by Alexey Stepin , Yaroslav Lyssenko
02/20/2009 | 09:05 PM
The graphics division of Advanced Micro Devices sticks to their schedule and releases new Catalyst driver versions once a month, which is reflected in their numeric indexes. As we have already found out, the first release of this year, version 9.1, did in fact ensure a significant performance improvement in certain games. However, it was not free from a number of issues. In particular, the video post-processing was disabled in for some reason, and there were some problems with enabling anisotropic filtering using the corresponding options in the Catalyst Control Center.
Since the second month of 2009 is almost over, it was time for ATI to release their next Catalyst version 9.2. And they didn’t keep us waiting any longer: the new ATI Catalyst driver was just announced following the recent 182.06 driver release from Nvidia. Of course, we are primarily interested in the possible changes that Catalyst 9.2 brings to certain games, and here is what it promises us:
Besides that, the new ATI driver should have the following issues fixed:
They have also fixed a number of issues with the Catalyst Control Center that many users have been complaining about. Moreover, they have also resolved some video playback issues and eliminated the problems with HDMI devices connection. You can check out the complete list of fixes in this driver version on the official developer web-site.
Anyway, we are primarily interested in finding out how the new driver is going to behave in contemporary games. That is why we ran a number of performance tests of the new ATI Catalyst 9.2 driver in the games mentioned by ATI as well as in a few additional titles, where the new driver may prove better or worse than the old one.
We are going to check out the performance of the new ATI Catalyst driver version 9.2 using the following testbed:
Both versions of ATI Catalyst driver were configured to provide the highest possible quality of texture filtering with minimal effect from default software optimizations. We enabled transparent texture filtering. As a result, our driver settings looked as follows:
This time we used the following games for our performance tests:
We selected the highest possible level of detail in each game using standard tools provided by the game itself from the gaming menu. We didn’t use any non-standard settings, except a few situations, described separately later in the article.
For our tests we chose two most significant representatives of ATI Radeon HD 4800 family:
This choice was determined by our intention to study the peculiarities of the new driver version with a classical single-GPU architecture, as well as with a dual-processor solution based on CrossFireX technology.
We ran our tests in the following resolutions: 1280x1024, 1680x1050, 1920x1200 and 2560x1600. We added MSAA 4x antialiasing to the standard anisotropic filtering 16x in all tests. We enabled antialiasing from the game’s menu. If this was not possible, we forced it using the appropriate ATI Catalyst driver settings.
Performance was measured with the games’ own tools using the original demos. We measured not only the average speed, but also the minimum speed of the cards if possible. The only exception was Fallout 3 that has no built-in tools for performance readings. That is why we used Fraps utility version 2.9.8 to record our performance measurements in this game. In the latter case we ran the test three times and took the average of the three readings for the performance charts.
The only game in our today’s test session using OpenGL not just failed to benefit from the new driver, but even lost a bit of speed in some cases. And although the average performance didn’t drop by more than 3.5% in the very worst case, this is overall very alarming news.
ATI didn’t promise Catalyst 9.2 to bring any changes to Fallout 3 game and our tests proved that it was indeed so. Only in 2560x1600 the average performance of Radeon HD 4870 1GB increased by about 2.5%. In all other cases the difference between Catalyst 9.1 and 9.2 didn’t exceed 1.5%, which is totally within the measuring error for the manual method using Fraps.
Instead of the promised 20% boost we suddenly lost up to 8% of the average performance at least for the single-processor Radeon HD 4870 1GB! The minimal performance also suffered. As for Radeon HD 4870 X2, the situation here is completely different: the average performance rate remained almost the same, while the minimal numbers increased quite noticeably, especially in high resolutions. However, the general performance level in 1920x1200 and especially 2560x1600 remained too low for us to speak of their practical value. Unfortunately, we have to state that ATI didn’t keep their word here.
In F.E.A.R. 2, however, we should give due credit to the ATI Catalyst driver developer team: the incorrect functioning of the CrossFireX technology has been fully eliminated and Radeon HD 4870 X2 performance increased by 250-350%! This is another piece of evidence how vulnerable contemporary multi-GPU systems are and how greatly their performance depends on the quality of the software part of the graphics subsystem.
As for the single-processor cards, their performance remained the same and the tiny differences we detected can be considered a measuring error.
ATI promised that new driver release would improve the performance in this game, too. In this case, however, they kept their word, though partially. Radeon HD 4870 X2 did better that it was expected to: instead of the promised 5% increase, the average performance gained additional 8-12% depending on the screen resolution. And as for the single-GPU card owners, they shouldn’t hurry to update their drivers with the new catalyst 9.2: the performance will remain the same at best, or will even drop.
As usual, trying to minimize the CPU influence in 3DMark Vantage we used the Extreme profile (1920x1200, 4x FSAA and anisotropic filtering).
Just like Nvidia, ATI didn’t promise any performance improvement in 3DMark Vantage with the new Catalyst version, however, we did detect a slight increase in Radeon HD 4870 X2 speed. However, judging by the individual benchmark results, it only occurred in 1920x1200 in Test 1. All other readings didn’t change, and the total score for Radeon HD 4870 1GB got even lower than with Catalyst 9.1.
I can’t say that our today’s driver express-test refuted ATI’s official promises of performance improvements with the new Catalyst 9.2; however, they did in fact come true only partially. As usual, we summed up all the results in two tables, each showing the performance gain or drop in percents:
First of all, they didn’t deliver on the promise of 20% performance gain in Crysis Warhead. The owners of dual-GPU graphics cards have got 2-3% improvement at best, while single-chip Radeon HD even slowed down quite noticeably. The second game mentioned in the release notes for the new Catalyst 9.2 driver, World in Conflict, the actual gain turned out higher than had been promised and reached up to 12% depending o the screen resolution. Unfortunately, it was again valid only for dual-processor Radeon HD models: 4870 X2 and 4850 X2.
Other benchmarks results confirmed our suppositions about ATI driver developers focusing their efforts on optimizing the dual-GPU solutions performance in the first place and adding software CrossFireX support for the new games. The results obtained in F.E.A.R. 2: Project Origin are excellent proof of that. ATI’s desire to provide fully-functional CrossFireX support is definitely a great thing; however, we find this approach to driver development a little one-sided. Moreover, contemporary multi-GPU concept once again showed that extreme dependence on the software support is in fact its major weakness.
The owners of single-GPU Radeon HD 4800 will not gain much in games by switching to Catalyst 9.2 from version 9.1. In some cases, they will even end up with lower performance numbers. Nevertheless, we would still recommend considering this release for one simple reason: it is free from a lot of issues and bugs that are not directly connected with the gaming performance, but have a steady negative effect on the driver stability, compatibility or image quality in games or during video playback.