9600k hwinfo64 effective clock readings inconsistent between OC's

[SGTChalk]

New here so I hope this is in the right place, and hasn't been answered already (all the threads I've found didn't give me a satisfying explanation). I've got two overclocks, one at 4.6 and one at 4.8, the 4.6 fully loaded in prime95 seems to read a stable 4.6ghz effective clock, but the 4.8 roams around between 4.5 and 4.8 usually averaging around mid 4.7's effective clock. I'm not getting errors, and none of the power/current/temp etc limiters are tripping at least as far as what hwinfo is telling me, so why would it be different between the OCs? It seemed to get better with more voltage (as I weeded out the errors as well, which are gone now), but I'm getting close to my temperature ceiling at 1.33v core and llc5--full avx load (no offset) maxing at 88c, set to throttle at 90c but not reading a throttle happening. It will at first stay at 4.8ghz effective, but then it drops down after awhile, and I can't tell if it's just because it's moving past the test where it's able to max out and the following tests don't utilize as effectively (seeing 98-99% c0 residency after it drops), or if it's hitting some limiter that it's not showing me. I know it counts the sleep states etc in effective clock and that's the point, which is mostly what I've seen in the other threads, but then why would it be maxed out at first and then drop while still fully loaded, and why would the other OC not do that, nothing should be sleeping right?

Edit: forgot to mention it does it with AVX off too, where it should have tons of wattage and temp headroom

[unclewebb]

@SGTChalk For a second opinion, exit HWiNFO and run ThrottleStop instead.

 

https://www.techpowerup.com/download/techpowerup-throttlestop/

 

It follows the Intel recommended monitoring method and does a good job tracking any tiny change in the CPU multiplier for each thread.

 

 

 

When testing at 4.6 or 4.8, the multiplier should be reporting exactly 46.00 or 48.00. Open the Limit Reasons window and watch for any boxes lighting up red which indicates throttling. You need to exit HWiNFO when using Limit Reasons so HWiNFO does not interfere with the results. 

 

Post some pics while your CPU is loaded so I can have a look. Also post a pic of the TPL window so I can see what your power limits are set to and a pic of the FIVR window. If V-Max Stress or Thermal Velocity Boost are checked, your CPU will start to slow down if the voltage is too high or if the CPU temp goes over 70°C.

 

 

Edit - Your voltage seems a little on the high side for 4.8 GHz.

This is only 10 threads of Prime95 Small FFTs with AVX enabled. It has not been thoroughly tested. 

Too much voltage will increase the temps significantly. 

 

 

[SGTChalk]

@unclewebb Here are the requested screenshots run on the 48x OC, both the mentioned boxes are checked, so it would seem then that it is throttling, just wasn't saying so in hwinfo, and that I probably don't have the thermal headroom for the albeit unrealistic torture test if they weren't. Anything below this voltage was giving me errors in prime95, which could probably tighten up with higher LLC and lower voltage, but after reading about it and the purpose of vdroop going higher LLC sketched me out which is how I ended up here. From what I've seen the higher end/core count processors reach higher ratio with lower voltage...bbuutttt I also saw people saying they were running 5.0ghz at 1.3v on this processor, I'm going to guess with an avx offset, lower ring, and tighter LLC. MSI was saying 1.31v for 4.6ghz on this processor (vs 1.21v for a 9900k), but I know they're going to trend higher than necessary, I was running stable at 1.27v at 46x LLC6. I guess the step from here is try with those boxes unchecked for 4.7 maybe? Since outside of prime95 it would never hit max temps like that. It's set up to throttle at 90c if that goes awry, pretty sure, I think..hope. All in all I think you answered my original question and thank you, it's reading lower and differently because it's secretly throttling.

 

AVX on:

AVX off:

[unclewebb]

1 hour ago, SGTChalk said:

it would seem then that it is throttling

The only throttling I see is thermal throttling because you set the PROCHOT temperature to 90°C in the BIOS. When ThrottleStop shows PROCHOT 90°C in red, that means the CPU thermal throttled. It might have only throttled for a few milliseconds but that was enough to trip this sensor. In your AVX On screenshot, you are running right on the edge of thermal throttling. If ThrottleStop is showing 48.00 for the CPU multiplier and Limit Reasons is all black, there is no throttling at the moment you took that screenshot.

 

In the Options window, if PROCHOT Offset is not locked, you can reduce PROCHOT Offset from 10 to 5 so your CPU starts thermal throttling at 95°C instead of 90°C. This will give you a little more temperature headroom while testing. Intel set the thermal throttling temperature to 100°C because they know their CPUs can run reliably up to this temperature. 

 

If your CPU needs lots of voltage to be stable then it is what it is. Not much you can do about that. My 10th Gen runs great at low voltage below 5.0 GHz but needs crazy voltage beyond that. 

 

The Thermal Velocity Boost and V-Max Stress boxes in ThrottleStop are not indicators. They just allow you to disable this type of throttling. I do not think the 9600K uses thermal velocity boost so whether this box is checked or not probably makes no difference. The 10th Gen can throttle 100 MHz at high voltage if V-Max is enabled. Either TVB or V MAX will show up near the bottom of the Limit Reasons window if either of these types of throttling are in progress. Everything looks OK. 

 

As long as your multiplier maintains a steady 48.00 in ThrottleStop, I would not worry about the effective MHz. Temporarily lower your turbo power limits and run a stress test and you should see things light up in Limit Reasons. Red boxes indicate throttling is in progress. Yellow boxes are a record of any previous CPU throttling. 

 

[SGTChalk]

Thanks for the help and information! I guess this OC will work alright then, I didn't see limit reasons light anything up, and the multiplier only varied at idle, at load it was rock steady in Throttlestop. I'll do what you said about the PROCHOT for some final stress testing, it shouldn't get anywhere near that level in normal usage anyway. 

[SGTChalk]

Kind of a weird update here that I can't explain really. I booted the computer up, confirmed it was still doing the weird supposed throttling that wasn't registering anywhere except effective clock in hwinfo, shut down, saved those settings, loaded up the 4.6 OC that had never done that, confirmed it still didn't do that, shut down, loaded the exact settings (and visually confirmed it was indeed the exact same settings before sending it) that were just minutes earlier doing that, and now it's acting exactly how it should, rock steady 4.8ghz effective, benching way higher than it had before on either OC in RealBench etc. So there was a performance hit from what hwinfo was reading, I'm just not sure what was happening, why, or what made it stop. I'm completely out of my depth on why saving and then loading the settings would change anything, when the settings are the same. I did however notice that on both OC's, when opening Throttlestop on a fresh boot, I've got a yellow PL1 and EDP Other in limit reasons, they stay yellow throughout stress testing and if I clear them they don't come back. 

[unclewebb]

@SGTChalk It is common for the CPU to trigger throttling flags when first starting up or when resuming from sleep or hibernate. These flags seem to get triggered before the BIOS has a chance to set the CPU power limits. That is why it is a good idea to Clear this throttling data out of the CPU before you do any testing. 

 

Be careful when using ThrottleStop. If you make any changes in the BIOS to the CPU multiplier, if your start ThrottleStop and it finds a ThrottleStop.INI configuration file, it will set the CPU to whatever settings were previously saved in that file. If you make any changes in the BIOS, it would be best if you delete the ThrottleStop.INI configuration file each time so any previously saved settings do not override what the BIOS set your CPU to. 

 

I like testing with Cinebench R20. I find the results are fairly consistent from run to run. 

https://www.techpowerup.com/download/maxon-cinebench/

 

If you ever find a way to duplicate the effective clock throttling, run R20 with this throttling problem and run it again without this problem to see if there is a difference in performance. You got me curious now. I will try doing some testing with HWiNFO to see if I can figure anything out. 

 

Edit - In Cinebench I am getting a consistent 5000 MHz Effective Clock across all threads. 

Same in Prime95.

 

[SGTChalk]

Well, I was able to replicate it, and it wasn't what I thought it was (I should know better than to jump the gun when troubleshooting, I work in the autonomous industry troubleshooting robots, oof) but I'm hoping you might have some insight on what's actually going on here. 

I started the computer in the exact configuration it was reaching full 4.8ghz effective last night, double checked that the Throttlestop.ini was still deleted, let it warm up while I sipped my coffee, started hwinfo and ran Cinebench R20, and it was doing the throttling again with no changes or even entering bios:

So then I closed hwinfo and started up Throttlestop (with no .ini) and ran again and it was scoring higher:

 

Then I closed limit reasons, left Throttlestop up and opened hwinfo just to confirm what I suspected, that it wasn't throttling now after having started Throttlestop:

 

Closed Throttlestop, deleted Throttlestop.ini, ran again and still not throttling, rebooted and again didn't touch Bios. Started hwinfo, ran Cinebench R20, throttling again. Ran Small fft's in Prime95 avx on, still throttling but as you can see hasn't hit temps where it should be throttling yet (no indicated throttles in hwinfo either, except effective clock): 

 

So this time I kept hwinfo running, stopped the test, opened throttlestop (with no Throttlestop.ini) and started the same test, small fft's avx on: 

 

So, it wasn't bios, I jumped the gun on that probably because of a preexisiting bias about this bios and being out of my depth here. It seems it has something to do with Throttlestop, even though I'm not touching any settings in there, just turning it on, without the Throttlestop.ini. It however also seems that Throttlestop was telling the truth, and so was Hwinfo, and unfortunately Throttlestop isn't catching whatever's causing the throttle (neither is hwinfo) before it's been run, because starting it prevents whatever throttle that might be. Any idea what it might be changing just by turning on?

Edit: thanks for the Cinebench R20 link, it is indeed far more consistent than Realbench!

[unclewebb]

1 hour ago, SGTChalk said:

it wasn't throttling now after having started ThrottleStop:

When ThrottleStop first starts up, it checks various CPU registers for any inconsistencies. The BIOS is supposed to set each core to the same values but sometimes there is a bug in the BIOS and it does not do this correctly. If ThrottleStop notices any issues, it quietly fixes the problem without saying anything. It does this automatic fixing whether it finds an INI configuration file or not. 

 

When testing with only HWiNFO, did you scroll down to see if any of the throttling flags are being triggered? There is a wide variety of reasons for throttling and Intel CPUs usually do a good job of reporting something if it detects any throttling. 

 

 

If you cannot see anything, try running CPU-Z, go to the About tab and press the Save Report (.TXT) button. Run a report after booting up but before you run ThrottleStop. Run a second report after running ThrottleStop. If you attach these two reports to your next post, maybe I will be able to see something that is changing. 

 

Time to do some more testing.

Edit - I get full Effective Clock without having to run ThrottleStop.

 

Just for fun, here is an interesting comparison. I disabled hyper threading and set my 10 core CPU to 6 cores and ran Cinebench R20 at 4800 MHz. My fake 9600K is a little faster than your 9600K but not much. I ran this test with the cache at 4500 MHz and the memory was at DDR-3600 CL16-16-16. Not much IPC improvement between 9th and 10th Gen. 

 

[SGTChalk]

Still not seeing any throttles in hwinfo before running Throttlestop, screenshots taken during a Cinebench R20 run where it was throttling before having run Throttlestop. I also ran prime95 after to check again, still nothing: 

 

CPU-Z log before having run Throttlestop, during a Cinebench R20 run while throttling:

CHALK1.txt

 

CPU-Z log after running Throttlestop, during a Cinebench R20 run while not throttling:

CHALK2.txt

 

I'll also be looking through those after sending this to see if I can find any differences, though from a cursory glance I can tell a lot of it is just going to be at the "thing look same or not same" level since it looks pretty complicated to me, haha.

That's pretty interesting about the fake 9600k. I would have expected more of a bump from your being stable at lower voltage (and therefore lower temps it looks like). Your ram is running at a higher clock, but mines running tighter timings (3200mhz, CL14-14-14). I think you might be getting a bump from the higher cache at 4500mhz, I'm on the auto setting which is indicated as 43x, but I'm pretty confident it's doing something else that isn't indicated or explained since if I manually set 43x ring the core voltage stays high instead of dropping down to the 0.8v range at idle--and uses a ton more wattage. Considering I spend all day on the computer, at $0.27/kwh I'm definitely trying to strike a balance between not compromising performance meaningfully but also not emptying the bank, haha.

Which actually leads to a completely unrelated question I've been having, hwinfo is showing C7 state residency a lot, but in the bios I have it set to stop at C3 for the faster wake times, I was/am noticing a stutter especially when scrolling after a pause in low performance usage when I had the full C state compliment active. This actually all kind of started because of stuttering issues in games, which lead to OCing in the first place because while looking into it I realized the auto settings in bios were driving the processor up to a whopping 1.38v at 46x (which unfortunately it had been doing unbeknownst to me for the past year 24/7 while on Folding@home, and which sadly I can't afford to participate in anymore since moving). It seems the consensus is 1.35v safe for this processor, and maybe up to 1.4v if your temps are real low, mine were a lot better at the time before this OC, thankfully. That may have some relation to why it seems to need a lot of voltage for 48x. 

Edit: this is the only difference that I understood that seemed meaningful, outside of this there were a lot of differences in the letter/number strings and plots that I don't know how to interpret: 

Not sure why it would be showing a second ring "core" while throttling. It's still idling voltage down and using less watts at idle after running Throttlestop so whatever Throttlestop is fixing isn't interfering with that auto ring setting in bios in any of the ways I've noticed it making a difference.

[SGTChalk]

Update: Ran into an error in Prime95 for the first time after running more tests at about the 45 min mark. No more room to up the voltage so I dropped it from 48x to 47x and from 1.33v to 1.3v, no errors yet but it's having the same problem. Ran 4.7 effective for maybe 10-15 mins on the fresh settings, then started throttling without lighting up any throttle indications in hwinfo. Run Throttlestop and now it's 4.7 steady, it's got plenty of thermal headroom trying to run this easier OC too. I'd flash the bios to see if that helped but I don't have spare hardware if that goes wrong, and I can't do without a computer right now while waiting for parts. So without knowing what Throttlestop is fixing I don't know what I can really do about it other than grin and bear it. 

[unclewebb]

@SGTChalk I was not able to figure out what kind of magic that ThrottleStop is performing. Just happy to see that it works. You can follow this guide if you need to add it to your Windows start up sequence. 

 

http://forum.notebookreview.com/threads/the-throttlestop-guide.531329/#post-6865107

[SGTChalk]

That's probably the best solution for the time being, and thanks again for the effort you put in to try to get this sorted out!

[Imannudein]

On 3/28/2021 at 6:04 AM, SGTChalk said:

That's probably the best solution for the time being, and thanks again for the effort you put in to try to get this sorted out!

Dear @SGTChalk,

 

Did you get to the bottom of this? I am also experiencing the same problem on my 10850k with Maximus 12  hero wifi. Effective clocks are all over the place. Run TS and that thing bench with effective clock same as discreet clocks. 

[unclewebb]

38 minutes ago, Imannudein said:

Effective clocks are all over the place.

When a CPU is not fully loaded, this is normal. When all threads are fully loaded, the Effective Clock should be steady for each thread. 

 

Boot up and run Cinebench R20. What are your Effective Clock speeds while Cinebench is running?

https://www.techpowerup.com/download/maxon-cinebench/

[Imannudein]

1 minute ago, unclewebb said:

When a CPU is not fully loaded, this is normal. When all threads are fully loaded, the Effective Clock should be steady for each thread. 

 

Boot up and run Cinebench R20. What are your Effective Clock speeds while Cinebench is running?

https://www.techpowerup.com/download/maxon-cinebench/

Hi @unclewebb,

 

This is perhaps the continuation of my thread earlier. I am at the office currently but I remember this; when CBr23 running, average Effective Clock as shown by HWInfo64 is 4645Mhz with one of the cores maxing out at only 4450mhz. If I open ThrottleStop, without touching anything, Effective Clock will be stable across all cores and the same value as discreet clocks at 4800Mhz. I'll post screenshot once I'm home later today. 

[unclewebb]

11 minutes ago, Imannudein said:

CBr23

I am assuming that you get better Cinebench results when the Effective Clock is at the full 4800 MHz? Definitely a weird problem. Almost like the CPU is throttling a little internally. Not yet sure why this is happening or how many other Intel computers this is happening too.

[Imannudein]

Just now, unclewebb said:

I am assuming that you get better Cinebench results when the Effective Clock is at the full 4800 MHz? Definitely a weird problem. Almost like the CPU is throttling a little internally. Not yet sure why this is happening or how many other Intel computers this is happening too.

Yes @unclewebb. On full Effective Clock, I get 16135 score whereas on the "weird" effective clock situation, 15200. VRM temp is okay at 53C. Same as @SGTChalk, no observable throttle limit being flagged. 

[Imannudein]

@unclewebb as promised, here are the screenshots:

 

1. Without ThrottleStop running. Effective clock averaged at 4600mhz

 

2. With ThrottleStop running, 4799mhz. I just open ThrottleStop and touched nothing. Not even the options. 

 

 

Needless to say I have ThrottleStop auto-run via Task Scheduler now. 

[SGTChalk]

Sorry to be so late getting back to you, I must have missed the notification email. I thought it was an issue with my admittedly outdated bios, but being that you're on a different generation and different manufacturer mobo that seems less likely. This looks exactly how mine did, especially with one core (core 0 for you as well it looks like?) in particular being noticeably lower effective clock than the rest of the already underperforming cores. Other than being able to second though I don't really have anything to add to the discussion, but I'd be happy to answer any questions. Also going to see if my friend with a 10700k is willing to check for similar behavior since it seems like a pattern is developing

Edit: it also sounded like you were talking about this issue in a different thread? Would be interested in reading through that, looked on your profile but I didn't see it

[Imannudein]

@SGTChalk thanks for your feedback. You may find the thread here: 

I really thought the problem was solved with the static manual OC. I'm now back on Auto Core Multiplier with MCE on but with 0.7mV negative offset. All good  with ThrottleStop running. It just has to be there....

[SGTChalk]

Checked in with my friend that has the 10700k, he's not having the same issue as we are. Might be worth noting however that he did his OC with intel extreme tuning utility on the software side, and not through bios like I did.

[Mister Woof]

I noticed some strange behavior, where my Effective Clock, even when reset during a full load stress test, was not saturating all clocks to 5100mhz. Instead they bounced around between 4900-5050mhz.

 

Turns out it was the setting "Energy Efficient Turbo", which was on Auto. When I changed it to "Disabled", which the BIOS recommends turning off when overclocking, and it is now 5100mhz on Effective Clock when reset during a full load stress test.

 

[Mister Woof]

Update: I think this is wrong. It started doing it again for some unknown reason after about an hour in windows when trying the test again. I started Throttlestop (didn't do anything, just opened the program) and then magically it is 5100mhz again for effective clock.

 

Something is weird about this. Feels like just having throttlestop on prevents this strange behavior that is intermittent. 

 

Or maybe the effective clock sensor in hwinfo64 is just not good.

 

@unclewebbI just made task to autostart Throttlestop with BD PROCHOT disabled. I think that there's some wizardry with your program here to just make things work the right way.

 

Thanks for your service.

[SGTChalk]

I went and looked for that setting, couldn't find it for my mobo, but I'm also not using turbo for this OC. Worth noting that once Throttlestop has been run, at least for me, even if it's closed the issue remains resolved until the next reboot, so I've just been having it run at startup and then closing it to save the admittedly itsy bitsy amount of resources it uses.

I've seen a similar sentiment about the effective clocks in hwinfo64 elsewhere, but from the testing I've done it's been extremely consistent with it's pre and post Throttlestop run readings, and those readings have been very clearly tied to huge differences in bench scores in Cinebench R20. Pre running Throttlestop my 9600k running 4.7ghz all core scores anywhere from 2400-2600s in Cinebench R20 with with hwinfo64 effective clocks reading roughly 4450-4550mhz core 0, and 4550-4650ish on the remaining cores, but after I run Throttlestop hwinfo64 reads a steady 4700mhz effective all cores, and it scores consistently in the 2800's in Cinebench R20.