Why is Prime95 failing?

[Slickone]

It always fails the small FFT's at #10. It sometimes passes the large FFT, but sometimes fails.

This is an old Compaq system, not overclocked. I've tried 3 different processors - a K6-2 400 (came with) and two K6-2 500's. I do change the jumper for correct speed after swapping CPUs. I've also tried each of the two 128MB DIMMs separately.
I use a thin layer of regular white thermal compound, just enough to make it white (as like Arctic Silver's website shows) spreading evenly with my finger covered with plastic wrap. I can all areas are making contact. It's a passive heatsink, but I've tried with a desk type fan blowing on it. I've bent the hold down bracket to make it put even more force on the heatsink but when mounted, you can still grap the heatsink and twist it, which I guess is normal.

 

[Kwint Sommer]

Originally posted by: Slickone
It always fails the small FFT's at #10.

Failing small FFT's is generally a sign of a CPU error. None overclocked CPUs that generate errors generally do so because they are overheating. However, if different CPUs failed on the same test then it is probably the motherboard causing problems.

Originally posted by: Slickone
you can still grap the heatsink and twist it, which I guess is normal.

No it's not. None of my P3s, P4s nor my new Core 2 Duo have insecure heatsinks.

 

[Pabster]

Originally posted by: Slickone
"...you can still grap the heatsink and twist it, which I guess is normal."

Er, no. Unless you're making keychains.

 

[Slickone]

Originally posted by: Pabster

Originally posted by: Slickone
"...you can still grap the heatsink and twist it, which I guess is normal."

Er, no. Unless you're making keychains.

Not very easily twisted, but it can be done.
So what's supposed to stop that? I took the M shaped bracket and bent it even more past the way it usually is, so it should be putting even more force on it. It now takes a lot of force to get on.

 

[Slickone]

Here is a picture of the grease on the CPU after removing the heatsink once or twice. Here is a larger pic, but after it had installed it again and moved it around, thus the disturbed areas where I touched it.
Heatsink.
Bracket (looks like it's bend itself back to normal a little).

 

[Slickone]

Ideas?

 

[Slickone]

Is it a bad motherboard, or did I do something wrong?

 

[Slickone]

^

 

[cmdrdredd]

Could really be anything. I'm guessing it's heat related. If it goes up too high it fails usually.

 

[Budman]

Have you tried running memtest?

didnt these old boards have L2 cache on the board? maybe they're going bad....

 

[Slickone]

Originally posted by: Budman
Have you tried running memtest?

didnt these old boards have L2 cache on the board? maybe they're going bad....

Well I ran memtest before and it ran with no problems, but ran it today just to make sure and it fails after awhile, up towards the upper 'area' (?) of the memory. Meaning, it has 256MB, and fails around maybe 230 and up. I'm not real good reading memtest results or how it works, but once it failed it then only kept going through different areas up the upper memory areas, flashing #'s like mad and every single location failed.
I tried running each stick by itself and got no failures. So is it DIMM slot #2 (or paths/controller to it)? Of course I can't run only a stick in slot 2 to check it.

Between the time when memtest ran OK and now, I swapped the CPU out a couple times. Maybe I screwed something up.

 

[Slickone]

^

 

[nullpointerus]

Originally posted by: Budman
Have you tried running memtest?

didnt these old boards have L2 cache on the board? maybe they're going bad....

I was just going to suggest this. I once had a board RMA'd due to bad L2 cache. It was a terribly poorly-made board that would only last a few months between RMA's even without any overclocking. I think it had VA-530 or something in the name.

 

[nullpointerus]

Originally posted by: Slickone

Originally posted by: Budman
Have you tried running memtest?

didnt these old boards have L2 cache on the board? maybe they're going bad....

Well I ran memtest before and it ran with no problems, but ran it today just to make sure and it fails after awhile, up towards the upper 'area' (?) of the memory. Meaning, it has 256MB, and fails around maybe 230 and up. I'm not real good reading memtest results or how it works, but once it failed it then only kept going through different areas up the upper memory areas, flashing #'s like mad and every single location failed.
I tried running each stick by itself and got no failures. So is it DIMM slot #2 (or paths/controller to it)? Of course I can't run only a stick in slot 2 to check it.

Between the time when memtest ran OK and now, I swapped the CPU out a couple times. Maybe I screwed something up.

Can you pick up an old K6-2 board on eBay for like $20 or something?

 

[Slickone]

Originally posted by: nullpointerus

Originally posted by: Budman
Have you tried running memtest?

didnt these old boards have L2 cache on the board? maybe they're going bad....

I was just going to suggest this. I once had a board RMA'd due to bad L2 cache. It was a terribly poorly-made board that would only last a few months between RMA's even without any overclocking. I think it had VA-530 or something in the name.

Is there no way to tell if it's the cache that's failing with one of the tests?

 

[jiffylube1024]

Possibly heat related and quite frankly those older CPU's may not be 100% stable at default settings anyways.

Is there any way to up the voltage to the CPU? If you can mount a fan near the CPU (or if it's not running that hot) then upping the voltage often fixes Prime95 errors.

 

[betasub]

Originally posted by: Slickone
Is there no way to tell if it's the cache that's failing with one of the tests?

Usually there's a BIOS setting to switch off on-board cache. My GA5-5AX even had a setting to switch off on-die (L1) cache: not surprisingly booting slowed to an absolute crawl.