Sign in with
Sign up | Sign in
Your question

Prime95 error with cpu/memory blend test

Last response: in Overclocking
October 26, 2006 9:35:29 PM

Hello all, when I run prime95 in the blended cpu/memory mode I always get an error a few minutes in. It says something like "result rounded to .5, expecting <.4".

I can run the cpu mode overnight without problems, and I can run the memory heave mode overnight without problems, and I can also run memtest overnight with no errors.

Proc: e6400 @ stock clock, core temp reports ~38C idle, ~44C load.
mem: Corsair TwinX2048-6400C4, manually set to DDR800 with 4-4-4-12 timing
mb: P5W DH Deluxe.
vid: powercolor x1950XTX
sound: x-fi fatality

I saw another poster was having a similar error and narrowed it down to his modem. Should I try moving my sound card down? right now it's in the first available PCI slot.

I should note that I haven't had any trouble in any other programs that I've notice. These include FEAR and Company of Heroes.
October 27, 2006 12:57:19 AM

Try whatever you think of... it cant hurt.

It sounds like you got another issue, and that the results wont be too plesent if your doing calculations with your computer that has to be right.

VideoGames can crash... Its no big deal. But if your serving Data, storing Media Files, Creating Video/Media Files, Working with Data, etc... then you better make sure prime never fails.

October 27, 2006 11:53:07 AM

Well, i now have some interesting news.

I actually went in and overclocked based on wusy's guide. My memory is set to use the SPD timings which cpu-z is reporting as 5-6-6-18. I'm running 1:1 with a 333FSB.

Ready for the punchline? I've been running Orthos for ~11 hours now and not run into any errors. Core Temp reports the temp never exceeded 48C, and my Vcore is fluxing between 1.336 and 1.352 based on cpu-z.

When the test failed, I had the mem manually set to DDR2-800 with 4-4-4-12 timings. Is it possible for memory to pass a memtest but still fail the blended test because if the timings?
October 27, 2006 3:47:30 PM

It could be the way in which the memory is accessed that is different. From my understanding, MemTests work to test only the memory.

The ultimate testing software should do one of two things: Attempt to limit the scope of the test down to a single aspect, or too test the system as a whole.

A test that works well at limiting the potential for outside causes to result in bad results allows us to eliminate and/or define key strengths, weaknesses, and defects.

Prime95 tests everything. It tests how the memory works, hence if the memory is defective it will fail. It tests how the memory is accessed, if how the access of the memory fails then the memory isnt bad your North Bridge is. It tests the communication from the processor to the North Bridge. It tests the Cache of the processor. It tests how the memory is handled logically on the processor. And it tests the ability for the processor to use the memory for calculations. And it does so in a complex way that uses a large portion of the processors capabilities. That results in heat, and a higher degree of failure due to stress then almost anything else I have tried.

However, I was running PRIME 95 all night at 318mhz FSB... But my computer would randomly shut down during gaming. I dropped it to 300, and I havent had a problem.