Overclock.net banner
1 - 20 of 514 Posts

·
Registered
Joined
·
221 Posts
Discussion Starter · #1 ·
Howdy there,

How are people's 9900KS OC results? All the reviews on the web seem pretty uninvested and don't really show how they stresstested their KS.

I'd love to hear about people's OC results, especially when there is even a previous 9900k to compare with.
 

·
Registered
Joined
·
71 Posts
hey there im looking for threads like this to read on lol . im waiting for my 9900ks but im on a asus x hero z 370 wifi board if i get 5 ghz @ 1.22 v ill be happy 5.2 ghz @ 1.3v or less
 

·
Registered
Joined
·
221 Posts
Discussion Starter · #3 ·
hey there im looking for threads like this to read on lol . im waiting for my 9900ks but im on a asus x hero z 370 wifi board if i get 5 ghz @ 1.22 v ill be happy 5.2 ghz @ 1.3v or less
Yeah, that's exactly what I'm hoping for, 5 GHz at something like 1.2 to 1.25, able to run non AVX prime and proper Stresstests on air because my 9900k required 1.28 to 1.3 volt to be prime avx stable.
 

·
Registered
Joined
·
1,058 Posts
Yeah, that's exactly what I'm hoping for, 5 GHz at something like 1.2 to 1.25, able to run non AVX prime and proper Stresstests on air because my 9900k required 1.28 to 1.3 volt to be prime avx stable.

Didn't you buy a binned 9900K? From what I recall, it was a pretty good one.
 

·
Registered
Joined
·
221 Posts
Discussion Starter · #5 ·
Yeah, that's exactly what I'm hoping for, 5 GHz at something like 1.2 to 1.25, able to run non AVX prime and proper Stresstests on air because my 9900k required 1.28 to 1.3 volt to be prime avx stable.

Didn't you buy a binned 9900K? From what I recall, it was a pretty good one.
Yeah, it ran at around 1.25v stable during most tests but required 1.28 to 1.29 volts to run Aida 64 fpu stably which bugged me.

I think it was an average good one but not good. I think it runs really well at 4.9.

But I have my own standards when it comes to stability so I'm hoping to improve my odds with the 9900Ks, if not, I can still send it back...

I mean, whatever people assumed to be 5 GHz stable was to me 4.9 so I'm hoping to get a chip that is better.

Mine will arrive on Monday.
 

·
Facepalm
Joined
·
10,320 Posts
Yeah, it ran at around 1.25v stable during most tests but required 1.28 to 1.29 volts to run Aida 64 fpu stably which bugged me.

I think it was an average good one but not good. I think it runs really well at 4.9.

But I have my own standards when it comes to stability so I'm hoping to improve my odds with the 9900Ks, if not, I can still send it back...

I mean, whatever people assumed to be 5 GHz stable was to me 4.9 so I'm hoping to get a chip that is better.

Mine will arrive on Monday.
Is that 1.29v bios set voltage or *load* voltage, and how was it measured? Maximus XI core voltage sensor? VR VOUT (Gigabyte, etc)?
 

·
Registered
Joined
·
221 Posts
Discussion Starter · #7 · (Edited)

Attachments

·
Facepalm
Joined
·
10,320 Posts
Got it.
I did the exact same test as you and got the same VR OUT in AIDA64 stress test (1.184v) but I had a lower power draw. Same settings: 5 ghz 4.7 cache, but I didn't use Turbo LLC like you did.
I used LLC=High with a higher Bios target voltage than you did (1.30v vs 1.275v). Improves on transient response (VMIN) while the RMS load vcore is lower from more vdroop (lower temps). This creates a stability increase.

I no longer use LLC Turbo in any of my profiles now.

Try this:
1.30v BIOS set.
Vcore LLC=High.
Run AIDA64 stress test like that.
Your VR VOUT will be the same but your power draw or temps should be lower.
 

·
Registered
Joined
·
221 Posts
Discussion Starter · #9 · (Edited)
Got it.
I did the exact same test as you and got the same VR OUT in AIDA64 stress test (1.184v) but I had a lower power draw. Same settings: 5 ghz 4.7 cache, but I didn't use Turbo LLC like you did.
I used LLC=High with a higher Bios target voltage than you did (1.30v vs 1.275v). Improves on transient response (VMIN) while the RMS load vcore is lower from more vdroop (lower temps). This creates a stability increase.

I no longer use LLC Turbo in any of my profiles now.

Try this:
1.30v BIOS set.
Vcore LLC=High.
Run AIDA64 stress test like that.
Your VR VOUT will be the same but your power draw or temps should be lower.
Oh man, thank you so much for doing the same test so I have some reference to go with by another user's good sample :)

You only did FPU? Because only FPU is the proper stresstest, not checking all the other boxes. To me Aida 64 in this mode is the best stress test to test your CPU, not too demanding like Prime and not too absurdly low profile like der b8uer's test.

Yeah you are right, I think a lower LLC would be better.

I just hope the 9900ks will give me something where 5.1 sits at this voltage so I can run 5 GHz more stably at a lower one or do you think my expectations are too high?

I wish reviewers tested the 9900ks with aida 64 in FPU mode instead of vague stability assumptions based on quick tests without telling what they've actually used :/
 

·
Facepalm
Joined
·
10,320 Posts
Oh man, thank you so much for doing the same test so I have some reference to go with by another user's good sample :)

You only did FPU? Because only FPU is the proper stresstest, not checking all the other boxes. To me Aida 64 in this mode is the best stress test to test your CPU, not too demanding like Prime and not too absurdly low profile like der b8uer's test.

Yeah you are right, I think a lower LLC would be better.

I just hope the 9900ks will give me something where 5.1 sits at this voltage so I can run 5 GHz more stably at a lower one or do you think my expectations are too high?

I wish reviewers tested the 9900ks with aida 64 in FPU mode instead of vague stability assumptions based on quick tests without telling what they've actually used :/
My chip isn't a good sample. It's average. Middle of the road. Unable to do 5 ghz FMA3 small FFT prime95 on Auto vcore and AC Loadline 1.6 mOhms and "Standard" Vcore loadline calibration because it gets too hot and crashes at 90C. And can barely do small FFT AVX Prime95 at the same settings (Auto vcore, LLC:Standard, ACLL 1.6 mOhms).

Yes I did "check FPU" only.
I'm aware that the other stress tests draw less power.
Have you tried check FPU with 1.30v bios set and LLC:High and checking your VR VOUT and comparing it to your previous 1.275v + Turbo attempt?

By the way, the "Julia" and a couple of the other FPU benchmark tests draw as much current as FMA3 prime95 ! WAY more than "Test FPU". It's very hard to get CURRENT IOUT or DTS temps even logged in HWInfo64 when running those because the entire system freezes during the most of the test no matter how stable you are. Probably why Aida doesn't include those in the stress tests, because people would just crash or reach 100C+ trying them for too long. Don't be surprised if you BSOD on those.

A 9900KS at its worst should have a +100 mhz bin advantage over an "average" 9900k (and far far more vs trash tier 9900k's).
 

·
Registered
Joined
·
221 Posts
Discussion Starter · #11 · (Edited)
My chip isn't a good sample. It's average. Middle of the road. Unable to do 5 ghz FMA3 small FFT prime95 on Auto vcore and AC Loadline 1.6 mOhms and "Standard" Vcore loadline calibration because it gets too hot and crashes at 90C. And can barely do small FFT AVX Prime95 at the same settings (Auto vcore, LLC:Standard, ACLL 1.6 mOhms).

Yes I did "check FPU" only.
I'm aware that the other stress tests draw less power.
Have you tried check FPU with 1.30v bios set and LLC:High and checking your VR VOUT and comparing it to your previous 1.275v + Turbo attempt?

By the way, the "Julia" and a couple of the other FPU benchmark tests draw as much current as FMA3 prime95 ! WAY more than "Test FPU". It's very hard to get CURRENT IOUT or DTS temps even logged in HWInfo64 when running those because the entire system freezes during the most of the test no matter how stable you are. Probably why Aida doesn't include those in the stress tests, because people would just crash or reach 100C+ trying them for too long. Don't be surprised if you BSOD on those.

A 9900KS at its worst should have a +100 mhz bin advantage over an "average" 9900k (and far far more vs trash tier 9900k's).
Yeah you're right. If I got a chip that ran at the same setting with 40 mV less, I'd be super happy.

Some people, especially Youtubers, claim to have a good sample but run stress tests that are a joke.

Usually by running non avx tasks, adding an offset and testing games...

The one year warranty is cracking me up though given how most laptops always run at nearly TJmax for years on years.
 

·
Registered
Joined
·
341 Posts
Most of the youtube stability testing is inadequate and sets a false expectation. To the point, you see people touting their cinebench runs as stable.

At the minimum, you should run 10 loops of x264 stress test and OCCT large with avx2 for 90mins

If you can get away with P95 non avx, go for it!

There's no practical use case for p95 avx outside of saying you did it. It'll require a de-lid and custom loop to manage the heat and you'll probably pull amps well above intel spec. I'd put p95 avx as a component stress test rather than a stability test. There is a difference between the two.

The whole point of stability testing is to confirm if you're going to spit out errors under different loads. That's where OCCT excels at compared to anything else I've tested.
 

·
Registered
Joined
·
245 Posts
Very interesting to check how much Intel has binned their KS version compare to the original 9900k.
I have an old stepping R0 9900k and I have tried the same test of Aida64 FPU only (with AVX).
Stable at 1.20v with no error after 6 minutes, but I don't understand why my power is still so high? (around 200w)
Is it because I use an air cooler? (more heat=more leakage=more power?)
Something else I don't understand is why such a huge difference between my VID (voltage asking by my CPU) and real voltage needed?
(Aida is showing 1.199v vs 1.398v!)
 

Attachments

·
Facepalm
Joined
·
10,320 Posts
Very interesting to check how much Intel has binned their KS version compare to the original 9900k.
I have an old stepping R0 9900k and I have tried the same test of Aida64 FPU only (with AVX).
Stable at 1.20v with no error after 6 minutes, but I don't understand why my power is still so high? (around 200w)
Is it because I use an air cooler? (more heat=more leakage=more power?)
Something else I don't understand is why such a huge difference between my VID (voltage asking by my CPU) and real voltage needed?
(Aida is showing 1.199v vs 1.398v!)
Power is high because CPU Package Power is ViD * Amps, not vcore * amps.
If your VRM supports current monitoring, you can get the real power draw from Current IOUT (Amps) and Power POUT (Watts).

The VID shown is also affected by the default VID written for that cpu multiplier, which is then affected by both Thermal Velocity Boost voltage optimizations (VID decreases 1.5mv every 1C drop, from 100C to 0C) and the AC Loadline written into the processor.
 

·
Registered
Joined
·
221 Posts
Discussion Starter · #15 ·
Very interesting to check how much Intel has binned their KS version compare to the original 9900k.
I have an old stepping R0 9900k and I have tried the same test of Aida64 FPU only (with AVX).
Stable at 1.20v with no error after 6 minutes, but I don't understand why my power is still so high? (around 200w)
Is it because I use an air cooler? (more heat=more leakage=more power?)
Something else I don't understand is why such a huge difference between my VID (voltage asking by my CPU) and real voltage needed?
(Aida is showing 1.199v vs 1.398v!)
****, 1.2 sounds like a badass CPU, might as well be a 9900ks to be honest.
 

·
Registered
Joined
·
221 Posts
Discussion Starter · #16 · (Edited)
Most of the youtube stability testing is inadequate and sets a false expectation. To the point, you see people touting their cinebench runs as stable.

At the minimum, you should run 10 loops of x264 stress test and OCCT large with avx2 for 90mins

If you can get away with P95 non avx, go for it!

There's no practical use case for p95 avx outside of saying you did it. It'll require a de-lid and custom loop to manage the heat and you'll probably pull amps well above intel spec. I'd put p95 avx as a component stress test rather than a stability test. There is a difference between the two.

The whole point of stability testing is to confirm if you're going to spit out errors under different loads. That's where OCCT excels at compared to anything else I've tested.
Yeah because they claim 5 Ghz can be easily achieved with any 9900k 😂 Especially jayzs2cents is a dum dum when it comes to that.

My first 9900k didn't even manage to pull off 4.9 with the Aida 64 fpu test .
 

·
Registered
Joined
·
163 Posts
Very interesting to check how much Intel has binned their KS version compare to the original 9900k.
I have an old stepping R0 9900k and I have tried the same test of Aida64 FPU only (with AVX).
Stable at 1.20v with no error after 6 minutes, but I don't understand why my power is still so high? (around 200w)
Is it because I use an air cooler? (more heat=more leakage=more power?)
Something else I don't understand is why such a huge difference between my VID (voltage asking by my CPU) and real voltage needed?
(Aida is showing 1.199v vs 1.398v!)
I'd be happy to have a 9900k (KS) of that kind like 1.2v for 5ghz.
 

·
Registered
Joined
·
221 Posts
Discussion Starter · #18 ·
Very interesting to check how much Intel has binned their KS version compare to the original 9900k.
I have an old stepping R0 9900k and I have tried the same test of Aida64 FPU only (with AVX).
Stable at 1.20v with no error after 6 minutes, but I don't understand why my power is still so high? (around 200w)
Is it because I use an air cooler? (more heat=more leakage=more power?)
Something else I don't understand is why such a huge difference between my VID (voltage asking by my CPU) and real voltage needed?
(Aida is showing 1.199v vs 1.398v!)
Power is high because CPU Package Power is ViD * Amps, not vcore * amps.
If your VRM supports current monitoring, you can get the real power draw from Current IOUT (Amps) and Power POUT (Watts).

The VID shown is also affected by the default VID written for that cpu multiplier, which is then affected by both Thermal Velocity Boost voltage optimizations (VID decreases 1.5mv every 1C drop, from 100C to 0C) and the AC Loadline written into the processor.
.

The same wattage is odd though given it is fed 80mv less. You think Asus' 1.2 is equal to gigabyte's 1.28? Almost 100mv offset? 😮
 

·
Registered
Joined
·
245 Posts
Power is high because CPU Package Power is ViD * Amps, not vcore * amps.
If your VRM supports current monitoring, you can get the real power draw from Current IOUT (Amps) and Power POUT (Watts).

The VID shown is also affected by the default VID written for that cpu multiplier, which is then affected by both Thermal Velocity Boost voltage optimizations (VID decreases 1.5mv every 1C drop, from 100C to 0C) and the AC Loadline written into the processor.
My gosh! You're right! The power value I was monitoring is not my real power consumption!
It's based on the theorical VID asked by my CPU!

But I've found the real values given by ASUS EC sensors (Embedded Controller) for both power and current.
And indeed, they are way lower!

So I did again the same test (Aida64, FPU only, with AVX) this time with only 1.19v and now my power consumption is about 160w (instead of my previous 200w based on high VID asked by the CPU)
And the current is 135 amps: check my screenshot.

You give me a good news today :) Because until now I based my OC to stay in Intel max safe spec, but with a wrong higher power value!
Which means I have more room to increase voltage and thus power :)
 

Attachments

·
Registered
Joined
·
283 Posts
9900KS at 5.1Ghz all core, 4.7Ghz cache, 1.23v BIOS with 1.172v under load.

Edit: I've tested up to 5.3Ghz in BFV and Cinebench but will do more testing for Realbench and AIDA64 stability. 5.3Ghz with my 280mm AIO will be my limit and I am willing to be the thermals under 100% AVX load are going to skyrocket. That said, these synthetic loads far outstrip my daily usage and gaming needs so 5.3Ghz would be a viable clock for me. I will still likely stick with 5.1Ghz at low voltage as my daily driver.
 

Attachments

1 - 20 of 514 Posts
Top