tl;dr Just set everything to max performance in CoreCtrl and you’re good to go.

Internet advice is usually of the form “X worked for me.” It’s extremely rare to get an explanation why it should work, a link to documentation, or actual measurements. After seeing a bunch of Internet advice to improve the performance of Elden Ring on Linux I decided to just try some of them and actually benchmark. The result was surprising, but comes with the usual caveats: my hardware, firmware, and software are probably different from yours, things change fast, and any of this might be obsolete before it’s published.

Specs

$ inxi --cpu --graphics
CPU:       Info: 6-Core model: Intel Core i7-8700 bits: 64 type: MT MCP cache: L2: 12 MiB
           Speed: 4508 MHz min/max: 800/4600 MHz Core speeds (MHz): 1: 4508 2: 4564 3: 4604 4: 4362 5: 4510 6: 4568 7: 4572
           8: 4415 9: 4338 10: 4401 11: 4402 12: 4401
Graphics:  Device-1: Advanced Micro Devices [AMD/ATI] Navi 14 [Radeon RX 5500/5500M / Pro 5500M] driver: amdgpu v: kernel
           Device-2: N/A type: USB driver: snd-usb-audio,uvcvideo
           Display: x11 server: X.Org 1.20.13 driver: loaded: amdgpu unloaded: fbdev,modesetting,vesa resolution:
           1: 3840x2160~60Hz 2: 1920x1080~60Hz
           OpenGL: renderer: AMD Radeon RX 5500 XT (NAVI14 DRM 3.42.0 5.15.32 LLVM 13.0.1) v: 4.6 Mesa 21.3.7

So an Intel i7-8700, AMD Radeon RX 5500, an extra “screen” which is really just an audio receiver, and the latest stable drivers.

Software

  • NixOS unstable
  • Steam stable
  • GNOME on Wayland
  • Various other software running in the background, like a browser, email client and IDE

Steam game settings

  • Compatibility set to Proton Experimental
  • Launch options set as mentioned below, always enabling Mango HUD for an FPS counter

Game graphics settings

  • Fullscreen
  • 1920×1080 resolution
  • Auto-detect best rendering settings off
  • All graphics settings at “high”

Methodology

  1. Load latest save
  2. While standing in the same spot, turn the camera around in all sorts of directions
  3. Note the lowest FPS which shows up in the Mango HUD in about 10 seconds

Results

  1. No extra options, just MANGOHUD=1 %command%: 35 FPS
  2. gamemoderun MANGOHUD=1 %command%: 33 FPS
  3. WINE_FULLSCREEN_FSR=1 MANGOHUD=1 %command%: 34 FPS
  4. AMD_VULKAN_ICD=RADV MANGOHUD=1 %command%: 33 FPS (no reference, sorry)
  5. ENABLE_VKBASALT=1 MANGOHUD=1 %command%: 34 FPS (no reference, sorry)
  6. VKD3D_CONFIG=no_upload_hvv,single_queue MANGOHUD=1 %command%: 31 FPS (no reference, sorry)

Doesn’t get much clearer than that: none of these make a significant difference with my setup.

Things I didn’t test:

  • After tweaking the CoreCtrl settings I saw no appreciable difference between X11 and Wayland performance, so I’ve only included Wayland measurements here. If you know of any reason why either of them should perform significantly better than the other for any specific settings, please let me know.
  • With a different FPS counter, since I assume its overhead is unnoticeable.
  • Disconnecting the receiver, since it’s necessary to get sound.
  • DXVK options won’t do anything since this isn’t a DX12 game.
  • Other desktop environments, since I don’t expect them to make significant difference.
  • PROTON_USE_SECCOMP is long obsolete.