NVIDIA GeForce DDR Roundup (March 00)

Overclocking Heat

It is a well known fact that chip producers are overly cautious when it comes to setting a running speed. By setting the stock speed of chips at speeds which they are certain that all chips can reach, companies such as NVIDIA take the responsibility out of ensuring that a chip runs as expected. By doing so, however, the true potential of a chip is never fully reached. It is for this reason that almost every GeForce card out there can be overclocked to some extant. To what extent depends on two factors: transistor conductivity (often referred to as chip quality) and heat.

The first limit mentioned is transistor conductivity, an aspect which is a function of the ratio between the transistor length and width. The fabrication process of chips results in variations in these two parameters, thus the maximum overclocked speed is a function of these. However, neither the manufacturers nor the consumer can tell which chips have maximum transistor conductivity. Thus, we are left judging overclocking potential on how well a card can deal the other limit of overclocking: heat.

How efficiently a DDR GeForce card deals with the additional heat produced by overclocking the GPU varies with the cooling method used. The majority of GeForce cards are cooled using the standard 4.5 cm x 4.5 cm heatsink and fan combo, leaving the differences between these cards determined by how the heatsink and fan are attached to the GPU surface. One manufacturer, Leadtek, actually uses a larger heatsink to aid in cooling. To rate how each of the cooling setups worked, we measured the temperature of the core using a thermistor from an ABIT BF6 motherboard which uses the Winbond W83782D hardware monitoring chip. This thermistor was then placed on the back side of the PCB directly under the GeForce core. A 30 minute loop of Quake III Arena was then run and the highest temperature reached, collected by Mother Board Monitor 4.12, was then recorded and shown in the graph below.

It was no surprise when the Leadtek WinFast GeForce 256 DDR Rev B took the crown for coolest running card. The extremely low temperature recorded by the Leadtek (7% lower than its closest competition) is no doubt due to the extremely large heatsink and fan found attached to the GPU with thermal grease and push pins. The heatsink on the card is unlike any other that we have seen on a stock GeForce card, be it DDR or SDR. Measuring 8.5 cm x 4.5 cm, the oversized heatsink seems to accomplish its cooling task better than any of the competing products. Heat transfer from the GPU to the heatsink surface is aided by the use of a dab of thermal paste, an aspect of the card that allows for full heat transfer to occur.

The other cards reviewed included the standard 4.5 cm x 4.5 cm heatsink and fan. The differences in cooling efficiencies here lie in the way that the heatsink and fan are attached to the GPU surface. With this in mind, it is no surprise that the ASUS V6800 came in second place as it uses thermal grease to attach the heatsink to the GPU. Commonly thought of as the most effective way to transfer heat from one surface to another, the thermal grease used allowed the core of the ASUS card to remain relatively cool. The Absolute Multimedia Outrageous 3D GeForce card came in at third place and also uses thermal grease to aid in heat transfer

Finally, the are the two other cards reviewed that use thermal glue to allow heat transfer from the GPU surface to the heatsink. Both Creative Labs and ELSA chose to use the glue and as a result heat transfer was less effective. In the case of the ELSA ERAZOR X^2, the temperature reached during the Quake III Arena run were higher than the cards using thermal grease. Unfortunately, temperature data from the Creative Labs 3D Blaster Annihilator Pro could not be taken, however the results are expected to be the same.

There is no question that temperature plays an important role in overclocking status, but transistor conductivity also plays a crucial role. Due to the fact that transistor conductivity will vary from chip to chip, we are forced to use temperature as a rating for overclocking potential. Just keep in mind that your results may vary.