how do you test if a system is stable when undervolting

bbzidane · Post by **bbzidane** » Wed Dec 02, 2009 5:31 pm

im curious to know what program people use to test for stability when undervolting.

i ran prime95 with 2 threads (for amd be2400), it ran overnight, with multiplier 10x at 0.95V, but it would still randomly crash my machine

to avoid random crashes, i now have multiplier 8.0 at 0.95V. this makes the system seem really slow to me

any advice?

bbzidane · Post by **bbzidane** » Wed Dec 02, 2009 5:37 pm

one more thing im wondering,
can a power hungry device cause less power going to another device?

for example, if a hard drive is spinning up/seeking a lot, could it cause less volts to be delivered to the cpu?

theycallmebruce · Post by **theycallmebruce** » Wed Dec 02, 2009 8:06 pm

Prime95 seems to be a widely used test for this purpose. I used it to determine the underclock voltage for mine, and it worked well for me.

As for advice, what is your end goal? Are you aiming to reduce maximum power consumption at full CPU load to bring down max fan speeds? Do you have a target CPU fan RPM or target temperature in mind?

What is the stock multiplier and voltage for that CPU, and what is the lowest stable voltage at the stock multiplier with your particular CPU?

Post by **MikeC** » Wed Dec 02, 2009 11:12 pm

bbzidane wrote:for amd be2400...... to avoid random crashes, i now have multiplier 8.0 at 0.95V. this makes the system seem really slow to me

any advice?

Is that CPU unlocked somehow? AFAIK, the multiplier is 11.5. How does it run at 8x?

for example, if a hard drive is spinning up/seeking a lot, could it cause less volts to be delivered to the cpu?

Very unlikely that a single drive would overstress your PSU unless you're running the PSU at its limit. The drive will not usually go over 10W; idle could be 4~7W. Of course, if you have a POS PSU, it could be worth trying a different PSU.

theycallmebruce · Post by **theycallmebruce** » Wed Dec 02, 2009 11:20 pm

I thought BE (Black Edition) meant unlocked multiplier?

Post by **MikeC** » Thu Dec 03, 2009 1:10 am

theycallmebruce wrote:I thought BE (Black Edition) meant unlocked multiplier?

BE prefix is AMD's code for low energy processors -- or was. Some series have an e at the end like 5050e. Black Edition models are referred specifically as Black Edition, afaik. None are 45W models.

rob77 · Post by **rob77** » Thu Dec 03, 2009 6:03 am

Black Edition processors have unlocked multipliers - to facilitate overclocking.

'e' suffix processors are low power variants

These tend to be at opposite ends of the spectrum, power-wise:

Phenom II X4 905e @ 65W vs Phenom II X4 945 Black Edition @ 125W

bbzidane · Post by **bbzidane** » Thu Dec 03, 2009 11:19 am

theycallmebruce wrote:As for advice, what is your end goal? Are you aiming to reduce maximum power consumption at full CPU load to bring down max fan speeds? Do you have a target CPU fan RPM or target temperature in mind?

What is the stock multiplier and voltage for that CPU, and what is the lowest stable voltage at the stock multiplier with your particular CPU?

i guess i just want to have it use as little power as possible, but still have decent speed when i need it. The machine is on 24/7, so i want it to idle at the lowest possible setting. im using a thermalright ultra 120 fanless. it does get warm in the case, but nothing too concerning.

the stock setting is 11.5x @ 1.25V

i have gotten the cpu to run 5x @ 0.75V for ~9hrs of prime95 and 10x @ 0.95V of prime95 for ~9hrs
but with normal usage of the machine, i would still experience random crashes

MikeC wrote:
bbzidane wrote:for amd be2400...... to avoid random crashes, i now have multiplier 8.0 at 0.95V. this makes the system seem really slow to me

any advice?
Is that CPU unlocked somehow? AFAIK, the multiplier is 11.5. How does it run at 8x?

i should mention that im not undervolting with the bios, but with software.
not sure if that means if my cpu is unlocked or not.

I use rmclock to tweak how the cpu is throttled. I have two settings configured, using the lowest multiplier (5x), and the highest multiplier at 0.95V.

Post by **NeilBlanchard** » Thu Dec 03, 2009 5:21 pm

What voltages and speed does Cool 'N Quiet do with that CPU? If it downclocks to 1gHz and runs at a low voltage when you are not using it, why not just undervolt at the stock speed?

Post by **MikeC** » Thu Dec 03, 2009 7:04 pm

I think you're confusing CnQ settings with full load settings, bbzidane. The CPU is multiplier-locked. When it's loaded, it will run at 200MHz x 11.3 -- 2.3MHz regardless of your rmclock settings.

When you run Prime95, that's puts the CPU at 100% load, and the CPU runs at stock voltage and clock speed. So it's perfectly stable. When you talk about random crashes, I'm guessing you're talking about when the system is at idle. This means that you are undervolting too much w/ rmclock.

I agree with NeilBlanchard -- you should try undervolting the CPU in the BIOS, and see if CnQ will still work. On some boards it doesn't work when CPU voltage is not at default. But if it does work, maybe let CnQ take care of lowering idle power instead of fiddling with rmclock. You need to up that low power state voltage anyway to get stability back.

Do you have some way of measuring AC power anyway? Because all this fiddling with power states and trying to reduce power is like shooting in the dark w/o a power meter. My guess is that your rmclock setting probably have maybe a 2W effect af the AC plug. Maybe more, but not much more. But if you reduce the power at full load, the difference will be bigger -- maybe 5-6W or more.

Of course how relevant this is to reducing total energy consumption (assuming that's what you're after) depends on how hard you run your machine. If you're stressing the PC to 100% load less than 10% of the time that it is on (true for the vast majority of non-gamers), then the effect will be extremely small... tho you can certainly argue that it all adds up.

Klusu · Post by **Klusu** » Fri Dec 04, 2009 1:43 am

Mike, you are wrong. Locked CPU means not higher. Any Athlon can run with lower multiplier (and lower V) 100% loaded.
As to the crashes - the problem is switching between states, MB's VRM may be not very good, try a bit higher voltage (one of the two or both). Test by switching manually with Prime running (or find some way to switch automatically).

Mats · Post by **Mats** » Fri Dec 04, 2009 5:12 am

bbzidane wrote:im curious to know what program people use to test for stability when undervolting.

Have you tried Intel burn test or LinX?

In my own experience, I used Prime95 on my Q6600 with Small FFTs, which showed that it was stable for 10 hours overnight at the current voltage. However, running Linpack with the most memory I had available (I have 64-bit XP, so I see all 4 GB of my RAM), it indicated instability. I had to raise the voltage up a couple of notches to get it fully stable.

Cistron · Post by **Cistron** » Fri Dec 04, 2009 6:38 am

MikeC wrote:Is that CPU unlocked somehow? AFAIK, the multiplier is 11.5. How does it run at 8x?

I always thought going below the set mulitplier isn't a problem? Contrary to increasing the multiplier.

edit: oh I see Klusu has answered this already. Disregard my overhasty post.

mrle · Post by **mrle** » Fri Dec 04, 2009 6:45 am

I had a similar issue with my old Athlon 64 X2. Prime tests were 100% stable, yet it would crash the machine in normal everyday usage. Turns out it was switching between states that caused the crashes, like Klusu said.

I solved it by manually setting the lowest possible voltage and fixed 10x multiplier in the BIOS, think it was around 1.0V, or maybe a little less, like 0.95. Also I disabled C'n'Q.

Post by **NeilBlanchard** » Fri Dec 04, 2009 6:50 am

Hi,

You should make sure that you are using the latest CPU driver -- this might eliminate the crashes. I think that Cool 'N Quiet is probably already doing what you want to do.

Klusu · Post by **Klusu** » Fri Dec 04, 2009 7:06 am

Wrong again. CPU driver does not matter, he uses another software, designed to override the CPU driver. Cool'n'Quiet is rather limited, no adjustments.

bbzidane · Post by **bbzidane** » Fri Dec 04, 2009 8:52 am

thanks for all the feedback

i stressed the cpu at the lowest setting by disabling all the other multipliers (still not sure if that is the right term), so it forces it to run 200mhz x 5, 1.0ghz, with prime95 running, which it is stable with voltage 0.75V.

so as some of you mentioned, it is most likely the machine hangs when changing state.

the odd thing is, when i was running vista ultimate 64bit, the setting i had then doesnt work now that im running windows 7 ultimate 64bit.

i'll try increasing the voltage at the lowest and see if that helps.

will let you know if things works out

Vicotnik · Post by **Vicotnik** » Sat Dec 05, 2009 2:35 pm

RMClock has a few timing settings that might help instability issues when changing states. I fiddled around a bit with those on my system but got tired of random freezes and went back to plain old CnQ.
Very hard to track down an intermittent problem that only shows it's ugly face once every other week or so..

bbzidane · Post by **bbzidane** » Sat Dec 05, 2009 3:57 pm

update

i ended up setting the multiplier and voltage in the bios
10.5x @ 0.95V
this was helpful as it stopped the computer from drawing ~90W from my ac/dc adapter that caused the fan to run

in windows, i increased the voltage to 0.85V for the lower multiplier and the system seems to be doing fine so.

fingered crossed, hopefully that is the end of my random crashes/hangs.

thanks for all the feedback/comments/advices.

how do you test if a system is stable when undervolting

how do you test if a system is stable when undervolting

Re: how do you test if a system is stable when undervolting

Re: how do you test if a system is stable when undervolting

Re: how do you test if a system is stable when undervolting

Re: how do you test if a system is stable when undervolting