Leadtek Geforce GT 220 Extreme Overclocking Experiment

Overclocking/Overclocking Tests by massman @ 2009-11-10

We pour some -180°C LN2 over NVIDIA´s 40nm GPU to see how high it will scale. With the help of some voltage modifications we are ready to breach some world records at the low end of the video card scale.

  • prev
  • next

Setup and Performance scaling


Massman's 3D on LN2 setup
Madshrimps (c)
CPU Intel Core i5 750
  • Cooler Master V8
  • KingpinCooling NB-1 LN2 container
  • MainboardGigabyte P55-UD6
    VideocardLeadTek Winfast GT220 1GB
    Memory 2x 2GB OCZ Platinum
    Storage Western Digital 250GB
    Chassis Cooler Master Lab Test Bench
    PSU Antec TruePower 1KW

    Performance scaling

    Before we start the actual extreme overclocking, it's important to find out what gives us the biggest increase in performance. Therefore, I removed the stock cooling of the card, switched to a more performing Zalman GPU Cooler sample, increase the voltage just a bit and started testing the frequencies using the 3DMark2001SE Nature subtest as reference benchmark. Although this benchmark is already 8, almost 9, years old, the Nature subtest still gives a good impression of the performance of a video card. Note that you could in fact use any benchmark to test scaling; as long as the difference in performance is significant enough, all that matters is the difference itself. Not the actual FPS.

    First up, GPU scaling. This includes increasing the core frequency and the shader frequency, which is linked to the core.

    Madshrimps (c)

    As you can see, a significant increase of the core/shader frequency is, well, almost redundant. The performance doesn't really increase and what we feared from the beginning (low-end card thus memory bottlenecked) may very well be true.

    Let's have a look at the memory overclocking performance scaling.

    Madshrimps (c)

    And, yes, our assumption turned out to be correct. This card is so heavily stripped from all of its features (48 shaders, 2 rops, 128-bit memory) that the real bottleneck has been re-located to the memory rather than the core. A pity, because this means that we will need a lot of luck to have well-clocking memory. On the previous page, you already saw that the memory isn't really high-end.

    Let's hope for the best and go on to the overclocking results ->
    • prev
    • next
    Comment from Massman @ 2009/11/10
    John, could you attach both PDF's to this thread? People who are no OCTB-member can't seen the voltmod thread.
    Comment from jmke @ 2009/11/10
    Comment from Massman @ 2009/11/10

    I'm actually tempted in buying another GT220 to check for overclockability :-)
    Comment from Kougar @ 2009/11/16
    Interesting read.

    Since the memory chip temperatures are obviously holding the core/shader speeds back, wouldn't it not be very hard to jury-rig an inverted peltier block, hot side on the chips? Wouldn't take very much, adjust the peltier voltage as needed to keep them just above zero centigrade... perhaps with a pot that allowed extra spacing. Or direct heating the other side of the PCB directly underneath the memory solder points, sort of like an inverse waterblock for the memory if it could be insulated from shorting.