Prime95

[Timdog]

I was testing my new OC with Prime95. I left the torture test run overnight. It failed in 7 hours, so I increased the voltage from 1.7 to 1.725. I left it run overnight again but it failed in 6 hours. The only difference was that I had MBM 5 running then to check my temperatures. Should I be satisfied with this stability or should I increase the voltage to 1.75?

My temps were 38C avg and 41C high.

[athayer187]

Well, if you want perfect stability, I'd try a little higher voltage or lower clock. Your other option is to test it as is with all the games or everyday usage you do. If it's stable for a week of hard regular usage, that's good enough for me. Sometimes prime detects small errors that would not cause a crash even under the heaviest gaming.

[Timdog]

I have been playing CS, listening to music, etc on it everyday. I think I'll raise the voltage to 1.75 and see if that makes any difference tonight.

[Jimbob7]

6 hours prime stable should be fine for games. But if u can get away with using more vcore and you are more stable do that. :)

[GlitchOfDoom]

if your folding try and get more stability, if your just playing games you should be fine where its at. Or if you have temperature headroom increse the voltage to 1.75

[ogboot]

the way i've always seen it, if prime produces an error then literally your machine is producing incorrect results. not something you want it to be doing, and if it is producing errors then it is especially detrimental to your chip.

EDIT: Failing gates are not cool.

[Filter]

to this day i still think prim95 is a flaky program.

[TheTick]

its not really flaky, it just does a damn good job of testing for stability. 100% stable, not just 95-97% which would allow us to play our games and what not.

[Oni]

I don't consider an overclock stable until I can run Prime for ~48 hours w/o an error. It's a pretty good marker to go by.

[Timdog]

Increased to 1.75 and it's stable now. Temps about 39C load. :D

[StaticXD00d]

Originally posted by Filter
to this day i still think prim95 is a flaky program.

Why do you think that? I've been running Prime for years on many machines, and never ever seen anything "flaky". The team I'm a member of (which was started by my friend) is currently in the top 10 for LL testing (#8) with a total of 1679 LL exponents tested in 2885.51 CPU years. That's a LOT of non-flaky priming.
:D

[Lancelot]

I always test (on HT systems) with both Prime95 and 1 SETI WU running. A system that flunks out of Prime95 even after a few hours is just not that stable. The thing is that Prime really is a b!tch as VERY often you'll find out you have to lower your OC considerably to get stability while you can do everything else without any problems. However I wouldn't have it any other way. I want my system to be able to run SETI for 24/7 and HATE it when I come home from work thinking she's stable cuz I gamed for hours the night before only to find her frozen up. Don't let all the nice screenies and high numbers your read about get to you. Come to your senses, lol! What OCing means to me is being able to enjoy a few hundred Mhz for free (don't matter if is 200 or 800) just as long as your system can ALWAYS handle EVERYTHING you throw at it.

[Jognt]

you threw prime at it, and it couldnt handle it ;)

[Lan_Of_Malkier]

I'm still not convinced about Prime. My machine would not pass when it was brand new, without o/c. Then it was off and on durring the various phases of moving up my o/c. Just not consistent. Add to that it is a Windows based proggy.

I fold on 3 machines 24/7 while doing other tasks and have never experienced freeze-ups. 100% cpu usage 24/7 is more than enough proof to me that my machines are rock-solid regardless of what prime says.

It's a useful program but don't put all your eggs in one basket. After all a fluke em field passing by or ion cluster can produce a flake in equations no matter how stable your system.

[Gnufsh]

Originally posted by Lan_Of_Malkier
I'm still not convinced about Prime. My machine would not pass when it was brand new, without o/c. Then it was off and on durring the various phases of moving up my o/c. Just not consistent. Add to that it is a Windows based proggy.

I fold on 3 machines 24/7 while doing other tasks and have never experienced freeze-ups. 100% cpu usage 24/7 is more than enough proof to me that my machines are rock-solid regardless of what prime says.

It's a useful program but don't put all your eggs in one basket. After all a fluke em field passing by or ion cluster can produce a flake in equations no matter how stable your system.
As for it being a windows program, there is also a linux version. Remember the primary purpose of prime95 is to find very large prime numbers (of the form 2^p-1, I believe). It happens to also be a good stability tester, and that function was enhanced by the author. I've not had prime95 give me errors that weren't hardware problems, buit then I ususally run it from linux, so it's less likely that the OS will give me problems. And yes, you should run other programs to test stability also. I use several including cpuburn and memtest86. But, it is my opinion that if prime95 produces an error repeatably (so it say, does it a coouple times in a row, reducing the chances that you picked up some random fluke), you should take it seriously and try to track down the cause. Often, a bump in vcore or a drop in OC makes the problem go away. If it does, then prime95 was correctly detecting a CPU pushed just a slight bit too far.

As far as 100% CPU useage is concerned, that doesn't tell you how hard the different functional uniits of the CPU are used. Maybe there is not much parallelism that can be extracted from the code by CPU, so not all the units are used. Maybe it contains enough of a mix of integer and FP code, that neither is unduley stressed. Prime95 tends to do a good job at highly stressing parts of the CPU, more so than most other programs. After all, most modern OSes technically always keep CPU useage at 100% by sending the HLT instruction when the CPU is idle (reducing power useage).

[Albuquerque]

Actually that isn't quite right... Windows XP only sends a HLT instruction to a processor if:

A. The processor driver allows it (any Intel after the P2, any AMD after like the K6, etc)
B. The power scheme is set to "Presentation" or "Max Battery"
C. The system Processor Dynamic Throttle Policy is currently in a "Degrade" state.

Your system must hit all three of those checkpoints before Windows XP will issue a Processor Halt instruction. The only other time it will issue a PROC_HLT instruction is in response to thermal values reported by the ACPI subsystem.

Quote directly from MS's Windows Native Processor Perforamnce Control document:
Degrade
The Degrade policy always runs the processor in the lowest available voltage/frequency (performance) state. Additionally, Degrade will utilize linear stop clock throttling under the following conditions:
• When the battery remaining capacity drops below a certain threshold (configurable in the registry), AND
• The system is not using the C3 idle state effectively; that is, the system is too busy to spend a certain length of time (configurable in the registry) in the C3 state.
The Degrade policy will never throttle below a minimum level, also configurable in the registry. See the following section for details.


Also :)
The current version of this paper is available on the web at http://www.microsoft.com/hwdev/tech/onnow/ProcPerfCtrl.asp
Not entirely related to Prime95 stability, but answers the question on XP halt instructions to the processor.

[Crash893]

Originally posted by Filter
to this day i still think prim95 is a flaky program.


its pickey but its definely not flakey

ive run 150.822 years of prime work for my team and ive never seen it do anything flakey

[Gnufsh]

I skimmed the doccument, and it seems mostly related to P-states and c-states. I certaintly wasn't talking about changing acpi states (at least I son't think so). I was referring to what programs like cpucool did on windows 9x computers and what linux does on all processors that support it. In fact, I searched the document you linked to and didn't find any mention of HLT. Although, maybe I'm way off base here.

[bottoneatrpi]

umm...back to prime....how can it be a flakey program if it works on stable computers?

All it does is perform mathematical calculations that it knows should result in numbers with certain properties (am I wrong?)


So....if your system is giving you answers to calculations that aren't right, it isn't prime95's fault, its the computer.


Every computer I have ever had (and I've had lots) have all been prime stable for at least 12 hours, maybe I should test longer but thats enough for me to sleep at night.

[superamd]

I have never had a computer prime for more that 7 hours my mobile 2400 with lax memory setting at default speeds (8x multi X 133) @1.55v did not prime for more that 7 hours it will prime for 7 hours @ 210X12 @ 2v on air with 57c load. go figure

I HATE PRIME95 'CAUSE IT TELLS THE TRUTH

[Steven4563]

prime95 is a great program to make sure a system is stable

[TheTick]

superamd, it could be the ram. My old crap generic ram wouldnt run a lot under its rating.

[cV]

Prime95 is a great program,b/c it'll actually tell you if your CPU/memory subsystem has problems.

For pure CPU stability testing, I still use CPUBurn on high. Never seen anything more taxing than that app.