The first thing I needed to do was to acquire a thermometer of sorts, so I went to one of the guys on my Engineering team, and borrowed a Fluke Digital Thermometer. The more difficult task was, of course, in trying to get the thermometer leads as close to the core as possible. Furthermore, in order to be fair, one would have to attempt to put the leads in the same location. I did take a screenshot of my Fluke Digital Thermometer, but it was severely overexposed. Here’s a photo straight from their website:

Fairly cheesy, but I think you get the idea.

• Windows 2000 SP2
• FIC AD11 motherboard
• 512 MB PC2100 DDR
• AMD Athlon 1.368 (144x9.5) GHz. CPU
• VisionTek GeForce3
• nVidia 21.81 Detonator drivers
• RivaTuner tweak application
• AMD AGP driver 5.22
• VIA 4-in-1 4.33
• Quake3 v1.30 1024x768x32 Max-HQ (unless otherwise noted)
• 3DMark2001 1024x768x32
• Serious Sam v1.04
• DronezMark
• Giants v1.4

I experimented with the GeForce3 to determine the highest, most stable core/memory settings possible. I ended up settling on 250 MHz (core) and 560 MHz. (memory). I’ll have more to say about this later. The basic procedure used was as follows: I used RivaTuner to overclock the GeForce3, allowing sufficient time for the temperature to settle before running the benchmarks. Even when you’re not running any 3D games/software, the temperature will still rise. This will be noted in the graphs as “No Load.” Once I recorded the temperature, I then fired up the benchmarking software. In the case of Quake3, I used a looping script to allow enough time for the temperature to settle. As for 3DMark2001, this benchmark takes long enough for the temperature to stabilize.

 Let’s take a look at Quake3.

I was rather shocked by the results, so I proceeded to redo the tests several times. Each time, I achieved fairly identical scores. No matter how you slice it, there's a massive difference between the 2 solutions. As you can see, a setting of 250/560 will give you a differential of 8.2 degrees under load between the 2!

 Let’s now take a look at 3DMark2001

Once again, we see similar numbers. This time, the differential under load is 7.7 degrees. Pretty awesome so far. Now, let's see how this translates in real-world applications.

Next Page: Benchmarks