Hercules 3D Prophet II GTS 64MB (7/9)

Antialiasing

Do 64MB of RAM offer any advantages over 32MB with antialiasing turned on? A valid question, seeing how performance is affected by memory speed and size with this feature, and one I feel is worth answering.

(If you have any questions about how anti-aliasing works, feel free to have a look at our review of the Hercules 3D Prophet II GTS 32.)

I ran the tests once under Quake3 (OpenGL) and once under DRAKAN (Direct 3D). The strange results under D3D are most likely due to some issues in NVIDIA's D3D driver implementation. If there is insufficient memory for the desired AA-level at a certain resolution or color depth, the driver automatically selects a lower level. Unfortunately it neither gives you control over this, nor does it indicate whether it is using the selected setting or one that is one or two levels lower.

AA-Quality Einstellungen unter D3D

I have strong reservations about the DRAKAN results, so I suggest taking them with a grain of salt!!! Although you can generally see where the driver decided to "switch gears" and go for the lower settings, we have no definite way of checking this!! This means these results aren't really very representative.

These test were run using version5.22 of NVIDIA's reference driver. Unfortunately, we had a problem with the 64MB version. The slider for the frequencies showed strange values and only offered settings between 0 and 100MHz. For this reason it was impossible to set the 3D Prophet II GTS 64 to run at the memory speed of 366MHz the final card will use instead of the set 386MHz of my prototype board. Even Powerstip couldn't change that, as the card crashed after a few seconds when "underclocked". All this means is that the 64's antialiasing results are higher than those of the final retail board are likely to be.

DRAKAN - NORMAL

This graph shows performance with antialiasing turned off.

DRAKAN - 2S AA

The fastest AA setting. The picture is only scaled vertically.

DRAKAN - 4S AA

4S AA. At this setting the complete scene is rendered at twice the resolution internally and then downsampled (2xFSAA), in this case with high-detail mipmaps.

As mentioned before, the "dynamic" driver makes testing in D3D rather problematic. You can see that these results are quite muddled. We would expect the 64MB card to have the advantage here, but the opposite seems to be the case.

Either there is a bug in the driver that only occurs with 64MB RAM, or the quality settings simply aren't set to the same level. For the sake of completeness I'll include the results here. The following Q3 test fortunately offer more reproducible picture.

Copyright: 31.05.2000 (Tanslation by Benjamin Kraft)-   RIVA Station 2000 - Lars Weinand
No Copy without Permission!

URL of this Article: www.rivastation.com/3dp2_64_e.htm - If you want to link to it, please use this URL! :-)