and the result comes up. IOW it shows that the AVX Offset has been tripped at least once since Windows started up. The benchmarks take 3 seconds or so and the clock reading always drops at about the 2 second mark. I have run the tests around 50 times since I identified the tests as the cause and the clock change and it always follows the same pattern.

 

[JayDeezy]

I am resurrecting this thread only because I just confirmed that whatever was causing my original issue with frequency drops recorded during Memory Read and Copy tests but not Write or Latency or any other test has finally been corrected. At the time, Martin was kind enough to help me understand some of the intricacies of HWINFO64 and how it interacts with the MOBO and CPU. Can't be certain about the cause of this resolution, any more than I can be certain about the cause of the original problem. However I did just update the BIOS on my Asus Maximus XI Code for the first time since the clocks started showing as lower last year. So I suppose this may tend to indicate that HWINFO64 was happily doing it's job the whole time and that my CPU frequency really was dropping for some inexplicable reason during 2 particular Memory benchmark tests and at no other time. This despite the fact that no other monitoring software showed any such drop?! Yet one more strong reason to recommend HWINFO64! Just figured I should tie a little bow on this one!

 

[Martin]

I think that some of the memory bechmarks (not sure if all) in AIDA64 run on one core only. That would explain why other cores frequency drops down as they are not utilized.