Need help interpreting Prime95 Error

[v3rax]

I havebeen running Prime 95 for about 5 hours on my PII 940. When I looked at it a few hours ago, I see this error on Worker#1 IMAGE . When I came back to look at, this is what I saw.

Does this mean I have a bad Phenom?

 

[Idontcare]

It definitely says the calculation did not complete as intended, so yes something is not right.

 

[aigomorla]

the point of a stress test is to stress the computer.

Not let it sit in the background as you browse webpages.

Sometimes by doing that you will run into problems.

Also are you overclocked?

 

[v3rax]

I wasn't doing anything on the computer while the test was being run. I left my house actually. The image I showed initially, YES, it was overclocked.

However, I reset my BIOS to stock, no OC at all and I ran the test again. I then walked away. I came back 7 minutes later and it had already failed in the same place again.

Shoudl I assume that one of the cores is bad?

 

[aigomorla]

Originally posted by: v3rax
I wasn't doing anything on the computer while the test was being run. I left my house actually. The image I showed initially, YES, it was overclocked.

However, I reset my BIOS to stock, no OC at all and I ran the test again. I then walked away. I came back 7 minutes later and it had already failed in the same place again.

Shoudl I assume that one of the cores is bad?

set your bios at factory recomended default.

Retest.

Core 1 fail is usually a sign of lack of voltage on the cpu. Or it could mean that you were loading multipul things and 1 of the cores had bad work.

 

[LOUISSSSS]

increase your voltage and try again. set your cpu speed to default and ram timings to default.
make sure everything has enuogh voltage.

aigo, i'm not sure browsing the internet affects stress testing. worse comes to worse, it'll just take longer time to complete the same amount of stress testing because firefox (I.E.) is stealing cpu cycles from (insert stress test here)
doing other tasks should error out your stess tests.. unless someone can convince me otherwise

 

[Idontcare]

Originally posted by: LOUISSSSS
aigo, i'm not sure browsing the internet affects stress testing. worse comes to worse, it'll just take longer time to complete the same amount of stress testing because firefox (I.E.) is stealing cpu cycles from (insert stress test here)
doing other tasks should error out your stess tests.. unless someone can convince me otherwise

Yeah it should not cause the stress test to suddenly be unstable. In fact as you point out, your browser app is stealing cycles away from the stress test...and presumably your browser is doing less power-consuming computations with those cycles than a stress test would otherwise be doing with them, so if anything you are making your system appear to be a bit more stable as it is being a tad less stressed in reality relative to the stress you thought your stress tester was putting the system thru.

 

[VirtualLarry]

I've had prime95 fail quicker while web-browsing in the background, versus just letting prime run. I attribute it to increased and varying loads on the system chipset, while loading the CPU to the max.

Regardless, if prime is erroring, then your system is unstable, something needs to be fixed.

Tried a Memtest86+ bootable CD image lately?

 

[BTRY B 529th FA BN]

If my overclock is unstable usually the sound drivers will cause a system freeze. Yes, I jam music while running prime95 WOOT!!!

 

[alkalinetaupehat]

I've ran P95 for two days straight before with no errors, but as soon as I ran a Lost Planet benchie or played a match in L4D, I BSOD'd.
Interpret that error lol. If I bump up vcore by a notch (like .00625v) it becomes stable through P95, linpack, L4D, w/e.
I haven't read too much about O'Cing with IMC procs, but the standard stuff mostly still applies, there's good advice in this thread.

 

[Idontcare]

Originally posted by: alkalinetaupehat
I've ran P95 for two days straight before with no errors, but as soon as I ran a Lost Planet benchie or played a match in L4D, I BSOD'd.
Interpret that error lol. If I bump up vcore by a notch (like .00625v) it becomes stable through P95, linpack, L4D, w/e.
I haven't read too much about O'Cing with IMC procs, but the standard stuff mostly still applies, there's good advice in this thread.

P95 does not test every instruction of the instruction set that your CPU is supposed to be capable of executing. P95 tests a subset of the total number of instructions.

The instructions P95 tests, and the thermal conditions the test itself generates, is a good enough proxy for determining whether the circuits signal/noise ratios are acceptably good enough to operate at the frequency and voltage you have specified.

But it cannot cover all instructions. F@H does some stuff that craters an otherwise p95 stable rig. I personally use a software package called Gaussian (quantum chemistry modeling) that craters an otherwise P95 stable rig.

P95 stable does not mean every feature of your CPU is functioning properly, just means that for a healthy number of instructions it does work properly.

P95 stability testing is merely meant to accelerate your personal time involvement in iterating your rig to a point of stability. Rather than waiting weeks to uncover errors in your programs of interest you can use p95 to uncover them in hours. But it won't uncover all of them.

P95 stable does not mean 100% stable, but being p95 unstable certainly means your rig is not stable.

 

[Kraeoss]

and what about occt ? does this test alot more than prime ? if so i guess my cpu is stable then 12hrs so far on OCCT 3.0 occt test largest cpu @ 3.015GHz and 1.5 vcore in occt 1.55 in bios

 

[alkalinetaupehat]

Originally posted by: Idontcare

Originally posted by: alkalinetaupehat
I've ran P95 for two days straight before with no errors, but as soon as I ran a Lost Planet benchie or played a match in L4D, I BSOD'd.
Interpret that error lol. If I bump up vcore by a notch (like .00625v) it becomes stable through P95, linpack, L4D, w/e.
I haven't read too much about O'Cing with IMC procs, but the standard stuff mostly still applies, there's good advice in this thread.

P95 does not test every instruction of the instruction set that your CPU is supposed to be capable of executing. P95 tests a subset of the total number of instructions.

The instructions P95 tests, and the thermal conditions the test itself generates, is a good enough proxy for determining whether the circuits signal/noise ratios are acceptably good enough to operate at the frequency and voltage you have specified.

But it cannot cover all instructions. F@H does some stuff that craters an otherwise p95 stable rig. I personally use a software package called Gaussian (quantum chemistry modeling) that craters an otherwise P95 stable rig.

P95 stable does not mean every feature of your CPU is functioning properly, just means that for a healthy number of instructions it does work properly.

P95 stability testing is merely meant to accelerate your personal time involvement in iterating your rig to a point of stability. Rather than waiting weeks to uncover errors in your programs of interest you can use p95 to uncover them in hours. But it won't uncover all of them.

P95 stable does not mean 100% stable, but being p95 unstable certainly means your rig is not stable.

My point was that a single program cannot determine absolute stability currently, though thank you for mentioning Gaussian, could you provide me a link to it? I'm interested in seeing if my OC is stable with that program.
Going off the thought of "The One Program To Rule Them All", would it be theoretically possible to create a program capable of testing every single aspect of a processor? Even if it was for a specific proc, such as an E8400?

 

[Rubycon]

Originally posted by: alkalinetaupehat

Originally posted by: Idontcare

Originally posted by: alkalinetaupehat
I've ran P95 for two days straight before with no errors, but as soon as I ran a Lost Planet benchie or played a match in L4D, I BSOD'd.
Interpret that error lol. If I bump up vcore by a notch (like .00625v) it becomes stable through P95, linpack, L4D, w/e.
I haven't read too much about O'Cing with IMC procs, but the standard stuff mostly still applies, there's good advice in this thread.

P95 does not test every instruction of the instruction set that your CPU is supposed to be capable of executing. P95 tests a subset of the total number of instructions.

The instructions P95 tests, and the thermal conditions the test itself generates, is a good enough proxy for determining whether the circuits signal/noise ratios are acceptably good enough to operate at the frequency and voltage you have specified.

But it cannot cover all instructions. F@H does some stuff that craters an otherwise p95 stable rig. I personally use a software package called Gaussian (quantum chemistry modeling) that craters an otherwise P95 stable rig.

P95 stable does not mean every feature of your CPU is functioning properly, just means that for a healthy number of instructions it does work properly.

P95 stability testing is merely meant to accelerate your personal time involvement in iterating your rig to a point of stability. Rather than waiting weeks to uncover errors in your programs of interest you can use p95 to uncover them in hours. But it won't uncover all of them.

P95 stable does not mean 100% stable, but being p95 unstable certainly means your rig is not stable.

My point was that a single program cannot determine absolute stability currently, though thank you for mentioning Gaussian, could you provide me a link to it? I'm interested in seeing if my OC is stable with that program.
Going off the thought of "The One Program To Rule Them All", would it be theoretically possible to create a program capable of testing every single aspect of a processor? Even if it was for a specific proc, such as an E8400?

I wish there was a way. If someone could make a CPU tester like a tube tester (for those of you old enough to remember them hehe) where you plug it in, strap on a cooler and push test and within a minute you are told the max speed it will run at a given voltage and temperature without producing errata - period!

I had a QX9650 that got up to 4.2GHz Prime95 stable for days. Thought it was fine. Then I started encoding videos. The encoding crashed at 4.2. Lowered to 4.0. Saw artifacts in the video output. Lowered to 3.8 (all at the same voltage) saw an artifact here and there. Lowered to 3.6 and it was perfect. Also I do surf and run Prime95 and never had it crash like mentioned before. If it can do everything stock it sure as heck better do it o/c or the o/c is useless to me.

 

[Thor86]

Simple testing with Prime95 for CPU/IMC/Motherboard/Memory stability testing:

1) LargeFFT testing for max heat/cpu stability testing (all cores).

2) Custom testing (depending on how much RAM you system has), use two instances of Prime95 if you are using a quad-core or more, and select 2 cores for each Prime95 instance, change the Torture Test setting to use FFTs within range 2048 min to 4096 max, and amount of memory just under 50% for each instance of Prime95.

If any of these tests error, then something is not setup properly or OC settings are not stable.

See more details on Prime95 usage here:
http://en.wikipedia.org/wiki/Prime95

-edit typo-

 

[Idontcare]

Originally posted by: alkalinetaupehat
My point was that a single program cannot determine absolute stability currently, though thank you for mentioning Gaussian, could you provide me a link to it? I'm interested in seeing if my OC is stable with that program.
Going off the thought of "The One Program To Rule Them All", would it be theoretically possible to create a program capable of testing every single aspect of a processor? Even if it was for a specific proc, such as an E8400?

Ah, sorry if it seemed like I didn't get your point, I was posting as a way to support that very point.

http://www.gaussian.com/

Gaussian is a software package whose purpose is to use various computational chemistry techniques (including ab initio) to compute properties of molecular structures. It's not a bench program, and there is no freeware download version of it like prime95 or F@H.

It is kinda like F@H in that it is computational chemistry, but I use Gaussian to directly do research (in my case developing light-harvesting molecules for the pursuit of photochemistry in splitting water into hydrogen/oxygen molecules) whereas most folks use F@H to be cpu cycle contributors. In my case my input files which I personally create for gaussian can cause system instability on an otherwise prime95 setup, so for me p95 is but one iteration of my stability test ensemble.

As for a "one program to rule them all"...of course it exists. They exist in droves, on tester after tester at Intel where they bin the CPU's for a given clockspeed/Vcc/TDP in the first place. And yes they are CPU specific, they have to be.

Now unfortunately the programs are proprietary, cost a veritable shitload of time and money to develop and require testbeds (million dollar pieces of equipment) to execute the code and monitor/capture the results...so no there is no freely available test program that does what Intel and AMD spend tens of millions of dollars to accomplish.

 

[BTRY B 529th FA BN]

Are you sure? I thought I could find it at the local convenient

 

[alkalinetaupehat]

Originally posted by: Idontcare
Ah, sorry if it seemed like I didn't get your point, I was posting as a way to support that very point.

http://www.gaussian.com/

Gaussian is a software package whose purpose is to use various computational chemistry techniques (including ab initio) to compute properties of molecular structures. It's not a bench program, and there is no freeware download version of it like prime95 or F@H.

It is kinda like F@H in that it is computational chemistry, but I use Gaussian to directly do research (in my case developing light-harvesting molecules for the pursuit of photochemistry in splitting water into hydrogen/oxygen molecules) whereas most folks use F@H to be cpu cycle contributors. In my case my input files which I personally create for gaussian can cause system instability on an otherwise prime95 setup, so for me p95 is but one iteration of my stability test ensemble.

As for a "one program to rule them all"...of course it exists. They exist in droves, on tester after tester at Intel where they bin the CPU's for a given clockspeed/Vcc/TDP in the first place. And yes they are CPU specific, they have to be.

Now unfortunately the programs are proprietary, cost a veritable shitload of time and money to develop and require testbeds (million dollar pieces of equipment) to execute the code and monitor/capture the results...so no there is no freely available test program that does what Intel and AMD spend tens of millions of dollars to accomplish.

Gaussian does sound pretty interesting, though it is pretty cost-prohibitive for personal use, assuming they would even sell an individual a license. You would also think that given the apparent awesomeness of the program, they could perhaps spruce up the site.

You know, I find it rather compelling to think of Intel providing a "super-binning" of chips, above their server and E.E. parts. Say they were to bin CPUs for a 3-5 year lifespan instead of the 10+ they currently do. I would reason to say a niche market similar to the one for i7 965's and the like would also exist for these processors capable of higher performance out-of-the box, but shorter lifespans that could potentially have better OC headroom as well.

Just a thought, though admittedly the binning of such processors could become too costly for the niche of potential consumers.

 

[Idontcare]

Originally posted by: alkalinetaupehat
Gaussian does sound pretty interesting, though it is pretty cost-prohibitive for personal use, assuming they would even sell an individual a license. You would also think that given the apparent awesomeness of the program, they could perhaps spruce up the site.

They do sell individual licenses. I was speaking literally when I said I was doing the research. I bought a license ($1k at the time) and I do the work at home on my personal computers. At TI we bought a commercial site license as well, that was $30k at the time.

Ha, I feel you on the website comment. It's one of those things though where the clientele for this product are geeky scientists types, and if we saw that $$$ had been spent on fancy flash-based website bling instead of hiring another coder to implement the latest density functional or basis set then we'd be pissed. :laugh:

Originally posted by: alkalinetaupehat
You know, I find it rather compelling to think of Intel providing a "super-binning" of chips, above their server and E.E. parts. Say they were to bin CPUs for a 3-5 year lifespan instead of the 10+ they currently do. I would reason to say a niche market similar to the one for i7 965's and the like would also exist for these processors capable of higher performance out-of-the box, but shorter lifespans that could potentially have better OC headroom as well.

Just a thought, though admittedly the binning of such processors could become too costly for the niche of potential consumers.

Lifetime binning is a touchy subject in this industry. Basically no one wants ever be in the position where there chips are dieing in the field, regardless of the reason, in any significantly large numbers perchance it becomes a national media spectacle. There is no price-tag that can be assigned a tarnished corporate reputation.

What you describe is a perfectly sensible and logical business plan, any chip maker could implement this strategy from ram to cpus to gpus to hard-drives, etc. But it isn't done because the negative PR aspects would be hyped by the marketing depts of the competition.

As soon as AMD marketing could find 10 or 20 people whose super-binned i7's died in 24 months (even if replaced under warranty) they'd be off creating slides and making marketing spin (as they are paid to do) to undermine Intel's image across *all* business segments to the best of their imaginations. Intel would do the same to AMD if given the chance, AMD to NV, Seagate to Hitachi, etc etc.

Basically its not done for a reason (super-binning that is) because its an invitation for a PR nightmare, real or contrived.

 

[AdamK47]

Originally posted by: aigomorla
the point of a stress test is to stress the computer.

Not let it sit in the background as you browse webpages.

Sometimes by doing that you will run into problems.

Also are you overclocked?

Are you saying the error is the fault of the user rather than the computer?

 

[aigomorla]

Originally posted by: AdamK47

Originally posted by: aigomorla
the point of a stress test is to stress the computer.

Not let it sit in the background as you browse webpages.

Sometimes by doing that you will run into problems.

Also are you overclocked?

Are you saying the error is the fault of the user rather than the computer?

im saying its possible.

But he said it was left alone, so its either something bad with his hardware, or bios needs to be updated.

 

[Rubycon]

Originally posted by: aigomorla

Originally posted by: AdamK47

Originally posted by: aigomorla
the point of a stress test is to stress the computer.

Not let it sit in the background as you browse webpages.

Sometimes by doing that you will run into problems.

Also are you overclocked?

Are you saying the error is the fault of the user rather than the computer?

im saying its possible.

But he said it was left alone, so its either something bad with his hardware, or bios needs to be updated.

Nothing the user can do (aside from pouring water in the computer or hitting the motherboard with a hammer!) e.g. NORMAL computing use - should ever cause an error traceable to system stability. Modern CPU's have LOTS of margin. Overclockers choose to run their systems with much less margin.

 

[Idontcare]

Originally posted by: Rubycon
Nothing the user can do (aside from pouring water in the computer or hitting the motherboard with a hammer!) e.g. NORMAL computing use - should ever cause an error traceable to system stability. Modern CPU's have LOTS of margin. Overclockers choose to run their systems with much less margin.

Funny story, I share this for your laughter at my past misery (not to appear as an annoying "yeah but..." reply to your post above) I use to use a USB flash drive a lot and it just so happened my floor had carpeting, I wore socks, low humidity, and I loved sweaters. You can see where this is going.

I use to invariably shock the crap out of my computer via the USB connection (you'd think it would be all grounded thru the case and out the PSU's ground line) and when this would happen the computer would go nuts and crap out a BSOD.

This was one of those things I would do maybe once every two weeks for months (winter months) because I'd eventually forget to discharge myself on the metal desk leg before putting in or pulling out the thumb drive at least once every other week.

Total user idiocy on my part, but had I known less about computers and ESD I'd probably have convinced myself it was the computer that was shocking me and that it was my rig that was unstable.