GTX 280 vs 8800 GTX - 18 Months of Innovation

3DMark06

3DMark06 is a benchmark created and sold by Futuremark. When using 3DMark06 I made sure to turn off AS and AA through the Nvidia control panel and I used all the default settings in 3DMark06.

That GTX 280 overall score is not very impressive. I don’t know why but I could never get 3DMark06 to give me a score over 15,000 like the other review sites. It is a tad frustrating but then again 3DMark is a benchmark and not a game (I use my video card to play games and not to “play” benchmarks), and the GTX 280’s game performance was much more normal. You can see that there was a small improvement in the Shader Model 2.0 and 3.0 scores which is expected. The CPU score went down but that always happens in 3DMark06. If you use a faster video card without changing your CPU at all your CPU score goes down.

Here are the FPS numbers for the 4 GPU tests. You can clearly see that Canyon Flight was the only test where the GTX 280 showed a significant improvement over the 8800 GTX. Watching all the tests, I got the feeling like my GTX 280 just was not trying very hard. The GTX 280 even turned out a lower FPS measurement than the 8800 GTX in the Deep Freeze test. Maybe my GTX 280 prefers to render balmy tropical environments.

3DMark Vantage

Happily my Vantage results were on par with tests run by other review sites. I ran these tests before I installed Nvidia’s Physx driver by the way. We will get to that bag of worms later on in the review. The GTX 280 overall score was 67% higher than the 8800 GTX and the GPU score was 91% higher. Good news for those types of guys that like to capture 3DMark world records I guess. Like I said before, if GPU speed goes up and CPU speed stays the same, CPU score goes down.

Here we have the results from the first 4 tests that Vantage runs. The Jane Nash test centers on a female spy (guess what her name is) that is escaping a watery enemy cave by jumping into a speed boat that transforms into a jet once it clears the cave. The boat/jet is emblazoned with the Sapphire logo by the way. I can’t wait to see this test run at 60 FPS at high quality someday. That would be awesome. I also wished that there was sound to go along with the whole thing. The GTX 280 trounced the 8800 GTX by an incredible 113%. That is the first measurement in this review (so far) that has doubled (over 100%) when compared to the 8800 GTX.

The New Calico test shows space fighters and their mothership gliding through space near an asteroid belt. Everything is peaceful and serene until the mothership launches three missiles at a highly populated planet on the other side of the belt. A huge area of the planet is vaporized in a huge explosion. Watching it made me a little angry because I don’t know who to side with emotionally. I don’t know who the bad guys are but I get the feeling that the aggressors are the evil race. Once again I would love to have audio to go with my visual tour de force, but this time it is more forgivable because there are no sounds in space. The GTX 280 rendered the New Calico massacre with 72% more frames than the 8800 GTX. 3DMark tests always take the same amount of time to run (usually between 60-90 seconds) but more or less frames are rendered during that set time frame.

The AI test and Physics test are (traditionally) CPU tests and they are both measured in operations per seconds. Please note that the AI test numbers have been divided by 100 so that I could keep it on this graph. If I had not done that the y axis of the graph when have been over 1,300 and you could not visually see the difference in the other bars on the graph. So the “real” values for the AI test are 1,282.7 OP/sec and 1,268.4 OP/sec. These tests were executed by the CPU which is why the GTX 280 values are lower than the 8800 GTX values.

This graph shows the 6 feature tests that Vantage runs.  Like the AI test from the last graph, I needed to divide the Texture Fill measurement by 100 to keep it on the graph. The GTX 280 out-performed the 8800 GTX in every test but it really sparkled in the Color Fill (57% better), POM (192% better) and Perlin Noise (145% better) tests. You can probably guess what most of these feature tests are testing, but let me define the POM and the Perlin tests for you. The POM test “renders a high resolution terrain height map using ray tracing in the pixel shader. Multiple lights cast shadows on the map.” The Perlin Noise test “stresses the computing power of your GPU by rendering multiple octaves of Perlin noise”. Perlin noise is something that is used to give digital creations a pseudo-random appearance by using visual details that are all the same size. It basically allows digital stuff to more closely resemble real stuff.