Every once in a while we see some very unique approaches to graphics card designs, even more so than just a slightly modified or fan-less heat sink (like ASUS's silent 6600 GT). Many times, HSF alterations on reference designs are to cut down on heat or noise that's generated during the card's operation, but some solutions are a bit more dramatic than others.

Liquid cooling with GPUs has been around for a while now and there is a lot of debate over how practical and effective it really is. It can definitely be impressive to see a gaming setup with its processors massively overclocked and cooled with liquid nitrogen systems, but for most people this type of system is very much out of their reach. Water cooling is a little more practical, but in order for it to be anything more than a novelty (i.e. used only be a small number of enthusiasts), it has to yield practical benefits without a lot of extra cost and trouble.

In principle, water cooling has some very nice benefits. For one, there is potential for these cards to run much cooler than air cooled cards, which would allow for high overclocks. Fan noise could also be cut down drastically, as water moving through small tubes is generally far quieter than a fan blowing air. However, one thing water can't get you is temperatures lower than the ambient room, but for most people that means 25 C or less, so that should be sufficient.

Today we are looking at two cards using ATI's current top-performing solution, the X1900 XTX. One of the cards is made by Sapphire, a company known for its innovative graphics card designs. This card features a slightly different take on water cooled graphics solution, and it aims to address the address the issues many people have with assembling water cooling systems. The second card is a more traditional design from Connect3D, using the same X1900 XTX chipset. It will also as something of a baseline performance metric, since absent special cooling configurations most X1900 XTX cards will perform similarly.

These two X1900 XTXs are very different from each other and show how much variation there can be between two versions of the same graphics card, and that should make things interesting. While the Blizzard X1900 XTX may overshadow its Connect3D competitor in terms of design, performance will still be similar in many areas, provided clock speeds are the same. We'll look at overclocking potential as well as noise levels, though, and there at least we should see some more substantial differences.

Connect3D Radeon X1900 XTX


View All Comments

  • gersson - Monday, April 10, 2006 - link

    Thank you for constantly deleting my posts especially if they have something completely harmless and neutral. I notice that they are usually erased when I am the 1st poster...

    This is my last post @ daily tech. Enjoy it while it still exists -_-
  • JarredWalton - Monday, April 10, 2006 - link

    Not sure if Daily Tech has deleted any of your posts or not, but I can tell you that there have been 13 comments on this article so far (this will the #14) and none of them have been deleted. None of them have been "voted down" either, so by default you should see all the comments. You might want to verify that your filter at DailyTech is set to -1 so that you can see all posts. Reply
  • LoneWolf15 - Monday, April 10, 2006 - link

    I think it's likely that a solution from Arctic Cooling (i.e., the Accelero X) or one of Zalman's coolers would do as good of a job, be reasonably quiet, and not have the drawbacks that a liquid-cooled setup would. It would be interesting to make a comparison as to how they would stack up.

    The Sapphire card looked interesting, but the results indicated to me it's a lot of sizzle without much steak.
  • Questar - Monday, April 10, 2006 - link

    "Of course, if the GPU runs cooler it should draw less power"

    This is new to me. Can you point me to anything to back this up?
  • NegativeEntropy - Monday, April 10, 2006 - link

    As you imply, the author is either mistaken (as the evidence shows) or was making a faulty assumption. For a given CMOS chip to consume less power one of two things needs to happen: lower the frequency or the voltage. This assumes no fancy technology to turn off part of the chip. Assuming this thing runs at stock voltage, it should consume slightly more power than stock due to the frequency increase (not counting the pump). Reply
  • Gioron - Thursday, April 13, 2006 - link

    Close to true, but not quite. The dynamic power of switching transistors is unnafected by temperature (its proportional to frequency and the supply voltage squared), but leakage current through an off transistor does decrease at lower temperatures and this current is a constant unaffected by the frequency. However, for most chips this isn't a huge difference... yet. As processes scale to smaller feature sizes leakage is increasing and some manufacturers have already run into problems with chips where 40% of their power is leakage and they suddenly had a chip using almost twice as much power as they expected, but most current chips see only a very small percent of their power usage as leakage current and making a small change to a small percentage of power is not likely to be noticable. Reply
  • DerekWilson - Thursday, April 13, 2006 - link

    This comment resulted from a conversation I had with Josh about temperature and power. As you have mentioned here, leakage current does vary with temperature. And, in any material, conductivity is affected by temperature as well -- which will also affect power. Two otherwise identical chips will absolutely draw different ammounts of power if one is very cold and the other very hot. Whether this ammount is significant is not addressed in this article, but the fact does remain that temperature affects power draw.

    And, as has been pointed out, with 90nm and below we will see this phenomenon increase (provided no one stumbles upon a miraculous answer to the leakage problem).
  • Sunrise089 - Monday, April 10, 2006 - link

    The Sapphire card seems pretty dissapointing. You give up an extra slot (right?) and pay ~$50 more to get almost no decrease in noise, almost no more of an overclock, and no less heat. Read that again: NO LESS HEAT! To me, a watercooling setup that runs equally hot as an air cooled setup isnt worth the leak risk, slight as it may be. Reply
  • JarredWalton - Monday, April 10, 2006 - link

    Two things: it takes longer to get to the same temperatures. Second, standard XT/XTX cards are two-slot solutions, so this just makes the second slot movable to either side of the GPU. But yeah, it's a bit underwhelming. At least it's a bit quieter - X1900XTX cards aren't very good on noise levels. Reply
  • mkruer - Monday, April 10, 2006 - link

    Does this seem completely pointless? Why not use heat pipes to draw more heat away from the core of the GPU and keep the fan heat sync in one assembly? This seems totally pointless unless you like having the possibility of your graphic card spring a leak. Reply

Log in

Don't have an account? Sign up now