![[H]ardOCP](/images/blank.gif){kind=small}
- Date:
- Monday , December 07, 2009
- Author:
- Mark Warner
- Editor:
- Brent Justice
- Google +1
DiRT 2 Demo DX11 Performance and IQ Preview
What are some new DX11 features DiRT 2 Demo brings with it, and what do they look like? Will DX11 cause a large drain on performance? These topics are briefly previewed today with the DiRT 2 demo and a handful of video cards.
Performance in the DiRT 2 Demo
The DiRT 2 demo has only two environments, each with a single race event. There is a Stadium race in Baja, California and a rally race in Morocco, which is on the northern coast of Africa. The Baja race is a very short "loop", at about ninety seconds per lap. The Moroccan race is just shy of four minutes long, and is a point-to-point race, meaning the drivers race from the starting line to the finish line, which is a different location. There are no laps in the Moroccan race. We tested performance in each race event and found them to be identical. Thus, for our purposes, the Moroccan rally is the better choice for measuring performance, as it gives us a larger sample of performance.
This demo also has, built into the game, a benchmarking mode. In this mode, the game loads up eight artificial intelligence (AI) drivers and races from the starting line to the finish line of the Moroccan rally. It is different from most benchmarks in that it is not actually scripted. The benchmark is different every time you run it. Sometimes the drivers will crash, and sometimes they will not. Sometimes the "Player One" driver (the one that the camera follows) will win the race, and sometimes he will lose spectacularly. We measured performance with FRAPS using both the benchmark race and by manually doing it ourselves, and we found there to be no performance difference at all. So, for the purposes of this quick peek at performance in the DiRT 2 demo, we used the game’s built-in benchmark mode.
When the full game launches, we expect that there will be other more interesting race events to use for performance testing, so we do not anticipate using the built-in benchmark for the retail product. But for now, using the demo, it is the best option we have.
NOTE: In the graphs below, our target average framerate for this game is shown as a dotted orange line at the 40 FPS mark. It is OK for framerates to go above and below this line, but we found that an average framerate of 40 FPS is the sweet spot for this game.
DirectX 11 Comparisons
Since the most interesting thing, from our perspective, about this demo is its DirectX 11 support, we thought it would be a good place to begin testing. All of the following tests were performed with the highest available in-game settings. We used the 2560x1600 resolution for every test for this game. The level of AA used in each case is indicated in the graphs.
Radeon HD 5870 -- DirectX 9 vs. DirectX 11
In this test, we wanted to see what kind of performance hit we were taking with the Radeon HD 5870 for the graphical improvements when running this demo in DirectX 11 mode, as opposed to DirectX 9 mode (the default API when a DX11 card is not present.)
At 2560x1600 with 8X MSAA, we saw a performance decrease of about 9% going from DX9 to DX11. With 4X MSAA, the difference was 16%. So there is definitely a performance hit for the added features of DX11, but it is not overly severe.
Radeon HD 5850 -- DirectX 9 vs. DirectX 11
Again, this test is to measure the performance degradation from going to DX11 from DX9, this time using the Radeon HD 5850 at 4X MSAA.
Here, we are getting a performance hit of about 15% when going from DX9 to DX11. We ran the same test with 8X MSAA enabled, and the performance decrease was about 8%.
DirectX 11 -- Radeon HD 5870 vs. Radeon HD 5850
Here, we were trying to determine how much faster the Radeon HD 5870 is in DX11 mode than the HD 5850. We ran this test with 8X MSAA and 4X MSAA with similar results. The data for 8X MSAA is shown below.
The difference between these two video cards is less than 5% with 8X MSAA at 2560x1600. With 4X MSAA, the difference was about 6.5%.
DirectX 9 Comparisons
Obviously, the NVIDIA GeForce GTX 285 and GeForce GTX 275 are not designed for DX11. They are DX10 video cards. Oddly, the DiRT 2 demo does not support DX10. It supports DX11 or DX9 modes only. Thus, with the GeForce video cards in this preview, we were forced to run in DirectX 9. We also tested the Radeon HD 5870 and HD 5850 in DX9 mode, so for those two, we had to force the game to run the DX9 renderer. To do that, we edited the "hardware_settings_config.xml" file located in the "[..]\Documents\My Games\Dirt2\hardware_settings\" folder. In that configuration file, there is a field which reads "forcedx9=false". Simply change that option to "true", save the file, and the game will be in DX9 mode until you change it back.
All of the following tests were performed with the highest available in-game settings. We used the 2560x1600 resolution for every test for this game. The level of AA used in each case is indicated in the graphs.
DirectX 9 -- Radeon HD 5870 vs. GeForce GTX 285
We wanted to directly compare performance between the Radeon HD 5870 and GeForce GTX 285, so to make it a truly "apples to apples" scenario, we had to do it in DX9 mode for both video cards. This test is shown with DX9 mode enabled and 8X MSAA.
At 2560x1600, with 8X MSAA, 16X AF, and maximum in-game settings in DX9 mode, the Radeon HD 5870 proved to be 23% faster than the GeForce GTX 285. With 4X MSAA, the difference was 15%. It appears that the Radeon HD 5870 is a good deal more efficient with 8X MSAA at this high resolution.
DirectX 9 -- Radeon HD 5850 vs. GeForce GTX 275
This is a similar test, but with the less expensive ATI Radeon HD 5850 and the NVIDIA GeForce GTX 275. Data for 4X MSAA at 2560x1600 is shown.
Here, the Radeon HD 5850 was 20% faster than the GeForce GTX 275. At 8X MSAA, the Radeon HD 5850 was 37% faster than the GTX 275.
DirectX 9 -- GeForce GTX 285 vs. GeForce GTX 275
For our last comparison, we wanted to see how much advantage the GeForce GTX 285 has over the GeForce GTX 275 in this demo.
With 8X MSAA enabled at 2560x1600, the GeForce GTX 285 performed 19% faster than the GeForce GTX 275. When we tested it with 4X MSAA enabled instead, the GTX 285 came out only 12% faster than the GTX 275. The hindered AA performance at 2560x1600 with 8X MSAA is likely due to the slower and narrower bus on the GTX 275.