Testing F1 2017 Linux port on AMD GPUs

You have likely read Liam’s thoughts and port review of F1 2017. I am going to complement his work by bringing you some numbers from the AMD side of things with my RX 580 and R7 370. So, hold onto your hats for we are benchmarking now!

Before we dive into the numbers let’s take a look at the system requirements and the hardware we will be testing the game on. The game expects an Intel Core i3-3225 3.3GHz processor minimum and recommends an Intel Core i5-6600K processor. RAM requirements are a minimum of 4GB and a recommended 8GB and for the GPU the game wants a 3rd generation AMD GCN GPU or a GTX 680 2GB minimum and recommends a GTX 1070. To open that “3rd Gen AMD GCN GPU” thing up a bit, that specifically means the R9 285, R9 380, R9 380X and the Fury cards. AMD VEGA is not supported but the Polaris cards (4xx and 5xx series) presumably are. For AMD the Mesa 17.2.2 version is required.

I will be testing the game on my main gaming rig, which has a Ryzen R7 1700 @ 3.7GHz, 16GB of 2133MHz RAM and an Asus ROG Strix RX 580 8GB GPU. I will also quickly touch on the GCN 1.0 support with the older Asus Strix R7 370 card. All testing was done on Arch (Antergos) using kernel 4.13.9 and Mesa 17.2.4. The CPU scaling governor was set to “performance” as recommended by Feral Interactive.

For the sake of simplicity testing was limited to 1080p resolution without anti-aliasing and/or anisotropic filtering. The numbers were acquired using the built-in benchmark with the default settings. Note that the built-in benchmark is not 100% accurate and changes slightly from one run to another.

The benchmark yielded the following numbers:

As you can see, the game does not scale all that much going from the Ultra Low preset to the Medium preset. The framerates do change a bit but the difference is not particularly huge. The largest performance drop comes from switching the graphics preset from High to Ultra at which point the minimum framerate dropped below 60 FPS for the first time. However, as far as I am concerned the game was still fully playable even on Ultra settings and maintained an average framerate of about 60 FPS. If you asked me, Feral could have put the RX 580 in the recommended GPU section without any complaints.

During benchmarking I also monitored CPU utilization via htop to see how much CPU horsepower the game was utilizing. The CPU usage was at its highest during the Ultra Low benchmark run and utilized up to 4 cores worth of CPU time. When the quality preset was increased the game began utilizing less and less CPU, presumably because of an increasing GPU bottleneck. On Ultra settings the game was only taking advantage of around 3 cores. The takeaway here is that games like F1 2017 can take advantage of up to 4 CPU cores, so owners of dual-core and quad-core CPUs might want to start considering moving to CPUs with higher core counts, particularly if they intend to run more demanding background tasks, such as streaming software, while gaming.

I also decided to take a look at how big of a difference the CPU scaling governor made, since I was initially feeling a bit lazy about switching the CPU to performance mode. The Ultra Low graphics preset was used here to emphasize the load on the CPU. The difference was as follows:

As you can see, switching from the default CPU scaling governor to “performance” yielded a tangible performance boost. The minimum framerate improved from 70 FPS to 78 FPS and the averages jumped from 93 FPS to 101 FPS. While these numbers might be somewhat insignificant on a 60Hz screen, they will be quite visible on high refresh-rate monitors and in addition give a little bit of extra headroom for potential frame drops. The game was fully playable even without setting the “performance” governor though, so if you are indeed feeling lazy about changing governors you can safely ignore it and still get a decent gaming experience.

Finally, I decided to take a look at the game on a GPU that is well below the hardware requirements, namely the Strix R7 370. Despite being a 300-series card the R7 370 is actually a first-generation GCN card, essentially a rebrand of the R9 270 which itself was a rebrand of the HD 7850. This card by default does not even have a functioning Vulkan driver, since AMDGPU is required for RADV to work. I manually enabled AMDGPU GCN 1.0 support and tried the game. It did start but when I started the benchmark I was greeted with the following message:

So yeah, apparently GCN 1.0 support in AMDGPU is still in such an experimental state that F1 2017 refuses to work properly. That doesn’t surprise me much, since the only Vulkan game I’ve managed to get working on the GPU is vkQuake. If you have a GCN 1.0 card like the R7 370, I would say don’t even bother with F1 2017. Just to reiterate, F1 2017 only uses Vulkan and there is no OpenGL backend.

So, all in all, if you have a system that falls within the recommended system specifications, F1 2017 runs quite fine and considering how well the game runs on my RX 580, I have no doubt the game would be playable on the slower 3rd generation GCN cards, such as the R9 285. Just make sure your GPU is actually one of the supported GPUs, since the GCN 1.0 and 1.1 (2nd generation) cards can have Vulkan driver issues that prevent you from playing the game entirely.