Here are my tests results for the FPS benchmark between Steam Snap and Steam DEB. I did not test Flatpak because it is still a mess. Some games did not work, others the performance was somewhat fixed at the refresh rate even though that was not set on the settings. Anyway, that one got skipped.
The settings were the following:
-
5 Tests for each game (10 for Shadow of the Tomb Raider since it had Native and Proton. Just out of curiosity)
-
Average FPS after 1 minute for each test. All tests involved either the same benchmark and settings or the same areas (if the game did not have a benchmark tool)
-
Snap version of Steam was average by recording the video, then taking note of the FPS counter and averaging that one.
-
DEB version of Steam was simply using MangoHUD and the stats had a reset just before starting the 1 minute benchmark. I also used the MangoHUD Logger to see the avg 1 minute results. On some game the time was a minute 1 second or a minute 2 seconds. On others it was exactly 1 minute. The exception was CS2 that the testing lasted 15 minutes and instead of 5 tests it was 3 tests. I also used the stats from the game to benchmark it.
-
Testing took MUCH longer because of the lack of MangoHUD on Snap version of steam (It only works on OpenGL. None of the games were executed in OpenGL. So you can imagine me watching each video and taking 10 minutes writing down values.)
-
Games that had DLSS used DLSS. Games that had RT used RT.
-
All games full graphics at 4K @ 144 using X11 (Not Wayland) and on Gnome 48 using Ubuntu 25.04 with Nvidia 570.86.16
-
A single video of the tests was not done because Steam Snap does not yet fully support MangoHUD on both modes, OpenGL and Vulkan. Once this happens again, I can do a side by side comparison.
-
CPU Governor (Power Mode) was on Balance (Not Performance) because I forgot to enable performance mode. Sorry.
-
GRUB for my PC had: mitigations=off nvidia-drm.modeset=1 ibt=off split_lock_detect=off split_lock_mitigate=0 nokaslr ignore_rlimit_data sched_migration_cost=512 pci=nommconf threadirqs umip=off clearcpuid=umip vdso=0 preempt=full irqbalance=128 usbcore.autosuspend=-1 audit=0
STEAM DEB
Metro Exodus - 151
Spiderman Miles Morales - 234
Read Dead Redemption 2 - 121
Shadow of the Tomb Raider (Native)- 181
Shadow of the Tomb Raider (Proton)- 146 (RT)
DOOM Eternal - 228
Dying Light 2 - 186
God of War - 162
CS2 - 191
Cyberpunk 2077 - 110
Hogwarts Legacy - 136
Days Gone - 181
STEAM SNAP
Metro Exodus - 153 (+2)
Spiderman Miles Morales - 234 (0)
Read Dead Redemption 2 - 121 (0)
Shadow of the Tomb Raider (Native) - 191 (+10)
Shadow of the Tomb Raider (Proton)- 156 (RT) (+10)
DOOM Eternal - 252 (+24)
Dying Light 2 - 185 (-1)
God of War - 165 (+3)
CS2 - 191 (0)
Cyberpunk 2077 - 109 (-1)
Hogwarts Legacy - 138 (+2)
Days Gone - 183 (+2)
The results show that, the dependencies included in the Snap version help games with their performance, so it is the opposite of what I was expecting which was that the DEB version would have better performance overall and because the Snap version was in some kind of container, it would have less. With that said, a difference of ~3 FPS or less just means they had very similar performance. There were only 3 cases that the games acted out in favor of the Snap version. For example Shadow of the Tom Raider on both modes had 10+ FPS than the Deb version. DOOM Eternal also had more than 20+ FPS for every tests I did. Why? I do not know.
I also found out the Snap version, at least on my hardware, had a higher 1% FPS and a higher 97%. Meaning, the lowest FPS was actually in the Deb version and the highest FPS was on the Snap version when doing the average. Only on 2 games did Snap have less FPS, which were Cyberpunk 2077 and Dying Light 2, but it was only by 1 FPS.
I rather do a live video streaming of all tests again once Snap supports, out of the box, MangoHUD for OpenGL, Vulkan and anything else in between to have a better visual tests of this.
I will of course use Snap in the future but my only reason not to use it right now is the MangoHUD. It has to support it completely or else I can’t use it for testing current settings or hardware versus future changes like a new ubuntu version, new driver version, new game update, new hardware change. And this is important for me to know if something happened between a diff and another so I can compare in the future. That is my only reason really.