High-quality sound was enabled in all tests. 32bit color was used in both tests.
I though long and hard about how to test the RAMsinks. Should I read the temperature difference between the standard heatsinks and RAMsinks and show the difference? I didn't think that would be precise as David Wood's
GeForce Heat article
concluded that each memory module produces varying degrees of heat.
Reading the temperature on the backside of the memory modules also proved fruitless as the readings would vary due to the fans that I have blowing directly upon the slots. I also wasn't sure about reading the temperatures of the heatsink itself. Would a higher temperature indicate the memory is running hotter and the heatsink is unable to dissipate the heat fast enough? Or would the hotter heatsink mean they are collecting all the heat while the memory modules remain cooler?
That hasn't left me many options on direct comparison between the standard and the RAMsinks other than the clock speed attainable by overclocking. The standard heatsinks allowed the memory to operate at 540MHz whereas the RAMsinks allow them to operate at 550MHz. At this speed I don't believe that heat was the limiting factor of operation of the 3.8ns memory (which are rated up to 526MHz), but rather the internal limitation of the memory modules.
I ran a battery of tests with different processor and memory speeds to illustrate the type of performance improvement is attainable from overclocking the GeForce3.
I'll begin with an examination of Evolva as it's DOT3 bump-mapping is an intensive process for graphics cards that actually perform the necessary calculations in lieu of being offloaded to the central processing unit. I also ran the test without bump-mapping to demonstrate the additional stress added by this feature.
The effects of bump-mapping add value, but as with many graphics quality enhancing features, it's employment reduces the performance of Evolva.
Notice the additional performance of the higher memory clock (settings 2 & 3) using the normal setting; the increased memory performance adds about half of the performance increase attained when both the graphics processor and memory are overclocked (settings 4 & 5). The same cannot be said about the test with bump-mapping enabled. DOT3 bump-mapping is processor intensive and as such the biggest proportion of the performance improvement arises when the graphics processor is overclocked.
Quake 3 Arena
Q3Bench beta 2.00 and the Four demo in Quake 3 version 1.31 were used for this series of tests. Trilinear filtering as well as high geometry and maximum texture detail were enabled to add additional stress the card. I also used anisotropic texture filtering for the ultimate stress test.
Once again, in the normal settings the increased memory results add 50% of the total increased attained by overclocking.
As opposed to antialising, where fill-rate is the all important performance factor, the above tests show that the graphics processor is the workhorse when anisotropic filtering is enabled. At both 1280x1024 and 1600x1200 an additional 10 frame per second were gained by the graphics processor and memory overclock as opposed to the results obtained by only overclocking the memory.
The higher anisotropic filtering results also have increased performance when the graphics processor is overclocked. The impact of various anisotropic filtering levels combined with the overclocked results are analyzed at my Quake 3 gaming resolution of 1600x1200.
In this careful examination of the performance results of the various filtering methods at a specific resolution, we see that the increased graphics processor speed is more of a performance booster than the increased memory speeds. For antialiasing higher memory speeds are the cornerstone of greater performance which is examined here.