LOSTCIRCUITS

Conclusion

The essence of running all benchmarks shown and a few dozen more is that there are enough applications that will show a benefit of using AGP8X, even if at this point the benefit is rather small. The only game we found that showed any difference, though, was Bioware's MDK2 and there, the difference was in the 1.5% range (208 vs 205 fps at 1280 x 1024 x 32 bpp)

Performance gain [in %] of AGP 8X over AGP 4X in professional graphics applications. The results show that AGP8X is a little better than its reputation, meaning that in high-end applications the average gain is somewhere between 2 and 3.6%.

The Pitfalls

Testing results are always based on the parameters that are changed and if the switches don't work, then any interpretation of results is pointless. This goes in particular for the lack of performance differentiation between AGP4X and AGP8X on the Granite Bay chipset. It is true that there is no difference between AGP4X and AGP8X except for DRV-08 but truth is that the difference is between AGP 2.0 and 3.0. We cannot make any further statements since it turned out that setting the BIOS from 4X to 1X did not show any differences either, not in 3DSMax nor in DRV-08. It is very easy to fall victim to dysfunctional BIOS switches but it appears as if with the ASUS P4G8X we have a classical example of such a scenario. If the switch doesn't work, we won't see a difference, it is as easy as that.

Again, that does not mean that there will be a difference if all the switches are working. In fact, our findings that there was no difference whatsoever in any other benchmarks even with hardware-forced AGP4X only reinforces our hypothesis of bandwidth vs. granularity as the major deciding factor for the ability to capitalize on AGP8X. It appears as if at this point nVidia's nForce2 chipset is the only player in the field that really can take advantage of AGP8X and that evaluations of AGP8X vs. 4X on other platforms may miss the point altogether.

Quadro4 980 XGL

Part of this review was dedicated to the Quadro4 XGL and whoever looked at the benchmarks knowing his or her own system performance will have gotten a kick out of these numbers. We don't want to go into listing 20 irrelevant competitors here but we still want to give some idea of the raw performance of this card in professional OpenGL applications in form of a single example: Advanced Visualizer or "AWadvs-04" that showed no difference regarding AGP settings.

A bit of apples and oranges but valid for illustration of the raw power of the Quadro4 980 XGL. The metric used is frames per second in AWadvs-04 taken from SPEC ViewPERF 6.1.2. and running on the Granite Bay chipset-based ASU P4G8X at 3.06 GHz. For the record, we were able to push the numbers up a bit further by overclocking the system to 150 MHz FSB or 3.45 GHz which gave us 492 fps.

Keep in mind that the competitors listed above are either outdated or consumer cards but somehow we need to put things into perspecive. We'll be looking into more professional graphics cards in the future.

Next Page:    => more =>