Undervolting performance testing

[Wing-Man]

What Undervolting what the best way to performance test to see when you have gone too fair.
Ive decided to make a second bios at stock speeds and under volt the crap out of my CPU but i dont want to have a negative impact on performance for games.

what the best benchmarks to use?

 

[Orcvader]

Cinebench works very well, I found Prime95 hits the CPU too hard so it begins to throttle. Cinebench will show speeds more closer to what you'll see in games.

 

[Wing-Man]

Cinebench works very well, I found Prime95 hits the CPU too hard so it begins to throttle. Cinebench will show speeds more closer to what you'll see in games.

that super, already have that one installed.

 

[humbug]

I capped my 5800X to 55 watts once as an experiment, it made absolutely no difference in games, it still boosted to 4.9Ghz.

 

[Wing-Man]

I capped my 5800X to 55 watts once as an experiment, it made absolutely no difference in games, it still boosted to 4.9Ghz.

i wont know how to do that? is there a power setting for max watt's?

 

[humbug]

i wont know how to do that? is there a power setting for max watt's?

Look for PPT in the BIOS.

 

[humbug]

I think its in the advanced PBO settings.

 

[Removed User 345456]

these days is it best to keep doing multiple runs vs the old days of running prime95 for hours?

 

[Wing-Man]

these days is it best to keep doing multiple runs vs the old days of running prime95 for hours?

prime as always been bad for cpu's and motherboard

 

[Wing-Man]

I think its in the advanced PBO settings.

i believe i have found the setting, the stock TDP for my cpu is 65w is 55w ok or should i play around?

EDIT:
So i turned on PBO to get to the setting and set 55w Max but now PBO is not letting the CPU boost because i have left all the other setting to auto.

EDIT / EDIT :
Found the power setting in a different menu so i can keep PBO off. its now boosting to 4.35 again and as dropped a few C's.
Thank you

 

[Removed User 345456]

prime as always been bad for cpu's and motherboard

bad as in what ott unrealistic loads? I always used it to see if an oc was stable? so is that irrelevant these days for testing undervolting?

 

[humbug]

I've had CPU's Prime stable only for them to crash in games... CPU's are far more complex these days than a Pentium 4.

 

[Orcvader]

I've had CPU's Prime stable only for them to crash in games... CPU's are far more complex these days than a Pentium 4.

This. The issue with Prime is it tests pretty much with a continuous heavy load, which is only really good for testing peak speed stability. Due to the nature of CPUs with rapid downclocking/boosting, you want to use a test that can test different load scenarios to make sure the CPU is also stable in those other points.

On my 3900X I did a lazy undervolt of 0.05v and PBO max boost + 200MHz, which boosted all core clocks considerably with a slight reduction in overall power consumption. I didn't mess with the PBO limits however.

 

[Robert896r1]

You can tweak Prime95 parameters to suit many use cases. From a continuous, current heavy workload to a transient heavy one and anything in between. Frankly speaking, a properly stable system is exactly that and requires testing with multiple test types.

On Ryzen, you need to watch your effective clocks not read clocks in hwinfo. It’s easy to clock stretch when reducing power limits.

 

[Wing-Man]

bad as in what ott unrealistic loads? I always used it to see if an oc was stable? so is that irrelevant these days for testing undervolting?

but it dose nothing of the sort. i remember having a system 24hr prime stable and it BSOD when i opened chrome..
its loads a system to a silly level all pushes past what is safe.

OcUK used to use it for the overclocked systems, the forums was full of people that have new systems what was crashing on boot...

 

[Removed User 345456]

but it dose nothing of the sort. i remember having a system 24hr prime stable and it BSOD when i opened chrome..
its loads a system to a silly level all pushes past what is safe.

OcUK used to use it for the overclocked systems, the forums was full of people that have new systems what was crashing on boot...

never heard of that happening back in the day? it was all we ever used to stress test for heat and stability upto the q6600 era.
so what actually does genuinely prove its stable these days now as I'm at a loss how something stressing the death past actual usage doesn't prove something is stable and how a Web browser doing bugger all in comparison could stress it?

 

[Robert896r1]

never heard of that happening back in the day? it was all we ever used to stress test for heat and stability upto the q6600 era.
so what actually does genuinely prove its stable these days now as I'm at a loss how something stressing the death past actual usage doesn't prove something is stable and how a Web browser doing bugger all in comparison could stress it?

https://www.ocbase.com/ This is a good piece of a software for stability testing.

To make it easy to identify where instability might sit, I generally, do the following:

- OC my ram and test and tune that with a stock cpu. To keep it simple, use the memtest in occt along with the large/avx2/normal test. 30mins for small changes, 1hr for when locking in something solid
- Then once my ram is good, I focus on the CPU aspects and use a combination of Y-Cruncher and large/avx2/extreme in occt. For YC, I'll run the following when it loads up: 0,1,7 and do about 5 cycles of that before doing an on hour of occt http://www.numberworld.org/y-cruncher/

The above has proven to be bulletproof for me on intel and amd: z390, z690, b550 and about 5 builds total between those platforms

A big gotcha on AMD. Clock stretching on AMD is very common and greatly exaggerated by undervolting. Clock stretching means what your cpu is running otherwise knows as "effective clock" is much slower than what you see in overlays which is generally "reported clocks."

Luckily hwinfo64 https://www.hwinfo.com/download/ has a seperate section under CPU for showing effective clocks. *That's* what you need to focus on for AMD. Intel doesn't really have these issue as we don't undervolt there and boost works differently.

 

[Wing-Man]

never heard of that happening back in the day? it was all we ever used to stress test for heat and stability upto the q6600 era.
so what actually does genuinely prove its stable these days now as I'm at a loss how something stressing the death past actual usage doesn't prove something is stable and how a Web browser doing bugger all in comparison could stress it?

there is no one test that dose it all. you need to test based on your use. if all you do is watch your tube and play games why do you use a test that hits the cpu with a constant heavy load, when want you need is constantly changing loads.

I use 3D Mark, RealBench, Cinebench R20 and aida64. i run them by there self in a loop and or more than one at the same time.
For ram you cant beat MemTest86 end off but ram dose one thing and only one thing. unlike a CPU that is constantly
changing work loads and types.

the first stress test is dose it post. but just because it posts doesn't mean it games.
when we are overclocking i keep bumping the multi until the system wont load to windows, that in itself is a test.

most CPU's will throttle when running prime if your not maxing the core how do you know its stable? then you open chrome and the cores hit full boost for 1 second and the system BSOF.

 

[eddiew]

it was all we ever used to stress test for heat and stability upto the q6600 era.

Tbqh, I still find it useful as a last resort. Was having erratic bluescreens on my faithful old 8700k, but P95 would reliably fail within seconds. Had to remove the overclock to get it to run overnight. Sadly it seems as though the old girl simply can't do 5ghz any more

(But CPUs have come a long way, and I feel like this is good motivation for a Ryzen 7000 later this year!)

 

[Wing-Man]

it was all we ever used to stress test for heat and stability upto the q6600 era

thats about 25 years ago.