AMD Mantle Vs. Direct X Battlefield 4 Multiplayer Benchmarks – Shows Massive Improvement Over SP
You might recall I have been ranting on and on about how Mantle will work wonders on Multiplayer gaming because the situation is so CPU bound. Well, I have finally gotten my hands on standardized Battlefield 4 Multiplayer Benchmarks with Mantle. Since Multiplayer is an extremely tricky thing to benchmark because the conditions change with every run, these benches are quite unique.
Battlefield 4 Multiplayer Benchmarks with AMD Mantle API Vs Direct X
The benches were given to us by Hardwareluxx. As always the preliminaries first. The specs of the test rig are as follows:
- Intel Core i7-3960X 3.3 GHz overclocked to 3.9 GHz
- ASUS P9X79 Deluxe
- ADATA XPG Gaming Series Low Voltage 4x 2 GB PC3-12800U CL 9-9-9-24
- ADATA S510 SSD 60GB
- Windows 8 Pro 64 bit
A complicated method was used to standardize the multiplayer benchmark. The map was Conquest Large and 64 players connected to it. The test rig was then manipulated by commandline after every 10 minutes going through every combination given below. The result was then extrapolated into an average using the frame response time to give the final output of FPS. Now without ado, the benchmarks, i have added the Single player benches as comparison to emphasize my earlier rants.
As you can see, Mantle API increases the frames per second more in Multiplayer than it does in Single Player. The reasoning behind this is simple. Single Player is almost always linear and relies mostly on the GPU (GPU Bound) relatively speaking. However in Multiplayer there are any given number of unscripted events taking place, not to mention physics calculations, so it puts more of a strain on the CPU. Because Mantle is brilliant at fixing the CPU-Bound situation you see a massive performance increase on the Multiplayer side of things.
You might have noticed that as the resolution increases the impact of Mantle decreases. Once again this is due to the face that as you increase the resolution, the scenario becomes increasingly GPU-Bound. The puzzles analogy works perfect here. If the CPU can only hand 5 puzzles of 10 arbitrary difficulty in one minute to the GPU, and the GPU can only process 1000 arbitrary difficulty in one minute, then it is being bottlenecked by the CPU = CPU Bound. Now there are two ways to remove the bottleneck. Increase the difficulty of the puzzle, say 5 puzzles of 200 Difficulty, which is analogous to increasing resolution and graphic settings. OR make the CPU hand out puzzles faster which is basically what the Mantle API does.
— Usman Pirzada (@usmanpirzada) February 16, 2014