Clock Watchdog Timeout

[toby12f]

System:

- EVGA RTX 3080 TI FTW3 Ultra Gaming OC
- Asus Rog Strix Z690-F Gaming Wifi
- Corsair Dominator Platinum DDR5 5600 MHz CL36 2x16 GB
- Intel i9-12900K
- Corsair H170i AIO (420mm radiator)
- Western Digital SN850 2 TB NVMe SSD
- SeaSonic Focus 850W 80+ Platinum

Apologies in advance if this is not the right spot for this question, in which case I would appreciate if you could point me in the right direction.

I keep getting CLOCK_WATCHDOG_TIMEOUT BSODs. I have tried with and without XMP II, with and without Asus AI overclocking, and even 100% default settings. I also thought it might have been because the motherboard jumper for additional CPU voltage wasn't 'enabled', so I moved it, but I am still getting this BSOD in <10 minutes on Prime95. The thermals on the CPU are all under 100°C on each core. Even with stock settings, the cores reach 5.3 MHz. I have been told this is an issue with the voltage, and I should raise VCore slightly, however I don't know how to do this. It even BSODs occasionally on idle loads, with only 1% of CPU in use. All my drivers are up to date, as is the BIOS. If the problem is indeed the CPU voltage, how do I go about raising it safely? Thanks again, any help is appreciated.

[Silent_Scone]

Chainbold wrote:
I disagree. 🙂 Intel itself is using P95 for their overclocking and stressing utility. They know why.

1. As stated many users do not use Prime and are stable as they need to be. This is a methodology that we endorse and have done for some time.

2. If in a commercial environment one could argue you shouldn’t be overclocking. No overclock is 100% stable.

3. Passing XTU Prime algorithms or conventional P95 without AVX doesn’t ensure stability. There are instances where certain platforms can pass P95 but fail in x265 encode, or other tasks. Any overclocker worth their salt will use a variety of tests. Real world AVX often makes more sense than synthetic.

4. P95 with AVX routines can degrade a CPU depending on the applied overclock and impede overclocking range.

5. Finding instability in games is no different to crashing Prime. You adjust the tune accordingly.

The take away ultimately is seeing things for what they are. If user A. Is stable at 5.3 all core and gaming, but cannot run P95, User B. Who is using P95 and cannot run 53 all core is welcome to do so, but shouldn’t try to instil on others that user A. cannot be stable, this is a falsehood. In a world where OC potential narrows constantly, impeding one’s performance to pass something you will never run makes less and less sense. Of course, it’s a free country - that doesn’t make either argument stronger or weaker.

[Chainbold]

Silent Scone@ROG wrote:
1. Â*As stated many users do not use Prime and are stable.
Â*

Others are not. It's a question of your testing standards and the level of stability you require.

P95 or similar Intel's XTU (that is based on P95) can be used in many ways, for softer testing, or hardcore. Go soft, or stay hard. 😉

[Jimbo93]

Silent Scone@ROG wrote:


2. If in a commercial environment one could argue you shouldn’t be overclocking. No overclock is 100% stable.

3. Passing XTU Prime algorithms or conventional P95 without AVX doesn’t ensure stability. There are instances where certain platforms can pass P95 but fail in x265 encode, or other tasks. Any overclocker worth their salt will use a variety of tests. Real world AVX often makes more sense than synthetic.

4. P95 with AVX routines can degrade a CPU depending on the applied overclock and impede overclocking range.

first hand observation is better than general rules
Any failed test that is hardware related is a failure, I dont mean to subscribe to any particular algorithm but P95 is one of a battery I would expect to pass before saying a system is without hardware fault lurking.
We are talking about a pass fail test where it is assumed thermals are kept in spec during the reasonable time it takes to pass or fail.

[Silent_Scone]

Jimbo93 wrote:
first hand observation is better than general rules
Any failed test that is hardware related is a failure, I dont mean to subscribe to any particular algorithm but P95 is one of a battery I would expect to pass before saying a system is without hardware fault lurking.
We are talking about a pass fail test where it is assumed thermals are kept in spec during the reasonable time it takes to pass or fail.

*

It’s an age old debate that brings out the same black and white thinking.

As I’ve said, there are users running higher clocks than you and are as stable as they need to be. Let’s not forget one can run Prime for 48 hours or longer and the system could easily throw a stop code later on. So how long is enough? What if someone ran 72 hours, are they more stable than you? How sure are you it won’t fall 6 hours after the test was halted? You catch my meaning :).

You certainly wouldn’t want to find out on some platforms. For maximum stability we want small FFT, right? To run a 5950x at an effective 4.4GHz we need 1.184v under load. This results in 274W through the die, closer to three times the stock TDP. Doesn’t sound too healthy, does it! ;).

[Jimbo93]

Silent Scone@ROG wrote:
*

It’s an age old debate that brings out the same black and white thinking.

As I’ve said, there are users running higher clocks than you and are as stable as they need to be. Let’s not forget one can run Prime for 48 hours or longer and the system could easily throw a stop code later on. So how long is enough? What if someone ran 72 hours, are they more stable than you? *

You certainly wouldn’t want to find out on some platforms. For maximum stability we want small FFT, right? To run a 5950x at an effective 4.4GHz we need 1.184v under load. This results in 274W through the die, closer to three times the stock TDP. Doesn’t sound too healthy, does it! 😉 *

You bring up an important point, and I think you do have to consider the extra pressure of the overclock vs. the length of the test. I wouldn't need to go more than 30m to an hour depending on amount of frequency boost I was trying to achieve.

[Silent_Scone]

Jimbo93 wrote:
You bring up an important point, and I think you do have to consider the extra pressure of the overclock vs. the length of the test. I wouldn't need to go more than 30m to an hour depending on amount of frequency boost I was trying to achieve.

Different strokes for different folks as they say. But you get my meaning. Not much point passing a stress test if your logic gates are as wide as the Grand Canyon after the fact! 😄

That kind of current on the above example is not something anyone should be endorsing.

[Jimbo93]

Silent Scone@ROG wrote:
Different strokes for different folks as they say. But you get my meaning. Not much point passing a stress test if your logic gates are as wide as the Grand Canyon after the fact! 😄

That kind of current on the above example is not something anyone should be endorsing.

I mentioned no frequency boost and did not intend to endorse any particular frequency/test regimen. In other words, one should research the specs vs. the stress expected.

[Silent_Scone]

Jimbo93 wrote:
I mentioned no frequency boost and did not intend to endorse any particular frequency/test regimen. In other words, one should research the specs vs. the stress expected.

Well, on the aforementioned platform with overclocking restrictions lifted, you shouldn’t run Prime AVX at all. It’s not something I’d recommend anyone do.

*

Chainbold wrote:
This would be a matter of concern. Is this an "assumption", or based on some research / evidence? I guess though that any extended stress testing will eventually degrade the CPU or RAM.

Rule of thumb one should try to stick below double the default TDP. Something that’s quite difficult to do with Prime AVX / small FFT. The above is a real figure taken from a real scenario. 270W through a 5950x package is brutal.


Again, just something else to consider when having to limit your OC potential.

[toby12f]

Silent Scone@ROG wrote:

*
It’s an age old debate that brings out the same black and white thinking.

As I’ve said, there are users running higher clocks than you and are as stable as they need to be. Let’s not forget one can run Prime for 48 hours or longer and the system could easily throw a stop code later on. So how long is enough? What if someone ran 72 hours, are they more stable than you? How sure are you it won’t fall 6 hours after the test was halted? You catch my meaning :).

You certainly wouldn’t want to find out on some platforms. For maximum stability we want small FFT, right? To run a 5950x at an effective 4.4GHz we need 1.184v under load. This results in 274W through the die, closer to three times the stock TDP. Doesn’t sound too healthy, does it! ;).

I’m sorry but if it blue screens within 10 minutes, there’s a problem. I don’t know how you’re arguing otherwise. I’m not even trying to run this for over 30 minutes. The way I look at it is simple: if it crashes on this test within 10 minutes, it’s sure as hell not going to even last an hour under normal load. And so far it has not. Either the quality of Asus boards is garbage this generation, or I have a defective one. There is no other logical explanation for why it cannot run a game whatsoever, no matter the BIOS settings.

[Silent_Scone]

toby12f wrote:
I’m sorry but if it blue screens within 10 minutes, there’s a problem. I don’t know how you’re arguing otherwise. I’m not even trying to run this for over 30 minutes. The way I look at it is simple: if it crashes on this test within 10 minutes, it’s sure as hell not going to even last an hour under normal load. And so far it has not. Either the quality of Asus boards is garbage this generation, or I have a defective one. There is no other logical explanation for why it cannot run a game whatsoever, no matter the BIOS settings.

There’s a problem with your memory OC, yes. One which is easier diagnosed with something other than Prime. At this point, all other clock domains should be stock. The post above outlines valid reasons to avoid using P95 at all, and doesn’t pertain to any one issue that can’t be resolved by other means (like a dedicated memory stress test ;))

* You need to be able to understand there’s a distinct difference there and are confusing the methodology with your own issue. Hence why I asked why you were even using P95 to diagnose a memory problem in the first instance.

HCI Memtest Pro

Karhu Ramtest

TM5

GSAT

To list but a few suites that will find memory instability better than Prime95.

Increase VDD and VDDQ to 1.4v and see if this helps further.

Set MCVDD to 1.25v (Extreme Tweaker>Advanced Memory Setting). You need to pick a suite from the above, I’d recommend Karhu Ramtest. Keep tuning until you’re able to pass for at least 1500-2000% coverage.