SuperPI as a stability test?

Z

Zinn

Guest
Under more relaxed voltages, my computer is OCCT and Orthos stable... but when running SuperPI to 32million digits, I get rounding or square root errors. That kind of surprised me... I thought OCCT was the best way to detect problems. Increasing the NB and CPU voltages just a tiny bit and the problem went away.

To get a true measure of stability, do you really have to use all these programs? Or is there really no "true" benchmark for stability?

:confused:
 
I run Orthos/occt and see if it errors out within an hour. Then I let it run for 24 just to see what happens. Truthfully, if it doesn't crash or have crackly sound or somthing. I let it run.
 
Depends which SuperPi/IOrthos test you ran since higher values use some of your RAM. Small FFTs in orthos stresses only the CPU while large FFTs stresses some RAM. Similarly with SuperPi.If you ran the 1M SuperPI and failed try relaxing the memory timings and bench again at the same CPU oc.

Edit: Run Orthos at priority 8/9 small FFTs for 10 straight hours and then 8-10 more for large FFTs to check for stability.
 
Or is there really no "true" benchmark for stability?
Click to expand...

I imagine Intel has a program that exercises all the functionality of the chip. I bet the military and other industrial/commercial entities have a "verification" suite they run to verify a machine is ready for a high reliability environment. Otherwise the ansswer is no, there is not any one program or even suite of programs offered to the community that has been accepted, at least in my mind, as "run this and if no errors you will never see a BSOD or freeze". "Stability" is ultimately in the mind of the user. Do/use what makes you happy/satisfied. So its your call.

Frankly Windows is so basically unstable and unforgiving in recovering from most errors, such a program would be a farce anyway.



If my machine will run all the typical stress tests (both thermal and operational) and my game and typical applications for as long as I want without error, I consider it stable. If what happened to you happened to me, I would leave the NB voltage raised a tad if that caused no other issues and the NB/MCH was still below Vcc max and no heat issues.
 
my last q6600 would run 2 x 32m superpi up to 3950mhz. orthos would fail after a few seconds at any speed above 3000. then again, occt would run fine up to 3700. so which is my stable 24/7 speed? i could run games for as long as i wanted and do anything else and never crash or blue screen at 3700 ish. for me, thats my
stable speed. i wouldnt take failure at any single program to mean you are not stable. but if you are failing more than 1, or failing 1 and getting blue screns, service errors, machine checks, etc, then you are not stable. also, stability is relative. if you are running telemetry for nasa, then i guess 1 week 4 instances prime would be ok. for just games and home stuff, a couple hours should be fine.
 
if you can go 24+hours of prime 8+hours of memtest 32, and loop 3dmark to infinity, I'd say you'd have a fairly stable system.

An unstable system can manifest itself in many ways, from BSOD to crashing programs to skipping music on your ipod when transfering files to bad movie encodes...

you may be able to play games with no issue, but not all games stress every part of the CPU.
 
Slade said:
and loop 3dmark to infinity,
Click to expand...

Yeah, but then you have to buy another system to actually use, while you are infinitely looping 3dmark on the first. Oops, you'd also need to stability test that one, so you'd actually need a third system, clocked at default, to actually use. ;)
 
that's strange. i have been able to run superpi on some very high unstable OC's. OC's that would BSOD with in seconds on starting orthos but, super PI would be fine so dont think superpi is a good stability test in my experience.
 
vanilla_guerilla
my last q6600 would run 2 x 32m superpi up to 3950mhz. orthos would fail after a few seconds at any speed above 3000. then again, occt would run fine up to 3700. so which is my stable 24/7 speed? i could run games for as long as i wanted and do anything else and never crash or blue screen at 3700 ish. for me, thats my
stable speed.
Click to expand...

IF Prime95 fails then no matter what way you look at it there is a stability issue with your system. It may never crop up during the lifetime of your normal system usage but it exists nonetheless.

I cant speak for occt but i can run SuperPi on my system @ 3.55GHz unstable. I get 16secs compared to 17.5secs @ 3.3GHz, but at the end of the day it isnt stable.

gesicht
that's strange. i have been able to run superpi on some very high unstable OC's. OC's that would BSOD with in seconds on starting orthos but, super PI would be fine so dont think superpi is a good stability test in my experience.
Click to expand...

Spot on.. read my previous stmt.
 
commodore said:
IF Prime95 fails then no matter what way you look at it there is a stability issue with your system. It may never crop up during the lifetime of your normal system usage but it exists nonetheless.
Click to expand...
my sentiments exactly. if the system is not superpi, orthos, prime OR occt stable at a given speed, it's not stable period. failing any one test is the same as failing all of them.

i suspect what you were saying about superpi stressing more ram at higher values was correct, as the system could crunch through 1M digits easily and reliably at the lower voltage, but would quickly fail trying for 32M.

i ended up turning the FSB speed down by 3mhz because I just couldn't stabilize it and keep my 4-4-4-12 timings. i know there's some headroom in there, so i'll probably tweak around some more once i have a bit more time.
 
mhz > timings

I'd sacrifice 4-4-4-12 timings to run 3.6 as opposed to 3.2 to run them...
 
You must log in or register to reply here.
Back
Top