Performance in Synthetic Benchmarks
The total score of GeForce 8600 GTS is quite high: it is close to that of Radeon X1900 XT, which used to be one of the best High-End solutions. However, we are talking about the 1024x768 resolution here, which doesn’t load the graphics accelerator enough to estimate the advantages and disadvantages of its architecture in full. We should discuss the results of individual tests in “eye candy” mode, as they will give us a much better idea.
In Game 1 test GeForce 8600 GTS outperforms Radeon X1950 Pro only in 1280x1024. In 1600x1200 these two cards level out in performance: performance of 16 TMUs of Nvidia’s newcomer working at 675MHz perform comparably to 12 TMUs of the Radeon X1950 Pro working at 575MHz. As for the GeForce 7950 GT with 24 TMUs and Radeon X1900 XT with 16 TMUs, our hero cannot compete with them at all, at least not in the benchmark where fillrate is crucial for the overall performance score.
Game 2 test is far not as large-scale as the first one, it serves mainly to demonstrate the GPU performance when working with geometry and scene lighting. Unified architecture of GeForce 8600 GTS proved quite efficient here. However, the card still yielded to Radeon X1900 XT and fell behind GeForce 7950 GT and Radeon X1950 Pro in resolutions above 1280x1024. We would also like to stress that G84 slows down as the resolution increases, despite the peculiarities of the Game 2 test from 3DMark05 suite. Performance of texturing and rastering units of the new graphics processor are still a step backwards compared with the previous-generation solutions from AMD as well as from Nvidia themselves.
The results of Game 3 test prove the evident once again. They reveal very vividly the drawbacks of the unified architecture with few execution units as well as the problems created by relatively low-performance TMUs. Of course, the new Nvidia solution outperforms GeForce 7900 GS, but it cannot get beyond the performance level of Radeon X1950 Pro, especially in high screen resolutions.