[xarx]

I’m trying to make work the Fatal Racing game, which executes fluently with the highest settings on my notebook. But on my tablet Samsung T580 the game has FPS <10, even on the lowest settings.

I have executed the https://magicbox.imejl.sk/forums/topic/benchmark/ and https://www.complang.tuwien.ac.at/misc/doombench.html. The tablet score is ~2250, the notebook score is 1300. So the difference is not big.

On PC, I use D-Fend Reloaded. The game takes ~10% CPU on EmulationCore=Auto and CpuCycles=Auto. When I change to CpuCycles=Value, I have to set Cycles=~150000 in order the game runs fluently.

On the tablet, I have tried many different settings, but the game is still unplayable due to very small FPS. I have tried also CpuCycles=Max. When I set CpuCycles=Custom, the biggest performance is for ~35000 cycles, but the FPS is still very small.

Why is the performance difference between MagicBox and D-Fend Reloaded so big, even though the benchmark difference is relatively small. Even though the CPU power difference is not so substantial?

[admin]

Score 1300 and 2250 is big difference. I can try the game on my shield tablet with score 1100 and will see. And you compare two very different operating systems and cpu architectures. See from table how big difference is between nvidia cpu and snapdragon. And see how was alex able tune his note 4. It is all about cpu and how is utilized by system.

Edit : I run it on my tablet, but I dont know how to control car. Car wont move. Probably I don’t know how to setup it. But I see how cars are driving around me and all is fluent. Then I switched to svga and it is a bit slower but seems still well playable. I used sound blaster 2.0 and svn core.

[xarx]

I see that the difference is not “negligible” :-). But as the notebook score is better no more than twice, I would expect that FPS would be higher no more than twice, with the same game detail settings. That’s why I was asking why there is so big difference in FPS between the two systems.

Also, as the benchmark is based on Doom “real-clicks”, I would expect the score to be linear – two times lower score means two times faster machine (from the Doom perspective). As this is not the case for the Fatal Racing, does it mean that the PC version of DosBox is more optimised? For instance, that it is able to make better use of the GPU? Or that the PC version is able to execute pieces of code in the notebook CPU directly, while the Android version has to emulate everything?

[xarx]

Before, I had measured the Doom benchmark on my notebook somehow incorrectly. Now I ran it again, and the score is ~310. I am not aware of any changes in dosbox settings, maybe the notebook switched to the better (of the two) graphics cards automatically. But now the performance difference between the notebook and tablet is very substantial, which could explain everything.

Edit: That low tablet score I obtained because I had limited the CPU cycles to 50000. Now, when I set CPU cycles to Auto, I got the much better score. That was my fault, sorry.

[admin]

I think, magic dosbox has more optimizations then desktop dosbox. Big difference is caused by cpu settings. Cpu power plan is different on tablet and on desktop pc. Or nvidia shield tv (which has not battery and is still connected to power) and shield tablet which optimizes cpu usage in runtime.

As was said in benchmark thread, switching to governor gains more stable cpu speed. And in expert settings you can set cycles=max 120%

Btw, can you help me with controls in that game? How can I get car moving?

[acrosstheline1]

With the arrows.

[xarx]

The only keys you need are arrows, esc, enter and Y. I’ve got left/right at the left-hand side and up/down at the right-hand side.

I’m really curious whether you’ll be able to make that game playable fast enough. With most details off, the game is still on the edge of being playable, for me.

[admin]

Ok, I tried on Shield Tv. At vga max settings is perfectly fluent. At svga max settings is almoat playable, however I feel fps drop. I used tweaks : sound blaster 2.0 instead sb16 and i set cycles=max 120%

[Al ex]

@xarx the dynamic recompiler for Android is far more complicated due to the big differences between ARM and x86 cpus.
The graphics card has nothing to do with it. Dosbox is running 100% on the cpu. Exception: Glide (no Glide wrappers for Android though), and hardware scaling.

[admin]

It is not only about difference between arm and x86. M-HT did very good job in translation. I tried x86 cpus included in mobile phones and they give similar score. Nothing special.

[xarx]

Ok, I tried on Shield Tv. At vga max settings is perfectly fluent. At svga max settings is almoat playable, however I feel fps drop. I used tweaks : sound blaster 2.0 instead sb16 and i set cycles=max 120%

Probably I had my game configuration somehow corrupted. Before, the sound was fitful/jerky, now it is clean. And the game is now “almost” playable. But the problem wasn’t fixed by setting SB2.0, the game now works even with SB16.

However, my SM-T580 tablet seems to be not powerful enough for this game. The game is playable only with compromises – VGA, reduced screen-size, restricted look-ahead, many features off, and the FPS are still not excellent. I feel like 20 years ago when my PC wasn’t powerful enough to play the game in SVGA.

According to the Doom benchmark, your Shield TV is more than twice faster than my tablet. That’s why the game is playable for you, but not for me.

Anyway, thank you for your help.

[Author]

Posts