When we posted about the revelation of the AMD Radeon RX 480, RX 470 and RX 460 graphics cards, we pointed out the odd Radeon RX 480 AOTS benchmark results that were presented by Raja Koduri, Senior Vice President and Chief Architect, Radeon Technologies Group.
The good news is we now understand how AMD derived those results. The bad news is it casts AMD in a rather bad light, and detracts from the great value proposition that the Radeon RX 480 brings to the table. But first, a quick recap of the Radeon RX 480 AOTS benchmark controversy…
Odd Radeon RX 480 AOTS Benchmark Results
In the final minutes of his presentation, Raja Koduri showed how two Radeon RX 480 graphics cards can beat the new NVIDIA GeForce GTX 1080 [Amazon] graphics card for far less money. Based on the sub-$500 price, the results are based on the 8 GB variant of the Radeon RX 480. Check it out in this short video clip :
In fact, if we extrapolate the results so that both cards have the same 98.7% GPU utilisation, the Radeon RX 480 would deliver a frame rate of 60.5 fps. That would make the Radeon RX 480 about 3% faster than the GeForce GTX 1080 [Amazon], which we know is just not possible…
How The AOTS Benchmark Was Conducted
Robert Hallock, Head of Global Technical Marketing, AMD, helped to clarify the controversial Radeon RX 480 AOTS results. First, let’s start with the technical details of the benchmark setup :
Testbed System Specifications
CPU : Intel Core i7-5930K
Motherboard : ASRock X99M Killer
RAM : 32 GB DDR4-2400
Operating System : Microsoft Windows 10 (64-bit)
AMD GPU Configuration : 2 x Radeon RX 480 @ PCIe 3.0 x 16 for each GPU
AMD Driver : 16.30-160525n-230356E
NVIDIA GPU Configuration : GeForce GTX 1080 Founders Edition
NVIDIA Driver : 368.19
Ashes of the Singularity Version v1.12.19928
Ashes of the Singularity Game Settings : Crazy Settings | 1080P | 8x MSAA | VSYNC OFF
Robert then revealed the actual AOTS benchmark results and his take on them, which we will post verbatim :
2 x Radeon RX 480 – 62.5 fps | Single Batch GPU Util: 51% | Med Batch GPU Util: 71.9 | Heavy Batch GPU Util: 92.3%
GTX 1080 – 58.7 fps | Single Batch GPU Util: 98.7%| Med Batch GPU Util: 97.9% | Heavy Batch GPU Util: 98.7%
The elephant in the room:
Ashes uses procedural generation based on a randomized seed at launch. The benchmark does look slightly different every time it is run. But that, many have noted, does not fully explain the quality difference people noticed.
At present the GTX 1080 is incorrectly executing the terrain shaders responsible for populating the environment with the appropriate amount of snow. The GTX 1080 is doing less work to render AOTS than it otherwise would if the shader were being run properly. Snow is somewhat flat and boring in color compared to shiny rocks, which gives the illusion that less is being rendered, but this is an incorrect interpretation of how the terrain shaders are functioning in this title.
The content being rendered by the RX 480 — the one with greater snow coverage in the side-by-side (the left in these images) — is the correct execution of the terrain shaders.
So, even with fudgy image quality on the GTX 1080 that could improve their performance a few percent, dual RX 480 still came out ahead.
As a parting note, I will mention we ran this test 10x prior to going on-stage to confirm the performance delta was accurate. Moving up to 1440p at the same settings maintains the same performance delta within +/-1%.
* Credit for this find goes to Acid 에이스. Thanks, mate!