FarCry2′s Broken ATi HotFix Drivers!

Nvidia Focus Group member Rollo brought FarCry2′sATI hotfix driver issues to our attention and we decided to explore them with these new drivers that apparently included diminished IQ and yet increased performance.  To be fair, AMD has posted a serious disclaimer on their site regarding these beta drivers and we would now have to suggest that you would have to be desperate to use them instead of the regular Catalyst 8-10 drivers:

“Note: This driver is provided as is and is not supported by AMD. It has not completed full AMD testing, and is only recommended for users experiencing the particular issue described above.”

Let’s jump right into it.  First of all, our test system uses identical setups of completely updated Vista32 that are installed on identical HDs - and except for the video cards - both set-ups are completely identical.  Three sets of the very latest drivers are compared, as posted in our table:

First we maxed out all FarCry2 DX10 in-game settings – “ultra” - and when we used AF, we forced it in the respective control panels, although we later learned it controlled only by in-game settings.  We used the short Ranch run from the FarCry2built-in demo and each setting ran at least 3 times – taking care to make sure all 3 runs were very close – and then the results were averaged.  We can clearly see that the 280GTX outperforms the HD4870-1GB when the regular Cat 8-10 is used.  However, when compared to the Hotfix performance numbers, the gap narrows considerably – especially at higher MSAA, where the Radeon starts to pull ahead.

From these results, we can see that 16AF is basically “free” and also that the 280GTX takes a harder performance hit with increasingly higher levels of AA applied than the HD4870 does until at 8xAA the performance gap narrows with the regular 8-10 drivers.  However, with the hotfix drivers, the HD4870 catches up to the 280GTX at 2XAA.  But does this performance increase come with a penalty?  Oh yes, read on!

EDIT: We have learned that anisotropic filtering is set as part of overall render quality. Setting control panel anisotropic filtering will not affect image quality or performance; the in-game options controls AF. 

So let’s look at the same screen shot – first rendered by GTX280 with progressively higher filtering applied – noAA/AF, then noAA/16xAF, further 2xAA/16xAF and on to 4x AA/16xAF and finally at 8xAA/16xAF; and then we compared IQ with the 4870 1GB Radeon using the same settings with regular Cat 8.10.  Even though the scene slightly varies from image to image as the gun barrel moves up-and-down while the plants gently sway from side-to-side, we can see the filtering is evidently applied correctly.

     GTX280 (178.24) IQ compared with HD4870 (Cat. 8-10)

HD 4870 – 8xAA, 16xAF

There you have it.  Check out the clearly missing and sometimes transparent rocks and missing textures – and this shot is with maximum filtering! The proof is in the pictures, as they say.  Clearly the hot fix drivers – while improving raw frame rates - may well be sacrifice texturing and IQ.  And these textures are missing across all levels of filtering.  So, AMD needs to know about this.  Clearly it appears to be a driver bug that we hope does not make it into Cat 8-11, next month.   Finally, we would like to say that the benchmarks chart’s figures do not accurately represent real game play for the 4870/1GB.  Unlike its bigger brother, HD4870x2 which plays FarCry2and its short Ranch demo run smoothly, 4870/1GB is not at all smooth as it hitches and lurches in many instances, from frame-to-frame – even when its minimum is well above 30 FPS.   Geforce 280GTX users will also find their FarCry2game play far more smooth than HD4870/1GB.  One possible solution for Radeon owners that has been suggested is to cap the frame rates to 25 or 30 as is done in UT3 games to smooth its frame rate when it is uneven.   You could open the console and type: gfx_maxfps 25