Question Overclock 9700K - Degraded?

[katulen]

Hi gurus.

First of all, specs:

i7 9700K
NZXT Kraken X72
Asus Rog Strix Z390-F GAMING
Asus Rog Strix RTX2070 O8G GAMING 8GB GDDR6
Corsair Vengeance RGB 16GB: 2x8GB DDR4 3000MHz
Samsung 970 EVO Plus MZ-V7S500BW 500GB (NVMe) - M.2 Card
Toshiba P300 3.5´´ 2TB - 7200rpm 64MB SATA-600
Corsair RM750x
NZXT H series H700i Window Black

I started to experience some instability with my 5GHz overclock.
I decided I wanted to start from scratch, but before I did so, I updated BIOS to newest version and it reset everything as well.

I am currently struggling to hit the same overclock as previously, and I wondered if my CPU might have degraded.

The overclock which were stable before was:

Multiplier: 50
XMP: Enabled
AVX: Auto
MCE: Auto
LLC: 5
Manual voltage: 1.385v
VCORE @ load: 1.350v
VCCIO: 1.250v
VCCSA: 1.150v

It passed 8 hours Realbench, 2 hours Prime95 26.6, Intel Burn Test & OCCT Linpack NonAVX.
Now it struggle passing 2 hours of Realbench, 15mins of P95 etc.

My temps has always been fine, hovering around 50-65 under load (gaming), so I doubt that should have caused any damage.

So what happened throughout the last 10 months or am I doing something wrong ?

Thanks in advance.

Best regards

 

[zx128k]

New BIOS versions can affect overclocking. You need to reset to stock cores and check the RAM is not erroring. Once RAM is fine, test with aida64. Once stock is fine, then start overclocking the cpu.

A new BIOS can force you to start from scratch. Thats why you never update unless there is a big issue.

The issue with your original stress test, is that you did not test the RAM with memtest at your max frequency. Sometimes as you increase the frequencies the IMC starts to become unhappy.

With my Ryzen system for example, I tightened the timings on the RAM and overclock it to 3800. This seemed fine at stock core frequencies. Once I enable PBO EDC bug the whole RAM overclock became unstable.

So I went back to stock and ran memtest. Everytime I ran memtest I got random results. Different settings would pass then not pass.

What I found was I had left the Termination block and CAD_BUS block timings for the motherboard to auto detect as they were the same as the DRAM Calculator for Ryzen version 1.6.2. Sometimes when I turned on the PC the system was stable and at other times the RAM would fail the test with errors.

Took a lot of testing to find the stable settings so that every time I boot my PC it will be stable. I could even get tighter memory timings afterward. Now I also could reach higher frequencies at low vcore values. I thought I had degraded the memory controller or cores but it was just the settings from DRAM Calculator for Ryzen 1.6.2 that were wrong and the latest DRAM Calculator for Ryzen agrees with my fix.

Unstable settings


Stable settings


Issue appears to have been fixed by changing CAD_BUS ClkDrv from 24 to 40. The power of one single setting.

 

[katulen]

New BIOS versions can affect overclocking. You need to reset to stock cores and check the RAM is not erroring. Once RAM is fine, test with aida64. Once stock is fine, then start overclocking the cpu.

A new BIOS can force you to start from scratch. Thats why you never update unless there is a big issue.

The issue with your original stress test, is that you did not test the RAM with memtest at your max frequency. Sometimes as you increase the frequencies the IMC starts to become unhappy.

With my Ryzen system for example, I tightened the timings on the RAM and overclock it to 3800. This seemed fine at stock core frequencies. Once I enable PBO EDC bug the whole RAM overclock became unstable.

So I went back to stock and ran memtest. Everytime I ran memtest I got random results. Different settings would pass then not pass.

What I found was I had left the Termination block and CAD_BUS block timings for the motherboard to auto detect as they were the same as the DRAM Calculator for Ryzen version 1.6.2. Sometimes when I turned on the PC the system was stable and at other times the RAM would fail the test with errors.

Took a lot of testing to find the stable settings so that every time I boot my PC it will be stable. I could even get tighter memory timings afterward. Now I also could reach higher frequencies at low vcore values. I thought I had degraded the memory controller or cores but it was just the settings from DRAM Calculator for Ryzen 1.6.2 that were wrong and the latest DRAM Calculator for Ryzen agrees with my fix.

Unstable settings


Stable settings


Issue appears to have been fixed by changing CAD_BUS ClkDrv from 24 to 40. The power of one single setting.

Thanks alot for your reply.

My system was unstable anyway, so I didnt mind updating BIOS, hence I could start from scratch with a fresh BIOS up to date.

Yes, my initial overclock (back 10 months ago) I did NOT test my memory, at least to the extend that you could qualify them stable.

However, when I suddenly ran into instability a few weeks ago, I disabled my OC and tested my ram with & without XMP enabled - using Memtest86, HCI Memtest & GSAT, and they are showing no errors what so ever. I have yet to test my ram in conjuction with my CPU OC.

I assume PBO EDC is AMD's equivalent of XMP right ?

I have done some testing throughout the night:

Multiplier: 50
XMP: Enabled
AVX: Auto
MCE: Auto
LLC: 5
Manual voltage: 1.390v
VCORE @ load: 1.350v-1.359v
VCCIO: 1.250v
VCCSA: 1.150v

For thermal:
Prime95 v26.6 small FFTS 15 MIN - Passed
Max temp: 97c

For stability:
Realbench v2.56 8 HOURS - Passed
Max temp: 92c

So by upping vcore 0.005v it will pass with no WHEA nor BSOD's.

What would you suggestion be to moving forward now?

Test my ram using CPU OC + XMP ?
And or reconfigure timings etc myself as you did?

UPDATE:
I have now done an hour of GSAT with CPU OC + XMP and 0 errors.
I am using GSAT, since I've read & heard that it is a more stringent and faster at picking up faulty ram than Memtest.

Thanks.

 

[zx128k]

Thanks alot for your reply.

My system was unstable anyway, so I didnt mind updating BIOS, hence I could start from scratch with a fresh BIOS up to date.

Yes, my initial overclock (back 10 months ago) I did NOT test my memory, at least to the extend that you could qualify them stable.

However, when I suddenly ran into instability a few weeks ago, I disabled my OC and tested my ram with & without XMP enabled - using Memtest86, HCI Memtest & GSAT, and they are showing no errors what so ever. I have yet to test my ram in conjuction with my CPU OC.

I assume PBO EDC is AMD's equivalent of XMP right ?

I have done some testing throughout the night:

Multiplier: 50
XMP: Enabled
AVX: Auto
MCE: Auto
LLC: 5
Manual voltage: 1.390v
VCORE @ load: 1.350v-1.359v
VCCIO: 1.250v
VCCSA: 1.150v

For thermal:
Prime95 v26.6 small FFTS 15 MIN - Passed
Max temp: 97c

For stability:
Realbench v2.56 8 HOURS - Passed
Max temp: 92c

So by upping vcore 0.005v it will pass with no WHEA nor BSOD's.

What would you suggestion be to moving forward now?

Test my ram using CPU OC + XMP ?
And or reconfigure timings etc myself as you did?

UPDATE:
I have now done an hour of GSAT with CPU OC + XMP and 0 errors.
I am using GSAT, since I've read & heard that it is a more stringent and faster at picking up faulty ram than Memtest.

Thanks.

PBO EDC is just an auto overclocking method for AMD cpu's. With my 2500k but not my 4930k, I had to increase core voltage after a period of time. So far everything seems good. Silicon lottery use up too 1.387V on the core. [1] [2] If you are at all worried about degradation, back off a notch. 1.39 vcore appear to be accepted as safe. [1] It could be that your overclock was on the edge (stable enough to pass tests at the start but still unstable) and just needed a little increase in vcore to become stable (a new BIOS can cause this as well). Hopefully from now on it will remain stable. I use Aida64 for testing, prime95 and memtest.

I would not increase VCCIO much above 1.250v for 24/7. [3]

[1] https://forums.tomshardware.com/threads/i7-9700k-5-1-ghz-1-39v-avg-is-it-safe.3536048/post-21361642
[2] https://siliconlottery.com/pages/statistics
[3] https://www.tweaktown.com/guides/8481/coffee-lake-overclocking-guide/index5.html

 

[katulen]

PBO EDC is just an auto overclocking method for AMD cpu's. With my 2500k but not my 4930k, I had to increase core voltage after a period of time. So far everything seems good. Silicon lottery use up too 1.387V on the core. [1] [2] If you are at all worried about degradation, back off a notch. 1.39 vcore appear to be accepted as safe. [1] It could be that your overclock was on the edge (stable enough to pass tests at the start but still unstable) and just needed a little increase in vcore to become stable (a new BIOS can cause this as well). Hopefully from now on it will remain stable. I use Aida64 for testing, prime95 and memtest.

I would not increase VCCIO much above 1.250v for 24/7. [3]

[1] https://forums.tomshardware.com/threads/i7-9700k-5-1-ghz-1-39v-avg-is-it-safe.3536048/post-21361642
[2] https://siliconlottery.com/pages/statistics
[3] https://www.tweaktown.com/guides/8481/coffee-lake-overclocking-guide/index5.html

Thank you alot for the info.

As you see in my OP max temps was 97c. I figured that my fan curve in my system was completely screwed, and some of my fans was only doing 40%, while CPU was hitting 97c.

I adjusted my fan curves and did some more testing.

With 5ghz @ 1.390v (1.350v @ load) I have passed 2x 8 Hours of realbench. Max temps in realbench & p95 small ftts is 87c. So I have a -10c decrease under load, which was pretty amazing.

I didn't know this, but I've been told that thermals can also cause instability. I thought thermals would only thermal throttle you, and not cause wheas/bsod's general instability.

So I started to figure, that maybe my system can be stable with lower voltages due to better cooling.

Besides that, I have dialed down VCCIO to 1.20v & VCCSA to 1.15v - no errors in GSAT.
However it wont pass IBT with vccio/sa changed lower. So not sure what to do about that tbh.

Also, there isnt a free version of Aida64 is there?

 

[zx128k]

Thank you alot for the info.

As you see in my OP max temps was 97c. I figured that my fan curve in my system was completely screwed, and some of my fans was only doing 40%, while CPU was hitting 97c.

I adjusted my fan curves and did some more testing.

With 5ghz @ 1.390v (1.350v @ load) I have passed 2x 8 Hours of realbench. Max temps in realbench & p95 small ftts is 87c. So I have a -10c decrease under load, which was pretty amazing.

I didn't know this, but I've been told that thermals can also cause instability. I thought thermals would only thermal throttle you, and not cause wheas/bsod's general instability.

So I started to figure, that maybe my system can be stable with lower voltages due to better cooling.

Besides that, I have dialed down VCCIO to 1.20v & VCCSA to 1.15v - no errors in GSAT.
However it wont pass IBT with vccio/sa changed lower. So not sure what to do about that tbh.

Also, there isnt a free version of Aida64 is there?

Interesting post on reddit about safe vcore.

View: https://www.reddit.com/r/overclocking/comments/dso1b8/whats_the_max_safe_voltage_for_a_247_overclock/?ref=share&ref_source=embed&utm_content=title&utm_medium=post_embed&utm_name=c01abc2152974988a422d0ca98a8fb6f&utm_source=embedly&utm_term=dso1b8

 

[katulen]

Thanks for the link.

After realizing my fan curve was utter crap, I have now adjusted it accordingly and getting a lot better results. It seems that the heat on my CPU caused my system to become unstable even at lower voltages. By controlling the heat better, I have managed to decrease my voltage quite a bit with same speed, 5GHz.

I have tested using Prime95 small ffts (15min) and Prime95 12-12 ffts in-place (1hour) & Realbench (8hours)

VCORE	VCORE @ LOAD	TEMP (max/avg)
1.385v	1.350v	87c/85c
1.375v	1.341v	86c/84c
1.365v	1.332v	85c/83c
1.355v	1.314-1.323v	84c/81c

1.355v passed in p95 both tests, however it failed Realbench 6.5 hours in with a BSoD.
So I have now upped the voltage 1.360v and is testing it through as well. Hopefully it will pass Realbench 8 hours, and when/if it does, I will check RAM to make sure the CPU OC is stable with the XMP, using GSAT, TESTMEM5 & Aida64 Extreme.

I am actually amazed about how much I could get my temperatures down, and therefor increase my stability. I was NOT aware that heat actually could make your system unstable. I had the idea, that you did thermal testing to make sure you were within the limits of tjmax and could go to longer stability tests without risking damaging your hardware.

But it seems getting my CPU down from 97c to ~85c has brought stability with it at lower voltages.

 

[lewinr]

not only does heat affect the stability of overclocked CPU and DRAM, heat also affects other components like VRMs that have a secondary impact on the stability of the system. I spent huge time trying to understand the source of instability in my CPU/RAM overclock by messing with timings in the BIOS, when finally it turned out to be a problem with heat dissipation around the VRMs and DRAM that was solved by aiming a fan directly at them.

When overclocking to the limits, even moving the PC to another room that might not have as effective air conditioning can impact stability. RAM tREFI timing is especially sensitive to heat.

 

[katulen]

not only does heat affect the stability of overclocked CPU and DRAM, heat also affects other components like VRMs that have a secondary impact on the stability of the system. I spent huge time trying to understand the source of instability in my CPU/RAM overclock by messing with timings in the BIOS, when finally it turned out to be a problem with heat dissipation around the VRMs and DRAM that was solved by aiming a fan directly at them.

When overclocking to the limits, even moving the PC to another room that might not have as effective air conditioning can impact stability. RAM tREFI timing is especially sensitive to heat.

Interesting.

So what kind of fan would you use to point for VRM or RAM for that sake?

 

[lewinr]

I just zip tied a standard 120mm case fan inside the case right over the DIMMs to constantly blow on them.