[SOLVED] 9900k first 5.0ghz oc attempt (vCore question and low timespy cpu score)

[heaterpeter90]

Good morning guys, I've attempted to manually oc my 9900k to 5.0ghz following one of the multiple guides you can find online. I'm using a z390 aorus pro as a mobo and a noctua nh-d15s for the cpu cooling. After several failed attempts even at high voltages, I found out stability with vcore at 1.35, vccio set at 1.20 and vcssa at 1.25 (probably the xmp profile was causing instability and not the vcore).

Now during timespy even with a fixed voltage I can see the vcore in the osd raising from 1.35 to 1.38. How is possible that the vcore would rise like in auto mode? My timespy cpu score also looks low compared to the average (got 11460). Temps during cpu test are around 85 degrees ( a bit too high with that cooler right?). That's all for now, hope i was clear and can get an advice on what I'm getting wrong here. Thank you

 

[Phaaze88]

I'm not going to be of too much help here, but...

Now during timespy even with a fixed voltage I can see the vcore in the osd raising from 1.35 to 1.38. How is possible that the vcore would rise like in auto mode?

That's most likely the result of your LLC setting. Static Vcore applications aren't immune to Vdroop.
Don't want such a high overshoot? Then lower the LLC.

As for the rest, the overclock probably isn't stable anyway - crashing isn't the only symptom of instability; poor benchmark scores are too.

Coffee Lake Refresh	All Core SSE Frequency	All Core AVX2 Frequency	BIOS Vcore	% Capable
9900K	4.80GHz	4.60GHz	1.275V	100%
9900K	4.90GHz	4.70GHz	1.287V	Top 91%
9900K	5.00GHz	4.80GHz	1.300V	Top 30%
9900K	5.10GHz	4.90GHz	1.312V	Top 5%

Source: https://siliconlottery.com/pages/statistics
That's Silicon Lottery's(company) binning statistics for the 9900K. Going by the above: If you have to crank that much more Vcore into your chip, it's likely not a 5.0ghz capable chip.

"To achieve the highest overclock, keep in mind that for your final 100 MHz increase, a corresponding increase in Core voltage of about 50 millivolts (0.050) is needed to maintain stability. If 70 millivolts (0.070) or more is needed for the next stable 100 MHz increase, it means you're attempting to overclock your processor beyond its capability. All processors reach a limit where an additional increase in Core voltage will not stabilize another 100 MHz increase in Frequency."
Source: CompuTronix's Intel Temperature Guide, if you're in doubt. I suggest giving it read.

 

[heaterpeter90]

I'm not going to be of too much help here, but...


That's most likely the result of your LLC setting. Static Vcore applications aren't immune to Vdroop.
Don't want such a high overshoot? Then lower the LLC.

As for the rest, the overclock probably isn't stable anyway - crashing isn't the only symptom of instability; poor benchmark scores are too.

Coffee Lake Refresh	All Core SSE Frequency	All Core AVX2 Frequency	BIOS Vcore	% Capable
9900K	4.80GHz	4.60GHz	1.275V	100%
9900K	4.90GHz	4.70GHz	1.287V	Top 91%
9900K	5.00GHz	4.80GHz	1.300V	Top 30%
9900K	5.10GHz	4.90GHz	1.312V	Top 5%

Source: https://siliconlottery.com/pages/statistics
That's Silicon Lottery's(company) binning statistics for the 9900K. Going by the above: If you have to crank that much more Vcore into your chip, it's likely not a 5.0ghz capable chip.

"To achieve the highest overclock, keep in mind that for your final 100 MHz increase, a corresponding increase in Core voltage of about 50 millivolts (0.050) is needed to maintain stability. If 70 millivolts (0.070) or more is needed for the next stable 100 MHz increase, it means you're attempting to overclock your processor beyond its capability. All processors reach a limit where an additional increase in Core voltage will not stabilize another 100 MHz increase in Frequency."
Source: CompuTronix's Intel Temperature Guide, if you're in doubt. I suggest giving it read.

Thanks I appreciate. I thought I wasn't lucky with my chip of course, but I didn't know would affect performance as well. Do you think try go from 5.0 to 4.9 would be a better idea?

 

[Phaaze88]

At the very least, start from 4.8ghz and try to work your way up from there, because if you're trying to do this without an AVX offset unlike in the chart - at the same frequencies, AVX requires more Vcore.

If you do use the offset, then start from 4.9ghz.
Test Vcore stability with at least an hour of Realbench and Cinebench R23, until you find the lowest possible Vcore - the longer they're run, the better.
Once you've found that, test cpu cooler thermal stability via Prime95, Small FFT, all 3 AVX options off - 15 mins is enough for this.

After you've passed both voltage and thermal tests, raise the Vcore 0.05v for the next 100mhz and repeat the voltage and thermal tests.
You basically repeat that until you hit one of the following walls:
-the next 100mhz needs more than an additional 0.05v to stabilize, or simply refuses outright.
-thermal tests exceed 85C. While Prime 95 produces a hard, synthetic workload, it's pretty close in thermal load to a more realistic one that uses AVX; Realbench uses AVX, but it is at specific intervals and not constant.

 

[heaterpeter90]

At the very least, start from 4.8ghz and try to work your way up from there, because if you're trying to do this without an AVX offset unlike in the chart - at the same frequencies, AVX requires more Vcore.

If you do use the offset, then start from 4.9ghz.
Test Vcore stability with at least an hour of Realbench and Cinebench R23, until you find the lowest possible Vcore - the longer they're run, the better.
Once you've found that, test cpu cooler thermal stability via Prime95, Small FFT, all 3 AVX options off - 15 mins is enough for this.

After you've passed both voltage and thermal tests, raise the Vcore 0.05v for the next 100mhz and repeat the voltage and thermal tests.
You basically repeat that until you hit one of the following walls:
-the next 100mhz needs more than an additional 0.05v to stabilize, or simply refuses outright.
-thermal tests exceed 85C. While Prime 95 produces a hard, synthetic workload, it's pretty close in thermal load to a more realistic one that uses AVX; Realbench uses AVX, but it is at specific intervals and not constant.

Shall I use an avx offset? I've tried now a cinebench r23 stress test trying to stay at 5.0ghz with 1.32v and it crashed after 4 minutes. If is not enough vcore, why I never get a bsod? It's possible that is something else like memories stability? (I have set a vccio of 1.21)

 

[Phaaze88]

Shall I use an avx offset?

This one's debatable.
I say use it until you find the cpu's max stable frequency. Then, if there's still cooling headroom to do so, work your way down to 0 offset.

I've tried now a cinebench r23 stress test trying to stay at 5.0ghz with 1.32v and it crashed after 4 minutes.

And you worked your way up to this? So you must've gotten 4.9 stable at 1.27v and 4.8 stable at 1.22v or something?

As I said earlier, some samples aren't capable of 5.0ghz.

If is not enough vcore, why I never get a bsod?

BSOD doesn't automatically translate to insufficient Vcore.
Some apps have automatic error checking and will stop the program if they are detected.

It's possible that is something else like memories stability? (I have set a vccio of 1.21)

Now it sounds like you are doing too much at once. One thing at a time. You can't rush this, or it'll just complicate troubleshooting.

 

[heaterpeter90]

This one's debatable.
I say use it until you find the cpu's max stable frequency. Then, if there's still cooling headroom to do so, work your way down to 0 offset.


And you worked your way up to this? So you must've gotten 4.9 stable at 1.27v and 4.8 stable at 1.22v or something?

As I said earlier, some samples aren't capable of 5.0ghz.


BSOD doesn't automatically translate to insufficient Vcore.
Some apps have automatic error checking and will stop the program if they are detected.


Now it sounds like you are doing too much at once. One thing at a time. You can't rush this, or it'll just complicate troubleshooting.

Thank you so much for all the replies. So now without rushing as you said, I lowered the clock to 4.9ghz with 1.30 . I know is an high voltage for that frequency but at least the r23 stress test didn't crash. I'm worried about the thermals though, I'm getting around 80 degress and with the noctua I don't think it's good (it means that with prime95 I would get a thermal crash for sure). I will try to settle 1.28 for 4.9ghz next time i will reboot and see how is going.

edit: I'm stable now with 4.9ghz at 1.27v. It doesn't like that bad right?