Discussion Core I9 9900K on asus ROG Strix Z390-E motherboard over clocking

germanium

Distinguished
Jul 8, 2006
93
9
18,565
This was an exercise in near futility until I started ignoring the advice to use a fixed voltage. With fixed user defined voltage I couldn't get the voltage low enough & stable enough that I could leave it that way on a continual basis.

I'm running my 9900k at 5.0 GHz with an AVX offset of -2. Currently I can run Cinebench 20 at 1.225 volts at 4.8GHz. Going to 4.9 GHz causes the voltage to shoot up to 1.386 volts & 1.412 volts at 5.0GHz. Unfortunately at anything above 4.8GHz the voltage offset for CPU/VID does not work & will not lower these voltages. At 4.8 GHZ I could lower the voltage to the point that it became unstable & would not run the benchmark long enough to get a reading. I did however run Cinebench 15 with an offset of .030 which yielded a Vcore of 1.278 volts successfully several times. This is running with the CPU/VID algorithm set to typical scenario setting as this produced the lowest stable voltage setting then I worked down from there to arrive at the best off set of -.030 volts. with these setting I was able to get the AVX mode heat down below 85 degrees Celsius. The integer math exemplified by the Cinebench 15 benchmark yielded 90 degrees Celsius with voltage at 1.278 volts.

Note that the CPU/VID offset setting does not always correlate with the actual voltages achieved. Actual voltages turned out to be even slightly more reduced than the setting suggested.

An interesting occurrence is that on manual fixed voltage I was unable to run Cinebench 15 consistently until I reached 1.36 volts where using typical scenario & .030 volt offset I was able to run Cinebench 15 consistently at 1.278 volts. Quite a difference. I am convinced that these are real numbers for voltage as the heat dropped dramatically from slamming its head against the 100c mark to 80c or less at least on the AVX code Cinebench 20, 90C on the Cinebench 15 when run at 5.0GHz.
 
Last edited:

germanium

Distinguished
Jul 8, 2006
93
9
18,565
Yes I have. Offset voltage mode works better at getting voltages as low as possible. Offset voltage is a subset of adaptive voltage that lets you set voltage parameters for the adaptive voltage which is based off of CPU/VID.

Changed my mind & am using adaptive voltage. For details see below the next post.
 
Last edited:

germanium

Distinguished
Jul 8, 2006
93
9
18,565
After giving it more time there was some instability when idle as the idle voltage was quite low as in .630 volts. Cinebench 15 also ultimately decided it did like the reduced voltage so I raised it up by reducing the offset from .030 to .020 & that seems to have done the trick. Runs a little hotter now but not 100degrees centigrade. About 80-90c now. 80 max on AVX code & 90 on advanced integer such as Cinebench 15. This is still way better than when I had fixed voltages as on some things it was hitting 100c & throttling which it isn't now.
 

germanium

Distinguished
Jul 8, 2006
93
9
18,565
Boy is it hard to get these thing perfectly configured. If I got the load voltage to where I could run Cinebench successfully without going over 90c temperature the idle voltage was too low & it would eventually crash when doing hardly anything using the offset voltage mode. If you lock the cache frequency to 4.7 GHz voltage would be too high & attempts to lower it made the computer more unstable at idle as idle voltage was too low.

I'm currently running variable cache frequency 1GHz-4.6GHz. I set the CPU/VID behavior to best case scenario. I also set the load line calibration to 5 in order to get the correct voltage droop under load so when computer gets initiated to run program but not yet loaded voltage is 1.350 then drops to 1.305 under load. this I did with adaptive voltage with .250 compensation for turbo & +.020 compensation to get to the 1.350 turbo no load voltage as well as prevent the voltage at idle from going too low & causing crashes when idle. This combined with loadline calibration of 5 prevented crashing on Cinebench 15 as well as prevented idle crashes & reduced the power needed to run whether idle or under load compared to running a fixed voltage without crashing.

When checking voltages looking at HW monitor do not use the voltages from the CPU/VID for each core, these voltages are just a request by the CPU for the external regulators to provide, motherboard BIOS actually determines what the processor receives & HW Monitor provides that closer to the top of the window.

I have to say Cinebench 15 is a very tough load to run successfully at maximum performance without crashing, much tougher for me to get best performance than Cinebench 20, Of coarse I do run the 20 version at 4.8GHz due to being an AVX2 benchmark.

Of special note the voltage you see in the BIOS is not the voltage you see in HW Monitor program inside windows so you have to boot to windows to observe what your voltages are under load/no load situations at max set frequency in order to make the proper voltage adjustment in the bios when using adaptive voltage. You also want to observe the idle voltage as well as stock the voltage is ever so slightly too low. Mine was .666 volts. I boosted it to .684 with the adaptive voltage offset compensation. This setting affects both idle voltage & peak turbo no load/load voltage.

It is definitely a challenge to get the perfect balance but this has definitely ended all my crashes & allows my computer to run cooler. Peak temp after single run of Cinebench 15 is 86c. Was 100c when was running fixed voltage. This was in part due to higher idle temps when running fixed voltage as well as higher power dissipation due to the high voltage needed to run the benchmark.

I went back & removed the AVX limitation & successfully run Cinebench 20 @ 5.0GHz without issue with peak temperature of 90c after single run. Idle temperature bouncing between 31-32c.
 
Last edited:

germanium

Distinguished
Jul 8, 2006
93
9
18,565
After all these changes & getting CPU stable I decided to go back & redo my memory overclock. Before all I could get before was 2.8Ghz with 16,18,18,36,2t timings. Now since I have the CPU stable I was able to get the memory up to 3.0GHz with 15,17,17,32,2t timings. Tested this as good with Memtest86 latest version.

Went back & was further able to boost memory to 3200MHz with 17,19,19,34,2t timings still at 1.35 volts. This is the highest I have been able to achieve so far which is 400MHz better than earlier before going to variable voltage to reduce power consumption. Though the timing are a little looser the overall performance is still better. Passed all tests with Memtest86 latest version which tests even for errors that rowhammer vulnerability could use to compromise PC.

Correction, I had to scale back the memory speed to 3.0GHz as i discovered that the BIOS had automatically boosted the voltage going to the RAM to 1.510 volts. I locked it to 1.350 volt & retried booting at 3.2GHz & it failed several attempts even with much looser timings but the 3.0GHz is good with reasonably tight timings stand.
 
Last edited: