what is the best way to stress test PBO undervolt stability for 7800X3D?

[rcarlos243]

I have a 7800X3D with Asus X670E-E and undervolted it using PBO with -20 on all cores.

 

I tried OCCT and it passed for 1 hour.

 

I heard that PBO causes instability when the PC is doing light load or at idle.

 

What is the best way to stress test to make sure my PC is fully stable with PBO undervolt?

[filpo]

1 minute ago, rcarlos243 said:

What is the best way to stress test to make sure my PC is fully stable with PBO undervolt?

I mean you've already done one of the stress tests so it should be fine, you could go higher on the undervolt if you want

There's loads, cinebench, prime95, CPU-Z, AIDA64 and Passmark to name a few

[Mark Kaine]

10 minutes ago, rcarlos243 said:

I have a 7800X3D with Asus X670E-E and undervolted it using PBO with -20 on all cores.

 

I tried OCCT and it passed for 1 hour.

 

I heard that PBO causes instability when the PC is doing light load or at idle.

 

What is the best way to stress test to make sure my PC is fully stable with PBO undervolt?

if it doesn't crash, then its stable...

 

 

asking for some hypothetical usage scenario makes no sense when you can just test that yourself,  does it crash, no? its "stable"...  does it crash at random times or at "light loads"? well, its definitely not stable! 

 

 

also by nature of the subject,  you cannot test for all usage scenarios, some applications will be more susceptible to out of spec settings than others simply.  you'll know it when you know it basically  ¯\_(ツ)_/¯ 

[RONOTHAN##]

8 minutes ago, rcarlos243 said:

What is the best way to stress test to make sure my PC is fully stable with PBO undervolt?

Core Cycler can help, though I've still had some systems crash when it will pass Core Cycler for a few hours. A stress test for this is pretty much just use the system as usual and see if it blue screens. If it does, it's unstable and should be lowered, while if not, it's fine. 

[rcarlos243]

9 minutes ago, filpo said:

I mean you've already done one of the stress tests so it should be fine, you could go higher on the undervolt if you want

There's loads, cinebench, prime95, CPU-Z, AIDA64 and Passmark to name a few

all the stuff you mentioned only stress for max load.

 

It does not address system stability when the CPU is either idling or at very low loads.

[filpo]

1 minute ago, rcarlos243 said:

all the stuff you mentioned only stress for max load.

 

It does not address when the CPU is either idling or at very low loads.

then leave it on standby for hours

 

6 minutes ago, Mark Kaine said:

if it doesn't crash, then its stable..

basically this but in a wallpaper scenario

[Mark Kaine]

1 minute ago, rcarlos243 said:

does not address when the CPU is either idling or at very low loads.

yes, and? idle/low load is likely the typical usage, so does it crash or does it not??

[Mark Kaine]

12 minutes ago, Mark Kaine said:

also by nature of the subject,  you cannot test for all usage scenarios, some applications will be more susceptible to out of spec settings than others simply.  you'll know it when you know it basically 

repeating this for the op, because i think its important to understand. 

 

eg. some games with certain anticheat etc will be more likely to crash with out of spec settings than others, hence it cannot be tested for.

 

and again, testing for low load scenario makes no sense because if you turn the computer on, watch a video , etc, it already *is* in a low load state. 

[rcarlos243]

2 minutes ago, Mark Kaine said:

yes, and? idle/low load is likely the typical usage, so does it crash or does it not??

 

It seems like you have nothing useful to add here.

 

Doing some research, I have found tools like CoreCycler, OCCT with Core Cycling, Prime95 Small FFT assigned to 2 threads, Curve Optimizer with per core stability, etc...

 

I am still looking for tools or methodologies to test for stability under idle/low loads.

[RONOTHAN##]

8 minutes ago, Mark Kaine said:

and again, testing for low load scenario makes no sense because if you turn the computer on, watch a video , etc, it already *is* in a low load state. 

And it's not even all low load states that will crash. There's a CO setting on my 5900X where it will be mostly fine at the desktop (1-2 bsods did pop up over a month, though I was also messing with memory settings throughout time so I'm not going to 100% attribute it to the CO, even though the memory settings I was running should've just worked), but would consistently crash within an hour of using an Xbox 360 emulator of all things.  

 

It's just impossible to stress test every single load state, since there are hundreds of them on PBO2 capable CPUs, and getting a chip to reliably stay in any one of them is next to impossible. It's the main reason I hate the curve optimizer and wish they gave a more Intel/Nvidia-esc version of undervolting and V/F curve optimizations. 

[Mark Kaine]

4 minutes ago, rcarlos243 said:

 

It seems like you have nothing useful to add here.

 

Doing some research, I have found tools like CoreCycler, OCCT with Core Cycling, Prime95 Small FFT assigned to 2 threads, Curve Optimizer with per core stability, etc...

 

I am still looking for tools or methodologies to test for stability under idle/low loads.

no, but it seems you have issues comprehending what i wrote.  

 

actually you should mark my answer as the solution because it pretty much covers all bases lol.

 

 

*read* it again, and maybe apologize for being so rude to people who try to help you...

[Mark Kaine]

8 minutes ago, rcarlos243 said:

I am still looking for tools or methodologies to test for stability under idle/low loads.

this cannot be "tested" for... because its already the normal state of a computer. 

 

your continued ignorance of this fact is baffling. 

[rcarlos243]

1 minute ago, Mark Kaine said:

this cannot be "tested" for... because its already the normal state of a computer. 

 

your continued ignorance of this fact is borderline trolling.

 

You are the one who is name-calling when u have nothing useful to add.

 

The problem with PBO is that when you use a very high negative offset, you can stess test the cores under max load and it would be fine as there is sufficient power at the top end of V/F curve, but as soon as it is on the bottom end the V/F curve the thing becomes unstable.

 

Several workarounds can be done such as disabling C-state, setting the minimum processor state to a higher value, but those are band aid fixes and does not fully adress the issue.