I installed a Powerlogix 250/250/1M card in the 8500 at work a couple of months ago. I couldn't wait to see how well this card did (that's probably why I enjoy visiting your Web site so much) so I ran all sorts of benchmark tests between the different Macs at work and was very impressed with the performance boost. By the way, I ended up running the card at 312/208/1M.

After reading about the 220/110 card, I decided to see how our card fared at 312/104/.5M. I ran a Ray Dream Studio benchmark (the ol' BareFeats Countach one) and found that the new .5M configuration was only slightly slower. I then decided to check it out with Photoshop, so I looked around for a CPU intensive Filter and found that the 'Stained Glass' Texture really worked the CPU. I loaded the Mallorca sample picture and set the Stained Glass filter at a CellSize/BorderThickness/LightIntensity of 3/1/5. At 312/208/1M it completed the operation in 27.5 s but when I set it at 312/104/.5, the time ballooned up to 90.3 s. I was shocked. I resized the image to several smaller sizes and once again performed the filter operation on them using both card configurations. Here are the results:

It seems to me that the reason the ratios on the first two image sizes are so large is that the image probably completely fits in the 1M L2 cache but not in the .5M. If you figure each pixel takes 3 bytes (24 bit color), then the raw image sizes are 609957, 502500, 406350, 321600, 245700, and 180900 bytes.

By the way, boosting the cache speed on the .5M config up to 208 had only a minor effect on the results. The cache size definitely made the difference.