Prime95 doesn't always mean stability

Just thought I'd post a comment that may have been posted before, but I have always used prime95x2 to test ultimate CPU stability on my last few P4 builds.

My P4C800E-D board has a 2.4 @ 3.4 that was (is) prime95x2 stable for 48hrs.

I began using "autoGK" (gordian knot) to compress MPEG2 video into MPEG4 (divx) and began getting lockups or reboots during the long runs.

To make a long story short, no change I made anywhere (mem speed, timings, dividers - memtest x86 good overnight) ,except lowering the bus speed from 283 down to 270, stopped the freezes / reboots.

So, my 2.4c / P4C800 machine has been running for over a year with no settings changed, and never a freeze or reeboot or anything, including running dual instances of prime95, until using this program.

I back off the speed and it will run GK all day long.

So, while I realize Prime95 may not predict stability for OCed graphics and the like for all you gamers, I had always thought it was pretty bulletproof for CPU/Memory subsystem testing.

I think P95x2 for 24hrs and at least 10 hrs of GK may be a better test for real stock-like stability.

- Chris

What is Gordian Knot, and is it free? I havent heard of it before.

Well I am all for the idea that it does not mean your system is stable. I have got it to run perfectly for 48 hours on 2 of my systems that are unstable otherwise. Its not 100% trustworthy. I only use it to heat up my cpu.

hibner said:

What is Gordian Knot, and is it free? I havent heard of it before.

Click to expand...

is program used to encode dvd-s to divx/xvid....and its free if im not mistaken..

No, Prime95 is NOT perfect, but is a pretty good indicator of stability, imo.

It also has the feature that you're actually doing calculations & results checking, which most CPU "torture" proggys DO NOT do.

Nothings perfect, but, imo, it's a pretty usefull stability tool, & I use it.

BTW, I've only found one proggy that will crash an HT'd Intel machine that's dual Prime stable, and that's the distributed computing program CPDN (Climate Prediction), FWIW.

As you stated, there's probably other proggys that'll crash a dual Prime stable machine, too, but not many that I've used (btw -I don't game).

Strat

<EDIT>

After re-reading your post:

1] Prime95 does only LIMITED memory testing - Memtest is a much better proggy for testing your sys mem (which you stated you did also).

2] To be dual Prime stable for 48hrs @283 FSB, and then require a drop to 270 FSB, is an extremely drastic drop, IMO. Something doesn't sound quite right...I could possibly see dropping 3 - 4 FSB; But requiring a 12 FSB drop, from 48hr dual Prime stable, seems pretty funky, IME.

But hey, you seem to have done the right troubleshooting techniques, so I won't debate the issue, but I've personally never seen anything requiring this drastic of a reduction on an overnite dual Prime stable machine, much less a 48hr one.

(Hmmm...maybe I should try running the proggy that gave you the probs)

Anyway, GL

Prime95 tests one part of the computer, but it's not a "one go" stability test that would tell you everything. There are also differant options in prime to set it to test the processor, memory or both. I like Prime95 alot, but it's only one part of a full test. I use memtest86 just as much as prime95.

-> test memory with memtest86
-> test processor with prime95
-> test with prime95 blend

That's my process. If I was into gaming and overclocking my videocard I would also run some testing (loop?) with a heavy game that really stresses the videocard and uses as much differant shaders etc as possible.

For me, prime almost always fails both instances at once. Last night I ran one at 8k (my L1 level) and another at 512k (my L2 level). Both failed at the same time about 3 hours into the night. Thats always seemed the flaky bit to me. Does the chip just hiccup every now and then?

i am using winavi encoder to test my cpu and it aseems ok

celeron 2.66 @ 3.7ghz v-core 1.475

zachysat

Well, I always use P95x2 with one instance running "small" fft and the other "large".

Memtest I always use before I even start messing with OS level stuff.

I also don't game.

The drop from 283 to 270 is not an absolute one - It may be OK at 273 for example, but I DID try it at 275 and it still had trouble with autoGK.
I set this machine up over a year ago, and didn't really want to take the time now to go in 1mhz increments, as I now have faster Procs to fiddle with.

BTW, load temps (H20) are ~35c after several hours.

System is fine every which way other than autoGK, and stepping down the FSB to that point is all that will make it stable.

Thanks for the feedback.

- Chris

Well... Prime95 is good, but it stinks for memory issues. Super pi should be used in conjunction with it.

Still, these tests will not iron out all quirks. When overclocking, there are often unforseen effects on the rest of the system... sometimes you have errors on the IDE or SATA controllers, the PCI-E bus can have problems (not locked), etc. You can't really test these with one program.

What Prime stress test do you have ran ?
13MHz FSb translate in nearly 150MHz CPU an helfy back-off.
283 is a pretty big overclock for the chipset, autoGK like all software encoding stress pretty hard cpu-chipset-ram.
IMO is the MOBO, pushed too far.
Prime in Small or Large FFT is pretty cpu centric stress test, Blend mode is way more all-around.
Loop 3DMark01+ prime95 small (affinity on CPU 1) for 4-5 hours is a pretty stressfull test all-around should crash if autoGK isn't stable at the same settings

mazzy said:

What Prime stress test do you have ran ?
13MHz FSb translate in nearly 150MHz CPU an helfy back-off.
283 is a pretty big overclock for the chipset, autoGK like all software encoding stress pretty hard cpu-chipset-ram.
IMO is the MOBO, pushed too far.
Prime in Small or Large FFT is pretty cpu centric stress test, Blend mode is way more all-around.
Loop 3DMark01+ prime95 small (affinity on CPU 1) for 4-5 hours is a pretty stressfull test all-around should crash if autoGK isn't stable at the same settings

Click to expand...

large is only cpu based if you have enough l2 for that to fit.

i've never ran Memtest, you can normally tell if its your CPU or RAM which is unstable when you play a game plus the errors you get within windows. I'll run Prime95 for 24hours, if thats ok, then i'll use the PC as I would normally, this being gaming and video work. If something craps out then i'll do some fine tuning as it were. But for me 24/48hours prime hasn't failed me yet.

people dont JUST dual prime anymore. they stress it out with dual prime and running something else also. for example i do prime and superpi 32m. then loop super pi

Jimbob7 said:

i've never ran Memtest, you can normally tell if its your CPU or RAM which is unstable when you play a game plus the errors you get within windows. I'll run Prime95 for 24hours, if thats ok, then i'll use the PC as I would normally, this being gaming and video work. If something craps out then i'll do some fine tuning as it were. But for me 24/48hours prime hasn't failed me yet.

Click to expand...

Wow! I'd never run windows for even a moment if I fail memtest...

Yeah, me either...

Anyway, I've always considered systems I've put together (at least lately) bulletproof if they could prime2x for 48hrs or so (Video subsytems excepted, and obviously Memtest approved beforehand...).

Thats the whole reason I posted this - I've not run into this before.

Remember, this is a system that has been on 24/7 for about 14 months, used everyday, and has never BSODd, frozen or rebooted for any reason except buggy SW, which I always "get to the bottom of", to make sure its not hardware related.

Weird that Auto GK does SOMETHING that P95 and others don't seem to do.

I would love to know what is causing the freezes/reboots, thats making me go from 283 down to ~270.

Not that it really makes a diff, but its the figuring it out part I'd like to do, as its odd.

- Chris

Just to be sure, although I'm not sure if it would cause this problem to arise, are you running Prime95 at the highest priority setting? I know the first time I ran Prime I didn't know that was an option, and since it was password protected, I assumed it didn't make a difference. Just a thought.

I think RAM instability is probably the most over looked aspect of stability testing.

After I was satisfied with memtest for a couple days, I ran Prime with my cache turned off for two and half days to make doubly sure my RAM was stable.

Well, I run P95x2 without any other apps running.

And in my experience, memory issus seldom cause re-boots, (but can cause the occasional freeze).

I've also dropped the divider to 3:2 just to take the memory issue out of it completely.

And I think if it was somehow memory related, I would have seen issues in the last year plus of up-time. This machine is used all the time.

Good suggestions, I think, though.

- Chris

Keep in mind that no single stability tester is perfect... different stress testers test in different ways, and different types of process can cause instability to exhibit itself where other types of process fail.

Ultimate, unbreakeable, bulletproof stability is in a sense, unachieveable, because a potential for error always exists (it is impossible to test for eternity).

I used to be a die-hard 'Prime95 activist'... recently I have changed my opinion of the program, and of stability testing in general. I have a memory overclock that is Prime, memtest, 3Dmark, and game stable... but it fails SuperPi.


Prime95 is an excellent stability tester, but it isn't perfect. No stability tester is perfect.