BF4 AMD Mantle Video Card Performance Review Part 1

AMD's Mantle API has been with us for just over a month now, and we have strapped a variety of video cards to the test bench to see what real world differences are being delivered to gamers within Battlefield 4. We will compare D3D11, Mantle, on various GPUs, looking at highest playable settings, frame times, and discuss our experiences.


Performance Comparison Page 1

We are continuing to use the Siege of Shanghai map for our performance testing. Siege of Shanghai is presented as a large conquest map with up to 64 players able to be connected at one time. The map provides a good representation of all of the eye candy that can be found throughout the game, including devolution of the skyscraper, water effects, tight city streets, and of course plenty of tanks.

For our testing procedures, we searched for a server with at least 60 players in a match and where the ping was under 50ms for a server located within 600 miles of the underground bunker. Prior to launching each setting change, we would confirm that it is configured in a running server, exit the game, and then reload the map to confirm that the settings were correctly applied, as there are times that in-game setting changes to not take effect if you do not reload the map.

We then played within the server performing various tasks, including capturing flags, blowing up tanks, and causing general mayhem as part of the team that we were placed on. We recorded the frametimes using the built in utility and then applied a formula to extract the average FPS over the course of the play session.

Remember, we cannot use FRAPS to record frames per second over time since it does not support the Mantle API. Therefore, we have to use the game's built in frametime recording ability to record the frametimes, and from that, we can calculate the average FPS. This means instead of the normal framerate over time graphs, we have to format these results into bar graph comparisons showing the average FPS. You can learn how to use BF4's frametime recording ability from this blog.

All the data below are being shown at apples-to-apples settings. We chose a setting that was not bottlenecking either video card, but still provided enough workload that the run-through was GPU dependent, and not CPU dependent.

AMD Radeon R9 290X CrossFire vs. GeForce GTX 780 Ti SLI

Article Image

First, we know prices are out of place currently with AMD GPU based video cards. If we compared these video cards by current street price, the performance comparison would not make much sense. Therefore, we are comparing the cards to the traditional comparisons for performance and MSRP pricing.

In DirectX 11 mode, the pair of Radeon R9 290X cards in CrossFire performs similar to the pair of GeForce GTX 780 Ti cards in SLI. Both platforms were playable and turned in an average of 80FPS during the course of our run-throughs.

Enabling Mantle in this configuration added a significant amount of performance, boosting the average frame rate up to 91FPS. This represents a 12.3% performance increase from Mantle alone. During the course of playing with Mantle enabled, it was very common for the frame rate to run in the 100-110FPS range.

After seeing the great results with all image quality settings, other than resolution scale set to maximum values, we decided that we should try increasing that value to see how much more image quality that we could add without impacting the gameplay experience. We started with setting the screen resolution to 150 (which renders at 3840x2400 internally before downsampling to 2560x1600) to see how much performance would get traded. Shortly after making that change, we observed VRAM usage increase to the moon,specifically over 4.5GB used on the 4GB cards. The game was beyond unplayable at that resolution scale, so we tried a resolution scale of 125 that renders at 3200x2000 internally and got much better results.

Although the results were much improved, these were not at all impressive with performance averaging under 60FPS. Based upon our experiences with our personal systems running R9 290X CrossFire cards running in Eyefinity (rendering a higher number of pixels than the single screen at 125 texture resolution), we were expecting the frame rates to land in the 60-80FPS range under Mantle, but that was simply not the case. The comparison to our Eyefinity performance, and difficulty even running a resolution scale of 110, gives us the impression that the resolution scale feature is not working properly under Mantle. This is unfortunate as there is certainly enough hardware capability to increase the image quality, however that may not be possible until Mantle is better optimized.

AMD Radeon 290X vs. GeForce GTX 780 Ti

Article Image

In DirectX 11 mode, the NVIDIA GeForce GTX 780 Ti turned in a 13.4% performance increase over the Radeon R9 290X. The Radeon R9 290X did not seem to perform as well as we have recorded in DirectX 11 mode during prior evaluations. Once we turned Mantle on, the average frame rate increased to 58FPS (an 11.5% increase) bringing us back to the expected level of performance out of the card.