Bunch of game performance tested on rk3588

[JSaurusRex]

running on an orange pi 5 with an rx 470 4gb running over pcie gen 2 x1. All tests should be mainly cpu limited

dnr = did not run (it crashed)
empty, didn't test
Steam means it requires the steam runtime, or at least that I ran it with the steam runtime.

Note: Fex is running with pretty unsafe settings for extra speed, didn't cause any instability anywhere but ultrakill where it generally lasted half a level vs box64's 2 levels. Box64 is generally running with the highest settings I could get away with, especially ultrakill is running with very aggressive settings.

I was testing on 2ef63f2 (Box64 2) but whilst testing 388d9c8 (Box64 1) released so I redid some tests with it.
FEX: 2509.1-1

Game	Engine	Steam	Free	Proton	Native	Box64 1	Box64 2	Box64 0.3.8	Box64 0.3.6	FEX	Details
Toribash	custom	yes	yes	no	-		62			62
Sword with Sauce	UE4	yes	no	yes	-		40			70
Tear Down	custom	yes	no	yes	-		80			117	Details performance gap is slightly wider here than shown i believe
DDNet	custom	no	yes	no	1300		dnr	dnr	dnr	1150	Details it kept crashing saying something about notifications
SuperTuxKart	custom	no	yes	no	166	128	128	dnr	dnr	137
Crysis Demo	CryEngine	no	yes	yes	-	16				20
Tesseract	custom	no	yes	yes	78		69			68
Sauerbraten	Cube 2	no	yes	no	370	330	350	320	327	dnr	Details 350 to 330 performance drop i think is just noise i think. Either way great performance and great stability from my short testing
UltraKill Demo	Unity	no	yes	no	-	135	85	77	77	150	Details very aggressive settings on box64 to match FEX, mainly ignoring the unity player safety flags. 33 ~ 50% performance hit without. I'd take the outlier 135 with a grain of salt. Also second level is very heavy on both, might be optimizable.
DogWalk	Godot	no	yes	no	50	23	20	20	dnr	45
Ravenfield Demo	Unity	no	yes	no	-	44	44	35	35	60	Details looking through the scope heavily impacts fps. Not sure if bug or just my setup. Don't remember this behaviour on x86 even on weaker hardware
Xonotic	custom	no	yes	no	240	166	dnr	dnr	dnr	164	Details crashes whilst loading into first level of the campain
MineTest	custom	no	yes	yes	50		33			30
Pseudoregalia	UE4	no	no	yes	-	34	34			50	Details In starting area, usually its worse, 20 ~ 30 fps vs 50 ~ 70

Overall I think unity games perform quite well. I ran all unity games here with all safety features off.. without probably 30% performance loss. UE4 games and Godot I think need some looking at, probably quite a bit of performance left over there.

[ptitSeb]

Thanks. That's quite some benchmarks already.

I need to check the games that dnr first. Strange that Xonotic was not working, it was working ealier, that's odd. Not brilliant overall for box64, looks like there is still a lot of work to do.

[ptitSeb]

I should have fixed the issue with Xonotic

[JSaurusRex]

Xonotic now works!!
I've added the numbers to the table as well

[ptitSeb]

Meh, disapointing fps tho.

[JSaurusRex]

I updated the fps numbers, I think I wasn't running it with the right settings. Honestly the amount of settings to change makes this a bit inconsistent. I try to run with either the most aggressive settings i can, or if thats not possible i use box64_profile=fastest.

[ptitSeb]

Thanks for the update.

Yes, I know there are lot of settings that can be adjusted. I want to leave a maximum of choice to advanced user to play with. The idea behind BOX64_PROFILE was to simplify this, as an quick way to check things.. The initial idea was to have huge database of game with optimium settings in the default box64.box64rc file, idealy with user-submited settings... but in reality, very few people submited new game profile...

[ptitSeb]

I would also suggest you create profile for your test game in ~/.box64rc so you can just run stuffs and settings gets automatically applied (if you don't do that already).

[JSaurusRex]

some more games

Game	Engine	Free	Proton	Box64	Fex	Details
Ghost runner demo	UE4	yes	yes	23	dnr	Details I might have gotten it to work with fex but its a pain in the ass, however the fact that performance of this AAA looking game is so close to pseudoregalia an N64 like game I think means there is something going on with this engine
Portal 2	Source	no	no	53	70	Details I did not create any kind of custom profile for this in .box64rc.. well i did but when i ran it it crashed my system. So this is without optimization flags set
Devil Daggers	custom	yes	no	477	474
Anger Foot Demo	Unity	yes	no	39	58	Details I think FEX has regressed in performance here, maybe there is a tweak Im missing but I swear i used to get 70~100 fps in the same scenario
Crumble Demo	Unity	yes	no	22	64	Details the last section of the first level has a whole lot of physics objects, performance was measured there during the introduction cutscene. Its a huge difference in playability.
Doom 2016	Id tech 6	no, but demo is	yes	46	62

I found an Unreal engine 4 game thats free, surprised to see it run after FEX failed.
Also all demos but ghost runner demo I found on itch.io

[ptitSeb]

Mmmm, pretty sure Doom 2016 runs fine with box64 (even Doom Eternal works, and even better)

[JSaurusRex]

Could be my setup but it would be odd for it to only work with FEX. I only have this setup to test it on tho

[ptitSeb]

Is It the version from Steam or the one from GoG? I only have the Steam one, I can retry tomorrow, in case there is a regression (but I doubt it)

[JSaurusRex]

Steam, running with vulkan. I think I tried opengl a while back as well

[ptitSeb]

Yeah, that's the version I also run. It runs fine (but the performances, somehow, is much better with Doom Eternal)

[JSaurusRex]

I had to upgrade from my i5 2400 to an i5 4460 to run Doom eternal I remember. Either is uses newer instructions or is just generally heavier all around. Odd that its faster..
I think I can also remember seeing Doom 2016 running in the release videos, so I'm surprised it doesn't run. Not sure if its on my side where it would be. Chromium also doesn't work with gpu acceleration but aside from that I cant think of anything that would impact it from running. Especially since FEX ran