Performance under ARM vs x86

[DerekCZ]

I have temporarily switched from my 64-bit, x86 laptop with a dual-core, 1.5GHz Celeron to the ASUS Tinker Board single board computer /w a 32-bit, quad 1.5GHz Rockchip CPU, the exactly same one as used in some models of the Chromebook.
I have, however, experienced much worse performance when playing on large maps - something that the Celeron had no problems dealing with.
So, I'd like to ask, if it may be either the ARM architecture or the 32-bit CPU slowing the game down.
I also have to note that now I have 2GB of RAM, as opposed to 8GB in the laptop, but I've never seen OTTD saturating the 2GB on either machine.
---
OT: I really can't express enough how impressed I am with the compatibility of this game over all the systems and platforms it can run on. I've never seen a game that would allow for crossplay multiplayer between Android, Windows, Linux Mint, MacOS and even the Armbian, all at the same time with no issues

[peter1138]

Yeah, no surprises. ARM is cheap, efficient, and slow.

[Eddi]

you can't really compare the "GHz" values of different CPU architectures. first, they re more or less made up values anyway, and the CPU architectures differ vastly in what they can do in one cycle

what openttd needs is mostly single-core integer operations performance, but also memory bandwidth or display performance may be limiting factors

[DerekCZ]

. . .

Display performance of the Tinker Board is superb, even when compared to the laptop - it can stream full HD, 60FPS videos from youtube and can play 4K using its default media player. But it might be the memory, given the slower bus speed and most likely longer latency, now that I think about it...

EDIT: oh, by the way, can OTTD take advantage of multiple cores?

[Gwyd]

No. Except when saving, I think.

[peter1138]

it can stream full HD, 60FPS videos from youtube and can play 4K using its default media player

By using a dedicated hardware video decoding engine that has zero performance benefit for gaming.

[Eddi]

Display performance is relatively easy to measure:

1. make an empty map, not too big
2. maximise the game window
3. let the game run in fast forward for a minute
4. note the amount of months passed
5. restart the same map, this time make the window as small as possible
6. run the game again in fast forward for the same time
7. compare the amount of months passed

for some extra data points, you can try disabling "full animation" or use different blitters from the command line

[jfs]

Maybe also worth mentioning that there are blitters optimized for SSE2 and SSE4 (x86), but none optimized for any ARM SIMD instruction sets.