Prime95 failing even on stock settings!

first off, I felt this was the best suited category.

I have a recently upgraded machine from XP to vista premium and on XP I was running a DFI x38 board with 4gb (3gb in XP), 1gb each in each slot 1066/5-5-5-15 patriot RAM with an e8400 intel chip. I had it at about 4ghz (445FSB x 9) and the memory still at 1066. The overclock ran 48 hours in prime95 with the default blend test with 2 threads. stable in XP; but now in vista it fails within minutes overclocked, and fails in under 2 hours on stock settings! what could possibly be the problem???

and before anyone asks, on stock settings I did manually set the RAM to the correct voltage after defaulting to stock settings in the BIOS.

any help is greatly appreciated!

have you checked the actual voltage when the box is up? my x38 (gigabyte) board's memory voltage sags once it boots, even if i have it set to the "correct" voltage, if i have all 4 slots occupied. i have to nudge the memory voltage up one extra notch with 4 sticks (along with my usual FSB and MCH increases).

^Wrong!

1. Memtest 86+ only tests RAM
2. SupperPI does NOT stress the CPU like Prime95 Small FFTs. P95, Ortho, OCCT are all about the same for stress testing. SupperPI is a benchmarking program and not meant as a true stress testing program.

====
@OP:
Like said check RAM first. Also check temps. If your CPU is over heating even at stock (due to incorrect HSF mounting) then P95 will fail. Also are you running Vista x86 or x64?

no,you are wrong
for memtest i am talking about memtest 3.8,not the old version
if you activate 8 memtest in 64bit system then you can stress your system to 100%
i am using that now
for superpi i guess you don't know a low about it too
there's something called hyperpi which could stress all 8 cores and complete the test in a short time instead of running for the whold night like prime 95
it does not have to be a true stress program to know whether or not your system is stable.

ya stressing your system at 100% for longer ensures better stability

and i have also tested memtest86 both in the operating system itself overnight as well as the deluxe version that runs off of a floppy on bootup and there were no errors in the tests. and i'm just running the 32-bit version of vista, fyi

PS what power supply are you using in this system?

nope... set everything manually (what i could manually, most of the default settings were already correct). and now it just freezes up while running prime95. i even had he multiplier at 6x for 2ghz and it still froze. i even upped the voltage a bit while underclocked and it made no difference. :-\

ok. i'l try the 1 DIMM and newer prime95 later. i have to go to work

i'll keep you posted

oh. and i do have the lastest bios update

but less time doesn't equate to being stable....That is the point of the multiple hour tests.

Memtest86(+) stresses jack all other than your RAM. It writes to RAM, reads it back and compares. Your CPU would have to be quite unstable to be unable to compare a bunch of patterns, in which case any real CPU stress tester would fail within seconds.

SuperPi is not HyperPi, you talk about one then the other. SuperPi is singlethreaded and inefficient; any barely stable CPU can pass it.

Prime95 Small FFTs provides a constant near-100% load which is why it is generally considered the best.

well i took out all but one stick of ram and ran prime 95 on stock settings and now windows just freezes about 5 to 10 minutes in. sometimes it BSODs but it doesn't stay up long enough for me to read it. i'm starting to think this is hopeless.

usually if you pass 300% without error on 9 memtest it's stable

^EXACTLY! That's what I was saying earlier. This has happened to quite a bit of people, esp. those who do vmods,etc for max OC that run Memtest86+ fine and crash and burn when loading OS, including command line Linux/DOS (NOW That is being REALLY unstable).

So if you stress both the CPU and RAM together and you get errors, which is the culprit?

That's obvious
for me becuz i am using i7 it's both
since the memory controller was built inside the cpu so maybe i need to give more voltage to it
anyway we can't always depend on the stress program in order to know what's wrong
how would you answer if i asked you the same question?
how would you know what's wrong when errors start popping out from prime95?

i could give that a try. i've just been running the default blend test with 2 threads. i'm sure it's not the memory as i've tested about everything memory related i could. i'm beginning to wonder if it's a failing power supply... :-\

Also note, after all P95/Memtest are software. Like everything, software can make mistakes. If OP doesn't BSOD for a while (~2-3 months),etc chances are he's OK with the OC.

ok i understand what you are trying to say now
i just thought that since the guy who posted this thread couldn't pass it he should try out hyper pi and memtest. i offered these two program out of my personal experience.