![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Stability and overclocking - what do you consider stable
- Thread starter thecold
- Start date
Many moons ago I spent time with Prime testing, but when I got my current rig (Haswell-E) I had a hard time getting it stable - I remember something about avx loads...
It turned out that prime isn't a stability indicator for my usage - gaming and browsing.
So getting to the point - is it stable for your workload? Yes - good job, now go use it. No - time to tweak it again.
It turned out that prime isn't a stability indicator for my usage - gaming and browsing.
So getting to the point - is it stable for your workload? Yes - good job, now go use it. No - time to tweak it again.
![Zarathustra[H]](https://cdn.hardforum.com/data/avatars/m/9/9298.jpg?1677728934){kind=avatar}
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 36,459
Are we talking CPU or GPU?
Either way, I use the old school method. I expect 48 hours of overkill load (Prime95, Intel Burn Test, etc.) without any thread crashing.
This way you've validated that the worst case is stable, can throw anything at it you want, and never have to worry about it again.
Anything less than that is not stable to me.
I also arrive there using the old school method without any shortcuts.
1.) Start at default settings, except set all fans to max.
2.) Raise multiplier by one step.
3.) Run brief 15 minute stability test.
4.) If it passes, go back to #2.
5.) If it fails, raise core voltage by a tiny amount and go back to #3.
6.) keep looping until you either reach your max safe temp, or your max safe voltage
7.) Back down to last known 15 minute stable point.
8.) Run 48 hour stability test at these settings. (Check in on it occasionally to save time. If one thread crashes 9 hours in, you don't have to wait until the end of 48 hours to take action)
9.) If it fails, back down to a step just prior to that, and go back to #8.
10.) If it passes, Note max core temp observed at final settings during load test.
11.) Configure fans to target that temp
12.) You are done.
This is a little bit oversimplified as it ignores RAM speeds and other voltages in modern CPU's other than the core voltage, but it is an example
I know, many people don't have the patience to do it right anymore, but if you follow this approach you will always land at the max stable overclock ant the minimum possible voltage.
Never take shortcuts and skip to a later destination.
Never accept a setting which is unstable at high loads using the justification that "I don't load it that high". If it doesn't crash in games, but crashes in load tests, its still not stable
Either way, I use the old school method. I expect 48 hours of overkill load (Prime95, Intel Burn Test, etc.) without any thread crashing.
This way you've validated that the worst case is stable, can throw anything at it you want, and never have to worry about it again.
Anything less than that is not stable to me.
I also arrive there using the old school method without any shortcuts.
1.) Start at default settings, except set all fans to max.
2.) Raise multiplier by one step.
3.) Run brief 15 minute stability test.
4.) If it passes, go back to #2.
5.) If it fails, raise core voltage by a tiny amount and go back to #3.
6.) keep looping until you either reach your max safe temp, or your max safe voltage
7.) Back down to last known 15 minute stable point.
8.) Run 48 hour stability test at these settings. (Check in on it occasionally to save time. If one thread crashes 9 hours in, you don't have to wait until the end of 48 hours to take action)
9.) If it fails, back down to a step just prior to that, and go back to #8.
10.) If it passes, Note max core temp observed at final settings during load test.
11.) Configure fans to target that temp
12.) You are done.
This is a little bit oversimplified as it ignores RAM speeds and other voltages in modern CPU's other than the core voltage, but it is an example
I know, many people don't have the patience to do it right anymore, but if you follow this approach you will always land at the max stable overclock ant the minimum possible voltage.
Never take shortcuts and skip to a later destination.
Never accept a setting which is unstable at high loads using the justification that "I don't load it that high". If it doesn't crash in games, but crashes in load tests, its still not stable
Last edited:
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 36,459
Couldn't disagree more.
Either it is everything stable for the worst thing you could possibly throw at it for 48 hours, or it isn't stable.
Question. What's your stand on WHEA errors that tend to plague Zen 3 platform with current BIOSes? I spend last weekend tweaking everything I could think of to get rid of them and I think I succeed. Problem is, before I noticed, I could run most challenging TM5 for few hours, no memory errors. On the same time, every now and then WHEA warring (not even an error) related to FLCK clock was recorded.
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 36,459
I have not played with a Zen3 yet
By WHEA error, do you mean that which displays on a blue screen? I'm pretty sure is consider any blue screen "not stable".
But it seems like you are referring to something else. I haven't seen those. It seems like there is something funky going on, but I couldn't tell you if it is a BIOS issue, software issue or something else.
If you run something like Prime95, do any threads crash?
I'm referring to this type of errors. It's not mine, I got rid of mine, at least with current setting, but before that I coud pass OCCT, Prime95, endless loops of Cinebench or TM5 no issues, yet windows event tracker was recording these warnings.I have not played with a Zen3 yet
By WHEA error, do you mean that which displays on a blue screen? I'm pretty sure is consider any blue screen "not stable".
But it seems like you are referring to something else. I haven't seen those. It seems like there is something funky going on, but I couldn't tell you if it is a BIOS issue, software issue or something else.
If you run something like Prime95, do any threads crash?
BR
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 36,459
Hmm
I'm really not familiar enough with overclocking Zen at all, let alone Zen3 to say for sure, but this does look like the type of random error that occurrs when you are right on the hairy edge of stability, and probably should either give it a tiny bit more voltage, or drop down the clocks a little.
I will have to defer to someone with specific experience though to make sure.
Last edited:
yea + the buggy BIOS
it was the same with Zen 2 platform, for me it was a good lesson that you may seems stable (some people with this issues pushed hundreds of hours of workloads and gaming through their machines) and yet, there's something fishy in the background, not enough to impact stability and performance (some of WHEA errors can do that) but enough to trigger warning event
Zarathustra[H]
Extremely [H]
- Joined
- Oct 29, 2000
- Messages
- 36,459
yea + the buggy BIOS
One thing I learned a long time ago is not to always assume that a higher clock is faster, but to bench it to make sure.
I've had many cases of my max stable overclock being slower than the next multiplier down
Sometimes as you reach the limit of what a system is capable of it starts having random issues. Maybe not enough to crash a thread in a stability test, but definitely enough to harm performance.
Maybe what is going on here is that the error correction in newer Zen chips can actually catch this and report it to the OS.
Not sure. It's an educated guess of a theory though
thesmokingman
Supreme [H]ardness
- Joined
- Nov 22, 2008
- Messages
- 6,617
Wait for a mature bios.
i used to run prime 95 and 3d mark overnight looping at the same time for stability check! overclkocked intel 2600k at 4600mhz [1.336V] for last 8yrs. The only thing that really seemed to stuff up stability was nvidia drivers ! it would start off a crash after a few weeks then it would gradually occur more often to the point of every 10 mins.over the course of a few months! it was worse just sitting there at the desktop doiing nothing and it would crash. sometimes it would crash playing games but it was so random. took me ages to figure it out ! like 6 months ! i was using a gigabyte GTX970 G1 gaming in sli ! found the late 380.xx series drivers were crap right up until early 410.xx after that they have been good! actually i had earlier ones too that caused instability but i didnt realixze what it was at the time!
if you can prime 95 it for 5 mins it will be close ! after that i would just run it overnight with 3d mark looping as well! if its still going in the morning then its probly ok ! and just use it for your normal activities ! good cooling is a must!
i have ryzen 5600x now so there is no point overclocking it to game with ! i just run ram at gskill ripjaws V C16 16 16 36 @ 1866mhz with fclk uclk and memclk locked at 1:1:1 [ it didnt like it at 1900mhz]
i fiddled with PPT TDC and EDC ! helps quite a bit in multithread loads but no better for gaming so i leave it stock!
hyper 212 black rgb runs cool ! low 60s C max load
if you can prime 95 it for 5 mins it will be close ! after that i would just run it overnight with 3d mark looping as well! if its still going in the morning then its probly ok ! and just use it for your normal activities ! good cooling is a must!
i have ryzen 5600x now so there is no point overclocking it to game with ! i just run ram at gskill ripjaws V C16 16 16 36 @ 1866mhz with fclk uclk and memclk locked at 1:1:1 [ it didnt like it at 1900mhz]
i fiddled with PPT TDC and EDC ! helps quite a bit in multithread loads but no better for gaming so i leave it stock!
hyper 212 black rgb runs cool ! low 60s C max load
drutman
Limp Gawd
- Joined
- Jan 4, 2016
- Messages
- 473
It is stable if I can run my PC all day with my software/HW/gaming needs and not crash or exceed max CPU Temps of 85 C.
I take measurements all day as a professional scientist when I am on leisure time I do not care to run benches or stress tests.
I take measurements all day as a professional scientist when I am on leisure time I do not care to run benches or stress tests.
drutman
Limp Gawd
- Joined
- Jan 4, 2016
- Messages
- 473
Yep measurebators assume this all the time, my GPU a 390X, runs 10 FPS faster at a 1200 vs stock 1500 MHz mem clock due to better timings at lower clock and lower thermals.One thing I learned a long time ago is not to always assume that a higher clock is faster, but to bench it to make sure.
I've had many cases of my max stable overclock being slower than the next multiplier down
Sometimes as you reach the limit of what a system is capable of it starts having random issues. Maybe not enough to crash a thread in a stability test, but definitely enough to harm performance.
Maybe what is going on here is that the error correction in newer Zen chips can actually catch this and report it to the OS.
Not sure. It's an educated guess of a theory though
spine
2[H]4U
- Joined
- Feb 4, 2003
- Messages
- 2,705
Totally, practically this!
I'm certain there's a stress test that would crash my PC, but why would I run it?
If there's an app you use that kills your OC, you are unstable, otherwise good to go. Handbrake will crash my 5.4 overclock with enough time, so I set it in the bios if all 8 cores are used, then it's only 5.3. Dead stable.
