Overclock.net banner

9900KS lottery thread

73699 513
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.
281 - 300 of 514 Posts

·
everywhere and nowhere
Joined
·
332 Posts
Ok, 5.3 Ghz Prime95 latest version AVX disabled STABLE:

Asus Maximus XI Hero

1.395 vcore Bios
1.323 vcore load
LLC7
PLL Bandwidth=0
CPU Standby=1.050
CPU Multiplier: x53
Sync All Cores

Passed Prime95 non-avx latest version for over 1 hour, 0 errors

I can't go any higher as VRM is getting in the 90-95 degrees range...

Temps: hottest core 86-87 degrees, ambient: 22 degrees. Delidded, direct die, custom loop.

Not exactly earth shattering, but it's stable. Now I just need to dial it in some more.
 

Attachments

·
Registered
Joined
·
244 Posts
Ok, 5.3 Ghz Prime95 latest version AVX disabled STABLE:

Asus Maximus XI Hero

1.395 vcore Bios
1.323 vcore load
LLC7
PLL Bandwidth=0
CPU Standby=1.050
CPU Multiplier: x53
Sync All Cores

Passed Prime95 non-avx latest version for over 1 hour, 0 errors

I can't go any higher as VRM is getting in the 90-95 degrees range...

Temps: hottest core 86-87 degrees, ambient: 22 degrees. Delidded, direct die, custom loop.

Not exactly earth shattering, but it's stable. Now I just need to dial it in some more.
Be careful, you are completely out of Intel spec! :eek:
A voltage of 1.32 during such heavy load of 264w is degrading your chip! :skull:
At this power level, your load voltage should be less or equal to max 1.20v!
Temperature is not the only parameter to monitor.
Only use this settings of 5.3Ghz for short benchmarks if you know what you are doing...
 

·
everywhere and nowhere
Joined
·
332 Posts
Ok, 5.3 Ghz Prime95 latest version AVX disabled STABLE:

Asus Maximus XI Hero

1.395 vcore Bios
1.323 vcore load
LLC7
PLL Bandwidth=0
CPU Standby=1.050
CPU Multiplier: x53
Sync All Cores

Passed Prime95 non-avx latest version for over 1 hour, 0 errors

I can't go any higher as VRM is getting in the 90-95 degrees range...

Temps: hottest core 86-87 degrees, ambient: 22 degrees. Delidded, direct die, custom loop.

Not exactly earth shattering, but it's stable. Now I just need to dial it in some more.
Be careful, you are completely out of Intel spec! /forum/images/smilies/eek.gif
A voltage of 1.32 during such heavy load of 264w is degrading your chip! /forum/images/smilies/skull.gif
At this power level, your load voltage should be less or equal to max 1.20v!
Temperature is not the only parameter to monitor.
Only use this settings of 5.3Ghz for short benchmarks if you know what you are doing...
Well, I’ve thought 1.52v is max Intel „spec”? It was pulling 191 amps, about 250 w, temps well in check you really think it’s out of spec and degrading my cpu?
 

·
Facepalm
Joined
·
11,379 Posts
Well, I’ve thought 1.52v is max Intel „spec”? It was pulling 191 amps, about 250 w, temps well in check you really think it’s out of spec and degrading my cpu?
Yeah you're way past the safe point, but it's unclear what will happen to the chip long term, but if you go by the intel doc sheets...remember while they have 1.520v VID max, they also specify 193 amps max *AND* a 1.6 mOhm VRM loadline. And you're using 0.4 mOhms as shown below. Yes, you gave max VID and Amps in your reply, sure, but you also forgot completely about the VRM Loadline. Refer to the 9th generation specification sheet.

These settings are based on "Auto" voltages being used and loadline set to 1.6 mOhms. The 1.520v VID value (I believe Asus allows this to be exceeded by default, judging from @criskoe 's testing with CPU VID values) is based on what AC Loadline can send to the VRM as a target voltage.

If this VID limit is respected on auto voltage, AC Loadline can only send up to 1.520v maximum. This is based on a complex combination of factors like base CPU VID (this stops scaling at 5 ghz), which is usually between 1.15 to 1.25v, thermal velocity boost voltage optimizations, which will lower CPU VID by 1.5mv every 1C temp drop, starting at 100C---or alternatively you can say it RAISES the base VID 1.5mv every 1C temp increase starting at 1C, and then AC Loadline boosting this (let's say this value is 1.420v before AC Loadline boost), depending on CPU Current load (the higher the load the higher the boost).

AC Loadline, is then limited to 1.520v in how far it can boost this. If "SVID Offset" (VRM command 33h) is enabled, this value is extended up to 1.720v.
Then VRM loadline vdroop follows to drop this target voltage down from that value. So if we start at 1.20v and use a 1.6 mOhm VRM Loadline, 193 amps becomes:

1520 - (193 * 1.6) = 1.214v.

So if you had a 193 amp load, Auto voltage, a 1.6 mOhm VRM Loadline and a 1.520v VID cap, your load voltage would be no higher than 1.214-1.220v.
If the 1.520v is bypassed via Serial VID Offset command (this is not the same as offset voltage shown in your BIOS btw), AC Loadline could conceivably go up to like 1.65v before vdroop, then at 200 amps you're looking at 1.315v after vdroop.

The one thing Intel does not state is under what conditions this is allowed. They only give this statement:
1.52 + Offset voltage= 1.72V (8 core processors ONLY; 4 and 6 cores are limited to 1.520v (Why?)

Then below:
15. An IMVP8 controller to support S82 65W and 95W VCCORE need to have offset voltage (33h) capability and potentially
VCCORE output voltage (VID +Offset) may be set higher than 1.52V

Anyway, this "offset voltage" is clearly not the same as 'offset voltage' in your BIOS.
Gigabyte labels this 33h command as "SVID Offset" option.

Enabling this actually causes a power cycle in the motherboard and disables all voltage changing, freezing the last voltage mode as active (no idea why. Bug?)
but then the 1.520v VID cap is removed.
Asus seems to have this removed by default, with no way to enforce it back on.

Anyway....
1.520v assumes there is no current going into the chip. Zero amps.

You can't just throw tons of current into the chip and go "well I'm below 1.520v anyway so I'm safe".
That's exactly how so many 2600K chips degraded (before anyone even knew what amps and loadline were even doing).

You're using what I calculated to be a 0.4 mOhm loadline (1395 mv - (0.4 * 191A) = 1318mv), 1395mv=bios set, 1318mv=load voltage, 191 amps=current amps of screenshot, 0.4 mOhm= LLC7.
Your vcore sensor has 12mv resolution so it will not show that value.

Intel specs call for a *1.6* mOhm loadline. It's listed under "DC Loadline" (DC Loadline is used for power measurements) but next to it it says "VRM Loadline" (loadline slope within the VR Loop regulation capability)

1520mv - (1.6 * 191 amps) = 1215mv = 1.215v.
What's confusing is that they don't specify a minimum VRM loadline at all. But if you go back and look at their older Core 2 documents, where the VRM Loadline is 2.1 mOhms, they actually show you the loadline slope with an actual graph, and even say its "-2.1mv / Amp" in writing, with no provision for changing it. After those days they started removing the graphs and compressing everything.

You can use that chip at those settings, yes. Just avoid stress testing. Use it for gaming or benchmarking. You will draw a lot less current.
 

·
everywhere and nowhere
Joined
·
332 Posts
Yeah you're way past the safe point, but it's unclear what will happen to the chip long term, but if you go by the intel doc sheets...remember while they have 1.520v VID max, they also specify 193 amps max *AND* a 1.6 mOhm VRM loadline. And you're using 0.4 mOhms as shown below. Yes, you gave max VID and Amps in your reply, sure, but you also forgot completely about the VRM Loadline. Refer to the 9th generation specification sheet.

These settings are based on "Auto" voltages being used and loadline set to 1.6 mOhms. The 1.520v VID value (I believe Asus allows this to be exceeded by default, judging from @criskoe 's testing with CPU VID values) is based on what AC Loadline can send to the VRM as a target voltage.

If this VID limit is respected on auto voltage, AC Loadline can only send up to 1.520v maximum. This is based on a complex combination of factors like base CPU VID (this stops scaling at 5 ghz), which is usually between 1.15 to 1.25v, thermal velocity boost voltage optimizations, which will lower CPU VID by 1.5mv every 1C temp drop, starting at 100C---or alternatively you can say it RAISES the base VID 1.5mv every 1C temp increase starting at 1C, and then AC Loadline boosting this (let's say this value is 1.420v before AC Loadline boost), depending on CPU Current load (the higher the load the higher the boost).

AC Loadline, is then limited to 1.520v in how far it can boost this. If "SVID Offset" (VRM command 33h) is enabled, this value is extended up to 1.720v.
Then VRM loadline vdroop follows to drop this target voltage down from that value. So if we start at 1.20v and use a 1.6 mOhm VRM Loadline, 193 amps becomes:

1520 - (193 * 1.6) = 1.214v.

So if you had a 193 amp load, Auto voltage, a 1.6 mOhm VRM Loadline and a 1.520v VID cap, your load voltage would be no higher than 1.214-1.220v.
If the 1.520v is bypassed via Serial VID Offset command (this is not the same as offset voltage shown in your BIOS btw), AC Loadline could conceivably go up to like 1.65v before vdroop, then at 200 amps you're looking at 1.315v after vdroop.

The one thing Intel does not state is under what conditions this is allowed. They only give this statement:
1.52 + Offset voltage= 1.72V (8 core processors ONLY; 4 and 6 cores are limited to 1.520v (Why?)

Then below:
15. An IMVP8 controller to support S82 65W and 95W VCCORE need to have offset voltage (33h) capability and potentially
VCCORE output voltage (VID +Offset) may be set higher than 1.52V

Anyway, this "offset voltage" is clearly not the same as 'offset voltage' in your BIOS.
Gigabyte labels this 33h command as "SVID Offset" option.

Enabling this actually causes a power cycle in the motherboard and disables all voltage changing, freezing the last voltage mode as active (no idea why. Bug?)
but then the 1.520v VID cap is removed.
Asus seems to have this removed by default, with no way to enforce it back on.

Anyway....
1.520v assumes there is no current going into the chip. Zero amps.

You can't just throw tons of current into the chip and go "well I'm below 1.520v anyway so I'm safe".
That's exactly how so many 2600K chips degraded (before anyone even knew what amps and loadline were even doing).

You're using what I calculated to be a 0.4 mOhm loadline (1395 mv - (0.4 * 191A) = 1318mv), 1395mv=bios set, 1318mv=load voltage, 191 amps=current amps of screenshot, 0.4 mOhm= LLC7.
Your vcore sensor has 12mv resolution so it will not show that value.

Intel specs call for a *1.6* mOhm loadline. It's listed under "DC Loadline" (DC Loadline is used for power measurements) but next to it it says "VRM Loadline" (loadline slope within the VR Loop regulation capability)

1520mv - (1.6 * 191 amps) = 1215mv = 1.215v.
What's confusing is that they don't specify a minimum VRM loadline at all. But if you go back and look at their older Core 2 documents, where the VRM Loadline is 2.1 mOhms, they actually show you the loadline slope with an actual graph, and even say its "-2.1mv / Amp" in writing, with no provision for changing it. After those days they started removing the graphs and compressing everything.

You can use that chip at those settings, yes. Just avoid stress testing. Use it for gaming or benchmarking. You will draw a lot less current.
Thank you for the extensive answer!

BUT... this is somewhat confusing - see this guide:
It's JJ from ASUS there, and he's using 1.450v with LLC7 and LLC8 for his 9900k, so this would be also massively out of Intel spec, right? How so then they're putting out this guide - clearly not aimed at OC pros? (check last 5 mins where he tries to stabilise 5.2 GHz, also on Maximus XI Hero).

As I have the temps well under control could this be the situation when I've actually reached silicon limit before thermal limit of my loop? If yes, what can I do more to get more out of this chip - SAFELY? With temps well before tjmax I was wondering about trying 5.4, now I'm scared! :)
 

·
Facepalm
Joined
·
11,379 Posts
Thank you for the extensive answer!

BUT... this is somewhat confusing - see this guide: https://www.youtube.com/watch?v=jOXH2YIMi2Y It's JJ from ASUS there, and he's using 1.450v with LLC7 and LLC8 for his 9900k, so this would be also massively out of Intel spec, right? How so then they're putting out this guide - clearly not aimed at OC pros? (check last 5 mins where he tries to stabilise 5.2 GHz, also on Maximus XI Hero).

As I have the temps well under control could this be the situation when I've actually reached silicon limit before thermal limit of my loop? If yes, what can I do more to get more out of this chip - SAFELY? With temps well before tjmax I was wondering about trying 5.4, now I'm scared! :)
Yeah I saw that. I guess he didn't know those were unsafe values. Keep in mind no one really knew or looked at the intel sheets and understood the interplay between these values. These are just electrical guidelines. Meaning if you follow them, your CPU will be longterm reliable. It doesn't say that the chip will degrade if you pass them nor by how much, only that long term reliability is no longer assured if you do it.

My 2600k degraded rather quickly by starting at just 1.38v with strong loadline calibration when doing 12 hour prime AVX tests on it. It passed 1.38v (Bios set) initially but eventually started crashing and needing more slowly.
That's all I can say. But using 1.45v bios set with Level 8 loadline calibration is very bad. Buildzoid already showed that that actually--hurts-- stability by using level 8 LLC, because your transients get torpedo'd and its the minimum transients that determine your stability. Look at his probinator videos for proof on that (he tested the Gene and other boards).

https://elmorlabs.com/index.php/2019-09-05/vrm-load-line-visualized/
 

·
everywhere and nowhere
Joined
·
332 Posts
So ok, attempt at 5.3 Ghz # 2:

Prime95 v29.8b6 (latest) AVX disabled smallFFT > 1hour STABLE, 0 errors

CPU x53
UNCORE x47
1.385v BIOS
1.261v LOAD
LLC6
VCCIO 1.312v (Auto)
VCCSA 1.408v (Auto)

Ambient 22 deg.
Max core temp 78 deg.
CPU current 169A/215w

I assume now it's "safe" or still out of Intel "spec"?
 

Attachments

·
Registered
Joined
·
537 Posts
So ok, attempt at 5.3 Ghz # 2:

Prime95 v29.8b6 (latest) AVX disabled smallFFT > 1hour STABLE, 0 errors

CPU x53
UNCORE x47
1.385v BIOS
1.261v LOAD
LLC6
VCCIO 1.312v (Auto)
VCCSA 1.408v (Auto)

Ambient 22 deg.
Max core temp 78 deg.
CPU current 169A/215w

I assume now it's "safe" or still out of Intel "spec"?
Looks safe to me. Temps look nice. The only thing would be VCCSA. It’s a little high for 24/7. Nice chip. You should try LLC5. It would give you more stability. Try 1.40v LLC5 (it should drop to1.243v-1.252v under load)
 

·
Facepalm
Joined
·
11,379 Posts
So ok, attempt at 5.3 Ghz # 2:

Prime95 v29.8b6 (latest) AVX disabled smallFFT > 1hour STABLE, 0 errors

CPU x53
UNCORE x47
1.385v BIOS
1.261v LOAD
LLC6
VCCIO 1.312v (Auto)
VCCSA 1.408v (Auto)

Ambient 22 deg.
Max core temp 78 deg.
CPU current 169A/215w

I assume now it's "safe" or still out of Intel "spec"?
You did a LOT better there.
Instead of pulling 200 amps, you're only pulling 167 and at a much lower load voltage. As you can see, more vdroop is not necessarily a bad thing if your transient response improves from a looser loadline calibration + a higher bios voltage!

Keep in mind that no one really knows what the 'true' boundaries are. They are simply empirical guesses based on the specifications we have access to.
I tried asking Intel about precisely this and one guy (who was still working there, on Twitter) said "he was under NDA for discussing the 1.520v and loadline thing".
An article recently said he's an ex engineer now so maybe I'll ask him again (and get ignored or yelled at) :/
 

·
everywhere and nowhere
Joined
·
332 Posts
You did a LOT better there.
Instead of pulling 200 amps, you're only pulling 167 and at a much lower load voltage. As you can see, more vdroop is not necessarily a bad thing if your transient response improves from a looser loadline calibration + a higher bios voltage!

Keep in mind that no one really knows what the 'true' boundaries are. They are simply empirical guesses based on the specifications we have access to.
I tried asking Intel about precisely this and one guy (who was still working there, on Twitter) said "he was under NDA for discussing the 1.520v and loadline thing".
An article recently said he's an ex engineer now so maybe I'll ask him again (and get ignored or yelled at) :/
Thanks! Kinda weird such significant information is kept under NDA for whatever reason... oh well. Do you think upgrading to let's say Maximus XI Apex or Gene would give me more headroom still? ie. lower vcore for the same clocks?
 

·
Facepalm
Joined
·
11,379 Posts
Thanks! Kinda weird such significant information is kept under NDA for whatever reason... oh well. Do you think upgrading to let's say Maximus XI Apex or Gene would give me more headroom still? ie. lower vcore for the same clocks?
Stepping up to a better board with better VRM's and more caps/resistors/inductors will always improve transient response and allow you to tighten vcore. Usually it's a small amount, and the minimum transient voltage dip is what determines your stable voltage, as those dips if they dip below what your CPU actually needs will cause random crashes.

Obviously, the Gigabyte Aorus Xtreme, eVGA Dark and Maximus XI Apex will be the best in their class. Only the Dark is tested in these tests. The Gene is tested though.


 

·
everywhere and nowhere
Joined
·
332 Posts
Looks safe to me. Temps look nice. The only thing would be VCCSA. It’s a little high for 24/7. Nice chip. You should try LLC5. It would give you more stability. Try 1.40v LLC5 (it should drop to1.243v-1.252v under load)
Tried LLC5, needed 1.44v to be stable (Prime95 latest version, 1hr, avx disabled), ended up with more power consumption than with LLC6! I think it's the board VRM not up to the task at this point (Maximus XI Hero).
 

Attachments

·
Premium Member
Joined
·
109 Posts
Finally managed to install my new Corsair fans (ML120 Pro) and completed Prime 95 Small FFT with AVX.

Test ran for nearly 9.5 hours with 0 errors. Definitely happy with the overclock and it is all dialed in now.

Temps reached a max of 85c with majority of testing done at 82c.

CPU-z core voltage was 1.154v. This is with a Vcore of 1.32v!
 

Attachments

·
Registered
Joined
·
55 Posts
Wasn't sure where to post this but I figured some people might be interested in the data to compare with their own CPUs. I was a little bored last night and decided to do a quick test to see what the voltage scaling is like on my 9900KS. I tested only the core speed, leaving cache and ram at stock values (4300 and 2133 MHz respectively). My goal was pretty basic so it's by no means a measure of full system stability - but it is definitely an indication. I looked for the voltage required to do 3 passes of Cinebench R20 without error/bsod. There are two exceptions to this, I didn't test 5400 MHz at LLC5 (the data given is an estimation based on LLC6 results) and only one pass at LLC 6.

The scores are purely for comparison between frequencies as HWiNFO, TurboV and CPU-Z were running during the test. With overclocked ram and cache I get 5619 points in RB20 (5300 core, 5000 cache, 4300 mem), ~200 points more.

Note 1: You can definitely see the increased undershooting of LLC6 in play with the average increase of 9 mV needed during load. Not sure what happened at 5.2 GHz but looking at the curve I suspect the LLC 5 voltage could've probably been 10 mV lower. Probably got a fluke bsod and increased the voltage.
Note 2: I'm running my cpu direct die under a custom loop so handling the 180A 1,35v load of 5.4 GHz wasn't an issue. Maximum peak temps at the 5 GHz setting was 55 c while 5.4 GHz LLC 6 ended up at 80 c. With the IHS and stock solder you'd probably get these temperatures at 1.32-1.34v which is nearly a 100 mV difference. Keep this in mind if you're doing your own testing - temperatures will have an impact on your scaling.
Note 3: The relative voltage is again mostly for comparison to see what the voltage increase would be at a current load equal to 5 GHz. Reality is that a given instruction will never draw the same amount of current at two different frequencies. RB20, P95, games, they'll all have an increased current under load as you increase frequency - dictated by your voltage increase.
Note 4: The set voltages at 5000-5200 LLC 6 are actually low enough that I'd occasionally get WHEA errors without any load.

I'm really suprised by the load voltage at 5 GHz, I'd love to see some other peoples values for this to compare with.
 

Attachments

·
Facepalm
Joined
·
11,379 Posts
Wasn't sure where to post this but I figured some people might be interested in the data to compare with their own CPUs. I was a little bored last night and decided to do a quick test to see what the voltage scaling is like on my 9900KS. I tested only the core speed, leaving cache and ram at stock values (4300 and 2133 MHz respectively). My goal was pretty basic so it's by no means a measure of full system stability - but it is definitely an indication. I looked for the voltage required to do 3 passes of Cinebench R20 without error/bsod. There are two exceptions to this, I didn't test 5400 MHz at LLC5 (the data given is an estimation based on LLC6 results) and only one pass at LLC 6.

The scores are purely for comparison between frequencies as HWiNFO, TurboV and CPU-Z were running during the test. With overclocked ram and cache I get 5619 points in RB20 (5300 core, 5000 cache, 4300 mem), ~200 points more.

Note 1: You can definitely see the increased undershooting of LLC6 in play with the average increase of 9 mV needed during load. Not sure what happened at 5.2 GHz but looking at the curve I suspect the LLC 5 voltage could've probably been 10 mV lower. Probably got a fluke bsod and increased the voltage.
Note 2: I'm running my cpu direct die under a custom loop so handling the 180A 1,35v load of 5.4 GHz wasn't an issue. Maximum peak temps at the 5 GHz setting was 55 c while 5.4 GHz LLC 6 ended up at 80 c. With the IHS and stock solder you'd probably get these temperatures at 1.32-1.34v which is nearly a 100 mV difference. Keep this in mind if you're doing your own testing - temperatures will have an impact on your scaling.
Note 3: The relative voltage is again mostly for comparison to see what the voltage increase would be at a current load equal to 5 GHz. Reality is that a given instruction will never draw the same amount of current at two different frequencies. RB20, P95, games, they'll all have an increased current under load as you increase frequency - dictated by your voltage increase.
Note 4: The set voltages at 5000-5200 LLC 6 are actually low enough that I'd occasionally get WHEA errors without any load.

I'm really suprised by the load voltage at 5 GHz, I'd love to see some other peoples values for this to compare with.
Excellent work there.
 

·
Registered
Joined
·
244 Posts
Wasn't sure where to post this but I figured some people might be interested in the data to compare with their own CPUs. I was a little bored last night and decided to do a quick test to see what the voltage scaling is like on my 9900KS. I tested only the core speed, leaving cache and ram at stock values (4300 and 2133 MHz respectively). My goal was pretty basic so it's by no means a measure of full system stability - but it is definitely an indication. I looked for the voltage required to do 3 passes of Cinebench R20 without error/bsod. There are two exceptions to this, I didn't test 5400 MHz at LLC5 (the data given is an estimation based on LLC6 results) and only one pass at LLC 6.

The scores are purely for comparison between frequencies as HWiNFO, TurboV and CPU-Z were running during the test. With overclocked ram and cache I get 5619 points in RB20 (5300 core, 5000 cache, 4300 mem), ~200 points more.

Note 1: You can definitely see the increased undershooting of LLC6 in play with the average increase of 9 mV needed during load. Not sure what happened at 5.2 GHz but looking at the curve I suspect the LLC 5 voltage could've probably been 10 mV lower. Probably got a fluke bsod and increased the voltage.
Note 2: I'm running my cpu direct die under a custom loop so handling the 180A 1,35v load of 5.4 GHz wasn't an issue. Maximum peak temps at the 5 GHz setting was 55 c while 5.4 GHz LLC 6 ended up at 80 c. With the IHS and stock solder you'd probably get these temperatures at 1.32-1.34v which is nearly a 100 mV difference. Keep this in mind if you're doing your own testing - temperatures will have an impact on your scaling.
Note 3: The relative voltage is again mostly for comparison to see what the voltage increase would be at a current load equal to 5 GHz. Reality is that a given instruction will never draw the same amount of current at two different frequencies. RB20, P95, games, they'll all have an increased current under load as you increase frequency - dictated by your voltage increase.
Note 4: The set voltages at 5000-5200 LLC 6 are actually low enough that I'd occasionally get WHEA errors without any load.

I'm really suprised by the load voltage at 5 GHz, I'd love to see some other peoples values for this to compare with.
Very interesting, thank you for sharing!
I also have a 9900KS and I've passed 3 runs of Cinebench R20 without error at 1.11v and max 70°c (with air cooling at 23°c ambiant)
I use mine at 5.1 or max 5.2 Ghz because I'm limited by temperature (and not voltage) but I think we should have about the same voltage scaling.
And I have also noticed that the higher is the frequency and the higher stability depends on temperature!
So it would be more accurate to mention temperature when comparing OC...
 

Attachments

·
Registered
Joined
·
619 Posts
After a few hours testing, think my chip is average but i'll throw this out there -

MB: XI Gene - Stable at 5180mhz (51 base clock with bclk increase) with LLC 5 - 1.395v Bios, 1.271 under load . System is on air - noctua nh-d15 , hit 87 deg while stress testing (blender) , much lower in games

Can pass / stress test stable at 5200+ but need quite a few increments on voltage and gain like 4-5 degrees in temps - not worth it for 20ish mhz

Assume I could get more out of this chip with a better cooling system but happy enough for now. Considering an AIO but it may only net me 4-5 degs which may not really achieve much in overclocking results
 

·
Registered
Joined
·
1,435 Posts
After a few hours testing, think my chip is average but i'll throw this out there -

MB: XI Gene - Stable at 5180mhz (51 base clock with bclk increase) with LLC 5 - 1.395v Bios, 1.271 under load . System is on air - noctua nh-d15 , hit 87 deg while stress testing (blender) , much lower in games

Can pass / stress test stable at 5200+ but need quite a few increments on voltage and gain like 4-5 degrees in temps - not worth it for 20ish mhz

Assume I could get more out of this chip with a better cooling system but happy enough for now. Considering an AIO but it may only net me 4-5 degs which may not really achieve much in overclocking results
I would say you're a little above average IMO :)
I know of CPUs not performing that well.
 

·
Registered
Joined
·
785 Posts
Man this thread makes me feel like I was totally robbed when I bought my 9900K. It requires 1.250vcore to hold 4.70GHz, 1.300vcore for 4.80GHz and I've been unable to get it stable anywhere above that.

My 9900K STOCK VOLTAGE goes as high as 1.370vcore just to boost two cores to 5GHz - and this has been the case on both Z390 mobos I've had it in.

This purchase was shortly before the announcement of the 9900KS and clearly Intel had already begun binning for the KS sku long before its announcement.

EDIT: These voltages are for AVX Offset of 0 which I think is essentially required these days (IMO). MANY games I've been testing on this CPU seem to be using AVX now (Battlefied 1/5, BatteFRONT 2, Path Of Exile...basically ALL Ubisoft games, the list goes on and on.) And with the new consoles having Ryzen-based CPUs I fully expect them ALL to be running AVX instructions to at least some extent. Path Of Exile uses AVX to calculate all its particle effects now, for example, and really hammers this 9900K in particle-heavy situations (even in towns for example, where the player-worn cosmetics are all particle-effect-laden)
 

·
Registered
Joined
·
466 Posts
Putting apart what we said recently, what do you think about these benchmark scores related at frequencies? for sure i didn't won silicon lottery and motherboard is not optimized for this CPU, 4.8 and 5.0 are stable with P95, Realbench, OCCT, blender and so on... 5.2 it's and hazardous OC in my situation also with thermal throttling on heavy software (and almost for everyone i think)... so, looking at these results, any opinion? could i check the CPU performances in these scenarios, better with other benchmarks? tks
 

Attachments

281 - 300 of 514 Posts
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top