I've got a P3-800EB in a spare machine which as far as most people are concerned is working fine. However, I got around to installing some new memory into it as it was going spare and managed to drop one of the sticks onto the floor (yeah I know, damn these short stubby fingers).
Anyway, after that I thought I'll run memtest86 to make sure all is well with the memory as I have seen RAM die after a fall. To my annoyance memtest flagged up some errors on test #5 [Block move, 64 moves, cached], but no other errors. This slightly bugged me as none of the burst or inversion tests caused a problem which technically are harder on the RAM circuits than moving blocks around. Anyway, to make sure the stick really was bad I took it out and tried it in another system and all was fine, no errors at all.
I put the memory back in the P3-800 and it was giving memory errors again only on test #5. Swapped memory around from different machines and the problem was still there. Obviously not a memory related problem anymore.
I re-ran memtest with the cache forced off and no errors occurred. Just to make sure I ran test #8 [Block move, 512 moves, cached] with standard settings and got errors but get no errors on any other test.
Have tried different CPU's in the board (P3-500 - locked 100MHz FSB, and a P2-350 @ stock and @ adjusted to 133MHz FSB) and get no errors. All slot CPU's, no slockets involved.
This leads me to think its not a board or RAM problem but a problem on the chip with the L2 cache. The only thing is, its only memtest86 which finds this error, and the more I run the test, the higher the error count increments with every pass. Originally it was finding under 100 errors on test #5 with every pass, now it is finding over a few hundred and in some cases over 1000 errors on each pass - all of which disappear with disabled cache.
Windows (2000 or 98SE) run without a hitch and test software such as Prime95 torture test (~12 hour run), Hot CPU tester, Seti@home, and encoding a DivX file all work with no errors reported and no problems with the output.
So, to the main question, is there any software out there which really hardcore tests a CPU for reliability in it's calculations and cache?
Anyway, after that I thought I'll run memtest86 to make sure all is well with the memory as I have seen RAM die after a fall. To my annoyance memtest flagged up some errors on test #5 [Block move, 64 moves, cached], but no other errors. This slightly bugged me as none of the burst or inversion tests caused a problem which technically are harder on the RAM circuits than moving blocks around. Anyway, to make sure the stick really was bad I took it out and tried it in another system and all was fine, no errors at all.
I put the memory back in the P3-800 and it was giving memory errors again only on test #5. Swapped memory around from different machines and the problem was still there. Obviously not a memory related problem anymore.
I re-ran memtest with the cache forced off and no errors occurred. Just to make sure I ran test #8 [Block move, 512 moves, cached] with standard settings and got errors but get no errors on any other test.
Have tried different CPU's in the board (P3-500 - locked 100MHz FSB, and a P2-350 @ stock and @ adjusted to 133MHz FSB) and get no errors. All slot CPU's, no slockets involved.
This leads me to think its not a board or RAM problem but a problem on the chip with the L2 cache. The only thing is, its only memtest86 which finds this error, and the more I run the test, the higher the error count increments with every pass. Originally it was finding under 100 errors on test #5 with every pass, now it is finding over a few hundred and in some cases over 1000 errors on each pass - all of which disappear with disabled cache.
Windows (2000 or 98SE) run without a hitch and test software such as Prime95 torture test (~12 hour run), Hot CPU tester, Seti@home, and encoding a DivX file all work with no errors reported and no problems with the output.
So, to the main question, is there any software out there which really hardcore tests a CPU for reliability in it's calculations and cache?
Facebook
Twitter