>5s delay when starting each arcade game

Ashpool

@hsamuels Whence selecting the drive, you've choosen the "boot" partition (not the ext4-/other one (the one windows won't recognize)), haven't enabled read allocated partitions only and the filesize of the resulting image is ~= to the size of the SD-Card? A "yes" on all three should result in a valid/useable image.

hsamuels

@Ashpool I believe the answer is yes for all three.

No matter, I'm going to start on a new SD from scratch. I've done it enough recently that it's a pretty smooth process!

hsamuels

I rebuilt the system on another SD card (actually, the one I was using on a different system that worked perfectly) and it still takes 10.5s for the game to load.

Ashpool

@hsamuels Just out of curiosity(!) -> if you have a spareable USB stick at hand [1], could you try the "retropie-mount" method?

1: In my experience with 32/64GB SanDisk and some "NoNames": Sometimes USB2 Sticks are performing faster on an USB2 Port compared to USB3 sticks from the same Brand/Label/Size (though those are faster on an USB3 Port) - so, if possible, it may be worth to try out both variants.

P.S.: I am now puzzled what normal/expected loading times would/should be -> Just taken some rough meassurements (subsequent starts, not initial 1st one) on my old Pi3B (no plus) official PSU/retropie build from official 4.8 and last updated maybe +-5 month ago on a few roms (run from usb-stick) in FBNeo (.7z) and lr-mame 2003 (.zip) (plus is my choice, but for the tests I've used pure 2k3) -> first time is from blank screen after starting rom to appearance of the "press key to configure"-OSD-box, second time is from thereon to the dissapearance of said box/rom starting.
For lr-mame2003 the times where ~ +-5.5s/7.5-8s (total ~12-14s) and for FBNeo ~ 5.5-6s/5-6s (total around 10-12s)

hsamuels

@mitu Can you suggest any debug things to try? For example, I could put an echo statement in runcommand-onstart.sh. What happens between the game select button being pushed and the launch menu appearing?

mitu

@hsamuels Try running the command to start the emulator outside EmulationStation and see how long it takes to start the actual game.

/opt/retropie/emulators/retroarch/bin/retroarch -L /opt/retropie/libretrocores/lr-mame2003/mame2003_libretro.so --config /opt/retropie/configs/arcade/retroarch.cfg "/home/pi/RetroPie/roms/arcade/1941.zip" --verbose --appendconfig /dev/shm/retroarch.cfg

hsamuels

@Ashpool

For lr-mame2003 the times where ~ +-5.5s/7.5-8s (total ~12-14s) and for FBNeo ~ 5.5-6s/5-6s (total around 10-12s)

It looks like you are seeing the same 5.5s between pressing the game select button and the launch menu dialog that I am.

I didn't measure the time between closing the launch dialog and the first signs of game starting. Rather, I disabled the launch menu dialog, and measured the total time between pressing the game start button and the first signs of life.

Either way, it seems much longer than what I had from a 2019 build on a Retropie 3 (not B). Unfortunately, I didn't measure that time. But the current launch time was long enough that I thought the thing was broken. And my son said the same thing.

In any case, we now have two users experiencing 5.5s from game select to the launch menu. Would anybody else care to time their systems and share the results, along with the SW and HW versions?

hsamuels

@mitu said in >5s delay when starting each arcade game:

@hsamuels Try running the command to start the emulator outside EmulationStation and see how long it takes to start the actual game.
/opt/retropie/emulators/retroarch/bin/retroarch -L /opt/retropie/libretrocores/lr-mame2003/mame2003_libretro.so --config /opt/retropie/configs/arcade/retroarch.cfg "/home/pi/RetroPie/roms/arcade/1941.zip" --verbose --appendconfig /dev/shm/retroarch.cfg

Thank you, @mitu. I tried a couple of different games, each one measured several times. They all clocked in at about 3.8s from hitting enter to first sign of game life (no launch dialog). This feels much closer to what I remember from my earlier build.

mitu

@hsamuels said in >5s delay when starting each arcade game:

Either way, it seems much longer than what I had from a 2019 build on a Retropie 3 (not B). Unfortunately, I didn't measure that time. But the current launch time was long enough that I thought the thing was broken. And my son said the same thing.

This is before support for Pi4 was introduced - there have been quite some changes since then. Looking at the runcommand.sh times, 5s seems about the time it takes to:

analyze the current resolution and change it if needed
set-up Retroarch's temporary config
start the emulator
exit the emulator.
cleanup-up
Now, it's probably less than 1 sec it takes ES to close down and invoke the runcommand , but I can see that adding up.

Looking at the a breakdown on times, I think the most time is spent running modetest to detect the current and available resolutions - used for Pi4 or other kms based systems. This takes less time on a kms system, but on a Pi3 that's not the case.

Short of maybe removing modetest or maybe put tvservice before it - to detect the current resolution - I don't see a considerable room for improvement here. Replacing modetest with something lighter is on my TODO list and maybe will be done as part of the Pi5 support additions.

hsamuels

@mitu Thank you!

It's good to know that I'm not just doing something wrong.

Does modetest need to check the current and available resolution each time a game starts up? Can it happen once at power-up?

hsamuels

I'm back on the original system, which now sports a Pi4. It has LED buttons, and a pacdrive tells which buttons to light up for each game. This happens in the last line of runcommand-onstart.sh.

It takes 3s between the game select button press and the LED update. But slightly faster than my Rpi3 system, about 8s, to get to the joystick message from the game select button.

mitu

@hsamuels said in >5s delay when starting each arcade game:

Does modetest need to check the current and available resolution each time a game starts up? Can it happen once at power-up?

Not sure if you still have the test system around, but we added a modification to runcommand to skip the modeset invocation on systems without KMS (i.e. Pi3 and olders). Give it a try to see if the loading speed is down to the previous timings from the older versions.