Corsair 2-2-2-5 & A8V works great, but beware...

Discussion in 'Asus' started by David O, Feb 22, 2005.

  1. David O

    David O Guest

    To all,

    I build a new system every few years. I'm not an overclocker at
    heart. For me, stability and reliability are the prime goals. I'll
    start by briefly presenting the lessons I've learned from my latest
    build, so that those who already understand such lessons will be able
    to skip my rather lengthy story.

    Lesson 1) Make sure your RAM is being supplied the correct voltage,
    especially when using those "special" low latency RAM modules that may
    be specified at higher than standard voltages. Your motherboard may
    not automatically supply those special voltages.

    Lesson 2) Your computer may seem to be running your OS and
    applications just fine even if your memory has occasional errors. You
    should test your memory with a good memory testing program, because an
    apparently fine running machine may have subtle memory errors that
    could, over time, insidiously corrupt your programs and data.

    Now for the story behind those lessons:

    I recently built a new system (Asus A8V Deluxe, Athlon 64 3500). Just
    for the hell of it, and quite out of character for me, I decided to
    try some low latency ram -- two sets of Corsair TWINX1024-3200XLPT
    (that's 2GB total). The install of XP Pro went smoothly on the new
    system and everything seemed to be working fine. Later that day I
    checked the RAM timings in BIOS and saw that they were not the 2-2-2-5
    of the Corsair spec. Instead, the BIOS had set the timings to a more
    relaxed spec. I therefore manually changed the BIOS memory timings to
    2-2-2-5 and rebooted.

    Great! Things seemed to be running just fine with the new RAM timing,
    and my computer was faster too -- a computation of pi to 1 million
    digits now took 2.29 seconds rather than the 2.44 seconds it took with
    the previous RAM timings. BTW, exactly what I'll do with those one
    million digits I don't know, but, if you are curious, a Google search
    of "Pi Computation" will lead you to some sites that will reveal
    exactly how far computing has come in the past few decades. Now, back
    to my story.

    After playing around with my new machine for a few more hours, I
    decided to find a program to better test my RAM. Why? Well, as most
    computer geeks know, the RAM test at POST is hardly exhaustive.
    Google lead me to http://www.hcidesign.com where I downloaded their
    free Memtest program. While running the program, I was quite
    surprised to find occasional errors reported (about one error per 200
    MB tested). Not acceptable!

    I located the Corsair data sheet for the RAM on their web site and saw
    that their timing tests were performed at 2.75 volts. Ah ha! The A8V
    BIOS had set the RAM voltage to the more standard 2.5 V. After
    changing the RAM voltage to 2.8 V in BIOS, I ran the Memtest program
    again for over an hour with no reported errors. Satisfied that I had
    fixed the problem, I then overclocked the machine (CPU + FSB) +5% for
    a margin test and ran the Memtest again. With no errors reported, I
    returned the computer to the normal clock rate. Margin tests like
    that are important because otherwise you wouldn't know if your memory
    was just teetering on the edge of failure, as it were.

    I think the important lesson here is that if I hadn't decided to
    perform the memory test, I could have gone for months or even years
    with memory that was generating occasional errors. Surely such errors
    would cause problems at some point, but being so rare I might well
    have chalked the problems up to buggy software.

    This is an important lesson, I think, for all of us who like to build
    our own systems and who sometimes like to push the limits of our
    hardware: A seemingly fine running machine may not be so fine after
    all. You can't judge a computer's health merely on the fact that it
    passes POST, boots the OS day after day, and runs your apps. Check
    that RAM with a good memory testing program and if you get ANY errors,
    find and fix the problem!

    Cheers
     
    David O, Feb 22, 2005
    #1
    1. Advertisements

  2. David O

    Ben Pope Guest

    Indeed.

    A good test for CPU stabilty is Prime95 torture test. This tends to throw a
    fit MUCH sooner than you would notice in just about any other software.

    Ben
     
    Ben Pope, Feb 22, 2005
    #2
    1. Advertisements

  3. David O

    Michael S. Guest

    GREAT points!!!! especially for new builders as well as some of us that
    just forget!!

    MikeSp
    --------------------------------------
     
    Michael S., Feb 23, 2005
    #3
  4. David O

    Paul Busby Guest

    Thus spake David O:
    Setting *all* FSB, CPU/DDR voltages & memory timings to manual is usually
    the best approach. Doing so forces one to research more but seems to get
    stable results with the best performance a lot quicker. Run memtest86 &
    Prime95 but not /all/ stability problems are caught with either. XP's ACPI
    Standby mode is shot on my PC at the moment, forcing me to use Hibernate
    instead. I'll get round to troubleshooting this if the anomaly remains with
    faster RAM. I've found this forum very informative:
    http://www.nforcershq.com/forum/viewforum.php?f=13
     
    Paul Busby, Feb 24, 2005
    #4
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.