AI7 Temps & Water Cooling Considerations

Discussion in 'Abit' started by Jim, Jan 8, 2004.

  1. Jim

    Jim Guest

    Abit AI7 (BIOS 14)
    P4 2.6C @ 2.85GHz (13 x 219MHz)
    Coolermaster Aero 4 HSF
    1GB (2 x 512MB) Kingston HyperX PC3200 (@ 219MHz, 2-3-2-5)

    The memory tops out @ 219MHz, and although the timings remain TIGHT, I can't
    get a single MHz higher reliably, even if I relax the timings COMPLETELY,
    kinda odd. IOW, it runs FULL BORE, no errors w/ Prime95 up to 219MHz, then
    totally collapses beyond that, no matter the timings, increased voltage,
    etc. But that's a minor issue, I'm happy to get the 10% OC given the

    I've had this setup for about three weeks, so far so good. But I'm dubious
    about the temp readings.

    Room temp is about ~25C...

    Aero 4 @ 2200 RPM: CPU idle ~38C, under load (Prime95), ~44C
    Aero 4 @ 3800 RPM: CPU idle ~32C, under load (Prime95), ~38C

    I've been an AMD user for several years (Athlon 1.2GHz, Athlon XP 2000+) and
    only switched to Intel recently after growing tired of AMD's lack of thermal
    control, and although I know Intel has AMD beat in the cooling department, I
    wasn't expecting it to be THIS LOW!

    It's not that I'm complaining, but if these temps are true, then I have some
    concerns about delving into the water cooling arena (a recent interest). I
    was considering the Swiftech H20-8501P kit because the CPU will NOT run
    reliably w/ Prime95 over ~240MHz. What I want to do is run AT LEAST 250MHz
    (1:1) and swap my Kingston PC3200 for some OCZ (or similar)
    high-performance, presumably PC4000 or better. I'll happily run 5:4
    temporarily once I get the CPU straightened out. But given the considerable
    cost of water cooling, and the temps I'm already getting, I'm leery of
    whether water cooling will actually improve by ability to OC. For pete's
    sake, at the current temps, why all the trouble in reaching beyond 240MHz?
    I know it could be the NB that's the problem too.

    I'm just looking for a sanity check here before investing anything into
    water cooling. Are these temps to be trusted? If so, why the lackluster OC
    of the CPU? I'm watching video after video at of one water
    cooling solution after another, and every time the test results are
    reporting these enormous OC's (e.g., P4 2.6C to 3.5GHz) and idle temps of
    ~38C, load temps of ~47C. Heck, I'm beating that already on air cooling! So
    what's the deal here? What temps are YOU getting?

    I assume if the temps are NOT to be trusted, my readings after water cooling
    would drop significantly, but how much?

    Btw, if you have any water cooling recommendations, I'm all ears. I like
    the Swiftech because the WBs look great, it's simple, no case modding, etc.
    A bit expensive, but given the apparent quality and completeness, perhaps
    worth the investment. Wondering if using TWO radiators might be worth the
    investment as well (although a little worried about excessive noise). I'm
    hesitant about something like the Thermaltake Aquarius because there's no
    chipset WB. HardOCP reported OC problems w/ the AI7 due to NB limits unless
    water cooled (or heavily air cooled). So a CPU+NB solution seems the
    minimum on this board. Only have a Ti4200 video card, and not really into
    video card OC'ing. Seen others recommending "custom" solutions as well,
    gathering WBs one place, pumps another, but what a hassle. Is something
    like the Swiftech H20-8501P likely to reach 3.5GHz (270MHz)? On ONE
    radiator? Two?


    Jim, Jan 8, 2004
    1. Advertisements

  2. Jim

    Tony Guest


    Jim, I also have an AI7 and 2.6C. I bought two 512MB sticks of Buffalo Tech
    PC3700, and cannot get it to budge past 210MHz (that's not even close to
    what it's rated for!). I have been scouring the newsgroups, Google, and the
    Abit forums to try and figure out what's happening. Unlike you, I've been
    using and oc'ing Intel based boards for quite some time, so my gut tells me
    there's something wrong here. Your temps, however, are not out of line. I
    still have the cover off my case, and my CPU/SYS/PWM is 23/22/26 at idle.
    Practically an ice cube.

    I'm starting to suspect one of the following things:

    1) The BIOS is not yet fully tweaked. Not impossible; the IC7 (as I've read)
    took quite a few BIOS upgrades before it was considered a great overclocker.
    OTOH, I have read posts from other AI7 owners that seem to have the board sm

    2) There is an issue with the NB on this board. Maybe the sample to sample
    performance of the NB design on the AI7 is such that some boards just can't
    do it.

    3) Maybe there's a problem using two 512MB sticks. I can't remember where I
    read it, but there have been comments about other boards that don't oc as
    well with two 512MB vs two 256MB sticks.

    I just got two 512MB Kingston HyperX sticks from Newegg. I'll be trying them
    out in less than an hour.
    Tony, Jan 9, 2004
    1. Advertisements

  3. Jim

    Jim Guest

    Thanks for the reply... I originally had 512MB (2 x 256MB) of Mushkin PC3500
    Dual Level II "Black", best in the business, the results were AWFUL, loose
    timings, 2.8v, nothing helped! I had to RMA it back to Newegg. I was able
    to pick up the HyperX 1GB real cheap, direct from Kingston, $176! Works
    fine, will OC slightly, but far from my original intentions. Just happy to
    be running 1:1 @ 220MHz w/ current setup. But disappointed, I was hoping
    the Mushkin would OC a little over PC3500, maybe even PC4000, but damn it, I
    couldn't even get it to hold 220MHz or so. I assume your Buffalo are BH-5,
    like the Mushkin, so I'm stumped. As I mentioned in the original post, the
    NB has been shown to be a bootleneck. Unless you're water cooling, I don't
    get your temps (or mine). YOU'RE IN THE 20'S?! The temp readings on this
    board have to be BOGUS! Even mine, in the low to mid 30's fluctuate far
    more than my AMD Athlon ever did, varying 2-3C over a period of 20-30
    seconds. I just don't trust the absolute readings on this board, no way,
    esp. after hearing your temps.

    Anyway, at least I'm not alone, and you may be right about the immature BIOS
    (but the updates have been slow in coming lately, not a good sign).

    Jim, Jan 9, 2004
  4. Jim

    Tony Guest

    Okay, this is going to get confusing, as I bottom-posted my last response...

    Keep in mind my BT sticks are PC3700. I should have been able to get 233MHz
    STOCK out of them, and could not even get past 210. Well, I'm cooking right
    now with the Kingston HyperX PC3500 (NOT 3700) and it IS running at its
    stock 217MHz without even having to touch the base DIMM voltage (2.6V). I
    had to bump the BT up to 2.8V just to get it to 210MHz. Multiple runs of
    memtest86 tests 3 and 5 show no problem with the Kingston. Running Prime95
    right now. Will let it run overnight.

    Did I just have bum sticks of Buffalo Tech? Maybe. They are NOT the BH5
    variety; there are the Micron chipped sticks. I did have one BT WB 512MB
    stick that couldn't even do as well as the Micron! Funny thing is I'd have
    great luck in the past oc'ing Micron memory, but that was back in the old
    days of PC100 and PC133 memory. Maybe things were easier back then.
    Interestingly, the Kingston are marked on the blister pack "Winbond", but
    with the heat spreader I don't know if it's BH5 or CH5 chips.

    I was also looking at the Mushkin, but they were out of my price window.

    Yeah, maybe the temps through AbitEQ are bogus. Then again, maybe they
    aren't. Remember, the cover is off, I have a Zalman CuAl7000 on the CPU, two
    fans in the PS, and two Panaflo 2950RPM fans in the case going full blast. I
    just wish Motherboard Monitor would get moving and update the app to work
    with these new uGuru boards.
    Tony, Jan 9, 2004
  5. Jim

    TomG Guest

    you could always set things up manually and use a 5:4 fsb to ram ratio which
    would allow you to push the cpu much higher while back-shifting the ram
    speed. while this might seem counter-productive, in most cases the increase
    in CPU speed more than makes up for the downshift in ram bus speed.

    you would need to set the NB to 800 and then set the ram speed to 5:4.


    Thomas Geery
    Network+ certified Abit Mirror <----- Cable modem IP
    This IP is dynamic so it *could* change!...
    over 120,000 FTP users served!
    TomG, Jan 9, 2004
  6. Jim

    Jim Guest

    Thanks for the input... in general, I feel a 1:1 setup produces better
    results UNLESS the asynch difference is truly significant. Does 250:200
    qualify? Not sure. But even if it does, the problem I'm having in even
    considering this possibility is that without more extreme CPU cooling, I
    can't get the FSB beyond ~235-240MHz (I tried 5:4, the CPU and/or NB is
    definitely the bootleneck). Of course, if I run 240 FSB, it underclocks the
    ram (e.g., 240:192). At least intuitively, this doesn't seem nearly as
    beneficial as running 220MHz (1:1). That's why I was considering water
    cooling to get the CPU to at least 250MHz, then run 5:4 temporarily until
    upgrading the RAM to PC4000 (1:1 nirvana). In the meantime, running 5:4 w/
    240 FSB doesn't seem to me to be the better choice compared to 220 FSB
    (1:1). Could be wrong, but seems unlikely. I noticed that when I did use
    5:4, Memtest-86 walltime went through the roof, it was awful, the async
    behavior seem to wreak havoc w/ the system.

    Jim, Jan 9, 2004
  7. Jim

    TomG Guest

    well, for what it is worth, a lot of us are using async ram timing with
    great success. I would recommend using some benchmark utilities to do
    testing. benchmarks are useless to the real world except for comparison
    testing for different settings. I would try Sandra memory bandwidth
    benchmark and some others of your choosing and see what sort of numbers you
    get and you may be surprised at the results.

    currently getting 244 out of mine and my ram would not do 1:1 at that FSB
    and the system is much faster in an async setup than running at the
    limitations of the ram.


    Thomas Geery
    Network+ certified Abit Mirror <----- Cable modem IP
    This IP is dynamic so it *could* change!...
    over 120,000 FTP users served!
    TomG, Jan 9, 2004
  8. Jim

    Tony Guest

    15 hours under Prime 95 Torture Test with no errors or warnings. I did have
    to bump VDIMM to 2.7v and VCPU to 1.55v for Prime95. Interestingly,
    memtest86 showed no memory errors at 2.6 volts.
    Tony, Jan 10, 2004
    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.