While both ATI and nVidia use a form of 'thermal throttling' in brand of GPU to stop them from being damaged by excessive heat due to overclocking (or other circumstances), nVidia's implementation of this feature tends to be a bit more sensitive than that found on ATI cards.

It's extremely rare to see artifacts or visual imperfections on a current or previous generation nVidia card. This is because the video core will manually slow itself down before a state stressful enough to cause video artifacts is reached. As readers might imagine, this feature makes determining the maximum overclock attainable by an nVidia card more frustrating than it otherwise would be.

When using the Coolbits hack to overclock an nVidia videocard, there is no indication that this throttling has been activated, so the only way to tell if you've pushed the card beyond its temperature limits is by noticing a decrease in benchmarking scores. If at x MHz the benchmark score increases as normal, but at x+5MHz it suddenly drops several hundred points, the thermal throttling has just kicked in. The frustration really sets in when trying to determine the precise point that the feature activates. Since the throttling is heat-triggered, it won't necessarily be the same value every time.

Best Practices for Overclocking an nVidia Card

To fully exploit an nVidia card's overclocking potential, you should follow this procedure:

1. Using the set of instructions above, enable overclocking on the nVidia videocard.

2. Increase the videocard's core speed only (do not raise the memory speed yet) by 3-5MHz increments, benchmarking at each increment and recording the score. Look for 'snow' (small white particles appearing intermittently) during each benchmark run as a sign the core's maximum speed has been reached. When you see snow, the benchmark refuses to complete, or when the benchmark score mysteriously decreases, the core has been pushed too far.

3. Power down the system for a few seconds to allow the card to cool, then reboot and return the card to its previous fastest 'safe' settings. Increase core speed from here in 1MHz increments until snow appears, the benchmark refuses to complete or the benchmark score decreases. The fastest stable score here is the nVidia cards maximum core speed. Note that due to nVidia's sensitive thermal throttling, the maximum core speed reachable with 5xxx or 6xxx series cards may vary with external conditions. With nVidia cores, it's usually best to 'ballpark' the figure slightly, backing off a few MHz to ensure that throttling does not occur. Don't forget to make a note of the maximum core speed and return the card to its default settings.

4. Following the same procedure as steps 2 and 3 above, increase the card's memory speed in 5-10MHz intervals, leaving the core speed at stock. This time, look out for 'artifacting' while running the benchmark. 3D artifacts will often manifest as jagged lines or flashing objects appearing briefly at certain points of the benchmark. Though thermal throttling should not be an issue when raising the memory speed, keep an eye out for decreasing benchmark scores anyhow. Once problems manifest themselves, return to the highest 'safe' memory speed and increase it in 1MHz increments until the memory's maximum speed is determined.

5. Now for the real test. Raise both the core and memory speed to the maximum values determined in steps 3 and 4, then run the benchmark. If snow, artifacts or problems running the benchmark occur, back both the core and memory up 1-2MHz then try again until the problems clear up. Once the card is running the benchmark stably, record the settings, reboot and start trying out some favourite games. If these appear to run ok, your nVidia video card's maximum overclock has been determined!

Due to nVidia's sensitive thermal throttling mechanism, it's also a good idea to reboot, back the core speed up about 10-20MHz from your maximum value then benchmark the card again and compare the scores. This will make sure that the thermal throttling has not triggered.