How many tests of Memtest are enough?

[blckgrffn]

Title says it all!

I am able to run 4*256 of Ballistix DDR500 @ 2.5-2-2-8-2T 240 Mhz 2.9V for 27 passes of the entire memtest suite, or about 8 hours or better, before I get four or five errors on pass 27.

Why run 4*256? Well, I picked up a gig of it from Crucial early last week for $110, that's why... the 256 alone was $45, the 512 kit was $55, so I thought that I would try it. We all know that 2T doesn't mean squat at this point, and I am still well above regular LL PC3200 in bandwidth.

I just want to know if getting errors after that long is acceptable or if I need to do something different.

Thanks!
Nat

 

[ryoanji]

My ballistix seems to be sensitive to the temperature. Do not know about yours but you might wish to either try fanning or rule out exogenous rise in the temperature, such as temp rising during the day, etc. etc.

Also, ASUS boards are notorious for conflicts with Ballistix DDR500 that is my I moved to DFI and much hapier right now.

 

[blckgrffn]

I wish that I would have known that before I started. I had to try many different configurations of sticks, reboot, etc. until it finally found its happy place. My ballistix runs HOT compared to my previous ram.

It seems conceivable to me that after many hours and pounding on the ram that probably is little like normal system usage that there can be an "acceptable" level of error, just wondering what that is... if prime fails after 16 hours, that is usually ok by me...

 

[Acanthus]

any errors are bad.

 

[blckgrffn]

Ok, I am going to try and back the voltage back down to 2.8V, right where it is rated at and see how that goes...

Nat

 

[bjc112]

Originally posted by: blckgrffn
Ok, I am going to try and back the voltage back down to 2.8V, right where it is rated at and see how that goes...

Nat

Good idea, I have heard of some G. SKill without heat spreaders push FURTHER with less voltage..

2.8V might do the trick.

 

[jpeyton]

My Patriot TCCD does 300FSB at 2.7v. It's old TCCD, which doesn't like high voltages; if I raise the voltage to 2.8v at any speed, it errors in Memtest. Newer TCCD responds well up to 2.9/3.0v.

If lowering the voltage doesn't help, you can try burning the memory in with Memtest. People report that repeatedly running their memory in Memtest produces less and less errors with each pass. This doesn't work if you have a massive number of errors, but you can try it if you only have a few.

 

[blckgrffn]

Thanks for the input, guys, I will keep it in mind... with any luck it will pass the 24 hour mark in memtest for me, and it will be prime and memtest stable, and you know what that means... time to sell it and move on

Nat

 

[uOpt]

memtest is not a very good test for actual system stability, it doesn't put enough load on other components and is probably not reproducing memory patterns closely enough.

FreeBSD's `make world` is the best hardware test I have ever seen, closely followed by compiling the mozilla and gcc source codes in a loop, where each new gcc compiles a new mozilla.

Prime95 is probably much better than memtest already.

 

[flexy]

i had situations where i run memtest #5 for 30 passes without problems....but with the same settings extremely instable in windows with prime95/occt.

In memtest you shouldn't get ONE error ever.

My philosophy right now is that memtest is literally worthless EXCEPT to check whether your ram is actually working and as a double-check eg. before you boot into Win. (So you dont mess up your Win installation)

The real testing starts in Windows w/ programs like OCCT or prime95.

 

[eplebnista]

Originally posted by: Acanthus
any errors are bad.

:thumbsup:

 

[Amaroque]

I see that you're running 2.5-3-3-8. Try running 2.5-3-3-10.

Ballistix like a Tras cycle time of 10 clocks at about 220 MHz and up. You might even be able to drop them back to 2.8v.

 

[Painman]

My 'Stix do NOT like to be overvolted, hopefully dropping to stock volts does it for you. That'll certainly help if your errors are heat related...