Random shutdowns when undervolting

Ok, I undervolted my F3Ka awhile back, just a notch or two. After a while, it shuts itself down when i open a program (usually) or randomly (not often).

I really think it should not shut itself down because the uv'ing i did, it was pretty minimal after all, and i ran orthos for a good while too.

So basically around any undervolting i do, causes system to shutdown itself without any warnings/bsods. If somebody had desktop with PSU issues, same kind of shutdown.

If anyone's kind enough to spare a few thoughts for solving this, i thank thee.

Oh, and it's not hardware issue. Runs fine without uv.

 

what program are you using? you might want to try an alternative (NHC or rmclock depending on what you are using now). e.g. for me, nhc was very unstable, but i have not had an issue with rmclock.

 

Going to test NHC then.

 

The FIRST thing I would say, is if it dies when undervolted, don't undervolt. Some CPUs take some undervolting, others don't. If your's doesn't then don't undervolt.

So if I have the cause right, it is hardware.

 

Do you mean invidual CPUs or as in CPU types?

 

this is exactly why i don't like people suggesting undervolting without proper explanation as to exactly what's happening to the processor.

People seem to think that undervolting is as simple as "Turn it down and go", but in reality there is more to it than just that! The voltage supplied to the processor is the foundation for the processor on how each of the transistors react at that multiplier and voltage level. The lower the multiplier and frequency, the lower the voltage required. However, not all processors can go below specifications. So basically you need to test each of the voltage levels for that multiplier, and for a modern day mobile processor you are dealing with a multitude of 6-8 different multiplier/voltage settings.

So, how do you test? You would need to run a stability tester, basically it's something that takes complicated mathematical formulas and basically puts your cpu to the test, if it fails any of the calculations, it would be deemed unstable. Personally, I use Stress Prime 2004 othos edition which can test both cores on a dual core processor(SP 2004 orthos). An unstable processor can cause complications such as bluescreens, application crashes/hangs, file corruption, and random reboots, along with other bad things.

Well it may sound like I'm exaggerating and really just spitting verbal spam here but it's really a more complicated procedure than it may seem and it takes a little more patience, time and research to actually take advantage of undervolting.

So take this as a word of warning...

 

Just one addition.

Many times it is reasonably safe to skip the time-consuming procedure of testing each multiplier, and just test the smallest and largest multiplier. Then, you can interpolate voltage values for the rest. Actually, RMClock does that for you.

I must confess I am very lazy and I just undervolt .2 volts, and if it runs stably for a while I'm happy, if it starts bluescreening I'm taking it back up a bit.

I know it's not the "technically correct" procedure, but it saves a heck of a lot of time

By the way, deattan, I mean individual CPUs.

 

maybe he undervolted the CPU to 0volts!

 

I know, and I admit it is tedious and most people won't do this but I just run those tests overnight by choosing one multiplier and voltage and see the results, eventually you'll know the limits as to the voltage for each multiplier. I've been in situations where a processor doesn't fail till after hours on end and you wonder what that problem was; whether some application hanged, or your computer didn't hibernate properly, or some program refuses to run...and we're talking very minimal stuff that you think might be ordinary, like 1 out of 50 times. But in reality it's your undervolt that is the underlying problem where after 2 hours of Stress Prime on multiplier 10x it fails a calculation. Unfortunately, the art of undervolting is just as tedious as the art of overclocking and although most times a 0.2 volt(which is quite a bit for some multipliers) decrease doesn't crash windows it may cause errors in mathematical calculation. Any inaccuracy in a processor sacrifices stability.

 

Oh boy I am using my system to run mathematical simulations for my work, so if it has computing errors all my results are wrong.

Seriously though, I believe the probability of getting wrong mathematical results while the system is not BSODing is very, very low.

 

actually! it's very common, I used to do heavy overclocking on all my previous desktops(Amd 4400+x2 & Conroe e6400 the most recent) and I would have no problems booting into Windows and even eluded BSODs or even program crashes when I pushed the processor to the absolute limit(we're dealing with FSB and voltage increase), but it would consecutively fail the "complex" prime calculations after about an hour testing. Everything else works fine, so I think your wrong on this one E.B.E

 

Hmmm... ok maybe it was wishful thinking.

I will do the test to make sure my simulation results are correct.

 

Do it when you have time, if your system isn't having problems its safe to assume that it's "fairly stable" so take your time while your at it. It took me about 4 months(on my w3v) to actually get all the proper voltage numbers down by squeezing any processing time to do the tests(can't really do anything while it's doing the tests as it pushes it to 100% cpu usage) between all the work nights and game nights It's also an excellent way to determine if your cooling hardware needs maintenance(thermal compound) or cleaning.

I'm still trying to figure out all the my c90s's voltage numbers...

 

Just thought I'd update. I ran Orthos (the dual-core version of StressPrime) for two hours + , until Self-Test 32K or so (whatever that means) at max multiplier. I got 0 errors, so it seems that the -.1V undervolt I am using is stable on my CPU.

Furthermore, when the CPU will be under load it will always run at max multiplier (except the transient in the beginning when the load is initiated), so I'm quite confident that my CPU will not generate errors. I might ran some more tests on the intermediate multipliers this weekend, to make sure.

Correction, it seems I'm running only -.1V undervolt. This means sometime in the early days of this laptop, it probably bluescreened on my at -.2V

Edit: Stable -.2V. 3:40 hours of StressPrime Orthos w/ Stress CPU, 3:40 more hours with Blend CPU and RAM. I've also tested the (single) intermediate multiplier that I use in-between the minimum and the maximum on the Power on Demand profile, and the interpolated voltage value given by RMClock is stable.

So that's what I'm going to use. Thanks again, DTX.