FinalBurn Neo vs MAME on RetroArch (Windows 10)

Hi everyone, still learning my way through RetroArch and its various cores, hope you don’t mind my very basic-level questions.

Premise: for arcade games, I read it’s recommended to go FinalBurn Neo first, then MAME if whatever ROM doesn’t work with FBN, finally MAME 2003 Plus as a last resort (I found out Battletoads still doesn’t work with anything else).

With all that said, I noticed how some CAPCOM titles in the FBN ROMset - although working pretty much fine overall - exhibit little graphical glitches/impefections. Couple of examples: Wonder 3 - during Roosters’ attract mode - shows a character incorrectly, while The King of Dragons suffers from some random flickering in the lower part of the screen. What puzzles me is, if I change said games’ core association from FBN to MAME (withoud doing anything else, so simply launching a ROM from the FBN set with MAME) the glitches disappear.

Is this a common thing, or I’m doing something wrong/missing something? Are CAPCOM arcade games supposed to work best on MAME maybe?

Thanks in advance! ^_-

1 Like

The issue with all these arcade roms and emulators is that they get changed. You probably have an older rom of Battletoads that won’t work in newer versions of MAME. And maybe these glitches you talk about in FBA can be fixed if you use the correct versions of roms FBA expects. Maybe, i’m not sure. Supposedly, CPS1 emulation in FBA should be perfect.

I use the latest roms and the latest MAME core (0.212) or standalone and play everything there. But i do keep a slightly older Neo-Geo set to play with the older FBA core, because it’s the only way to play the Metal Slug games with overclocking enabled, without having to re-enable it again. In MAME, the overclocking settings don’t get saved.

2 Likes

Thanks!

Yeah, I’m really new to the whole emulation scene, so it can get a bit overwhelming to me, at times.

Supposedly, I’m using the latest (…the only?), correct FBN set - I’m not using FinalBurn Alpha - but I reckon I just put its contents into a dedicated folder and called it a day. I didn’t update/change anything else with it, while for example I’m now reading it’s recommended to rebuild the entire set using specific dat files and a program called clrmamepro (which is new to me, so I’ll have to look for some basic tutorial), in order to get it fully working with RetroArch.

I find console emulation so much easier, but hey… I guess that’s how it’s all supposed to be after all. :smiley:

Historically, FinalBurn was the first emulator to run CPS1&2, however i wouldn’t say it’s perfect or superior to recent MAME, especially since accuracy is not our main goal. There are other reasons for using FBN though, like lower input lag or near-perfect savestate support (which allows even lower input lag through readahead, and good netplay support)

Also, i’m assuming the OP is using the right romsets with FBN, it wouldn’t launch otherwise.

I’m willing to investigate this, but i can’t find any of those issues, could you post screenshots or something ?

2 Likes

@BarbuDreadMon, thank you for checking out!

Sure thing, I provided some screenshots, which I hope will help in some way.

This is Wonder 3: MAME (https://i.imgur.com/CIHXhC1.png) and FinaBurn Neo (https://i.imgur.com/aq1mAt7.png). As you can see, on FBN, the gun isn’t cut as it should be, but ends up behind the text. The same ROM file has been used, here, from the FBN ROMset I mentioned earlier, just associated to the MAME core (currently 0.212) for RetroArch.

This is The King of Dragons: the video (https://www.youtube.com/watch?v=iq9YCPgaUEI&feature=youtu.be) is very short and the flickering seems to be quite random, as it can be more severe than that… Nonetheless, you can see the bottom part of the screen - around the Cleric’s name, portrait and weapon - blinking from black to normal. Once again, the ROM file is the same.

I should add that I’m coming to emulation after almost 20 years of collecting and gaming on almost each and every consoles, with the last 3/4 spent having fun with the Framemeister and the OSSC, trying to get the best video quality out of my retro systems. This is to say that probably my eye is just naturally prone to catch minuscule imperfections, and probably little bugs like those above aren’t even worth considering, while working at a new version of an emulator that doesn’t even aim to perfect accuracy.

As another example, I’m one of those people immediately noticing stuff like this (Warzard: https://i.imgur.com/hpoNjqS.png and https://i.imgur.com/ZjvEXfy.png… Take a look at the vertical blank lines), whereas the remaining 99,999% of the game just works perfectly.

Do you own the cabinet to confirm this ? There is no evidence FBN is the one in the wrong here, while searching on youtube i found videos showing both cut & uncut gun, however i don’t know their sources (well, most arcade videos on youtube are from either MAME or original cabinet, so it’s unlikely that https://www.youtube.com/watch?v=Ax8Agt5AhDY is from FBA/FBN though). Furthermore the way the box is cut with a big blue border on the right (while there is no border on the other 3 sides) in the MAME version looks kinda weird imho.

The video is private, i can’t watch it.

Nice catch !

1 Like

Apologies for the video, it should be ok now.

About Rooster, unfortunately, the only physical version I own is on the SEGA Saturn, and that’s a port rather than emulation, but that bit is cut like on MAME there, so I assumed that was the correct way for it to be displayed. You’re right, it’s not even a clean cut, but old CAPCOM games used to have kinda rushed graphical details here an there, so the effect on FBN just caught my attention more (the gun ends up right behind the text, looking more confused in my opinion).

One thing I noticed is that if I watch recent emulation videos on YouTube, it’s way more likely for me to find the same glitches that I experienced (Warzard’s for example), compared to videos from 5/6 years ago… So I’m wondering if different specs and OS may be a factor too (as in Windows 10 compared to Windows 7, for instance), in conditioning how an ever evolving emulator outputs consistently working ROMs.

Interesting, that’s typically what i explained above, we don’t target for perfect accuracy, there is actually a “hack” (i.e something that shouldn’t work like this on real hardware) to lower input lag by 1 frame in FBNeo, and it seems this hack is actually the cause of this flickering. Thanks for the report, i’ll talk with the other members of the FBNeo team about disabling this hack for kod and its clones.

If SS’s Rooster has the gun cut, then that’s probably the correct rendering, thanks for confirming.

Using the same binary, with software rendered emulators, it shouldn’t happen, you can get differences when :

  • different compilers or compilation flags are used
  • emulator is hardware rendered (i.e. directly using the gpu for rendering)
1 Like

Ah, I got it, thanks for clarifying!

And I think I noticed the effects of your ‘hack’ in action almost immediately, and it all makes more sense now. FBN felt more responsive right off the bat, to me, compared to Mame (both running in RetroArch), even though I couldn’t really point that out for sure, or come up with any reason for that… Not ‘faster’ per se, just snappier (apologies for the crude terminology!).

Thanks again. ^_-

I believe the MAME libretro core has 3 frames of input lag on CPS1, while FBN has 1 frame, except for captain commando and now king of dragons (should be available on the buildbot in a few hours), those 2 have the “hack” disabled so they have 2 frames of input lag. Disabling this hack didn’t totally fix the flickering in kod (i’m wondering if some of this flickering wasn’t present on the real hardware…), but it’s still a lot better now.

2 Likes

Sounds great, thank you so much for helping out. :+1:

I tested two games, Gunforce 2 and 3 Wonders on both Mame 212 core and fbaneo. I used frame advance to measure the lag frames and they are the same in both cores. I also tested the older fba core and its also the same.

Is there something im missing or need to configure to make fba faster?

That’s not a CPS1 game and i was talking specifically about CPS1.

You mean the MAME core has only 1 frame of input lag on CPS1 now ? That’s surprising, it wasn’t like this in the past.

You can use readahead which should work flawlessly in FBNeo.

@BarbuDreadMon Through more extensive tests - and in spite of what you mentioned about accuracy - it still seems to me FBN perfoms noticeably better than MAME with Neo Geo games, at least on my specific setup. Speed feels accurate, BIOS selection via direct option is great, and I didn’t stumble in any graphical glitch yet (whereas, with the latest to MAME master cores, I noticed random junk lines during 2020 Super Baseball’s intro, just to name one). Pseudo-CG animations - such as Blazing Star’s area 2 and third Terry’s stage in Garou - also play perfectly.

Next I’ll try a few games I’ve always found a bit problematic on older builds of both emulators, such as Riding Hero and Goal! Goal! Goal! / Super Sidekicks.

Both MAME and FBneo give me 3 frames of lag according to frame advance. In CPS1 games.

Sure, i can enable read ahead on FBneo but it’s not like the core is faster by default.

I can see the portrait of my character changing next frame on sf2’s champion select, so i assume the input lag inherent to emulation code is 1 frame, and any additional frame of input lag are probably inherent to the game itself.

I tested the game myself in both MAME and FBneo myself. While you are right that the portrait is changing next frame in FB and takes 1 additional frame for MAME to do it (which again is faster than any other case), the actual selection rectangle takes the same amount of frames in both cores to change the character.

Also, in gameplay the frames of lag are the same, i get 3 frames of lag and i can see the character preparing to jump on the 4th frame in both cores. Other moves may have one extra frame of lag, but it’s the same in both cores in all cases.

I don’t know why the portrait is changing faster in FBneo, but so far that’s the only difference i saw between the 2 cores. Tested a few other games like 1944 to see how fast the plane shoots and it’s again the same. So i wouldn’t declare FBneo being faster than MAME, let alone having only 1 frame of lag. Both cores have 3 frames of lag in 99% of cases.

It does though : if results are visible next frame, that means the input data is sent successfully to the game when it draws next frame, i believe that’s the very definition of input lag in emulation, what happens beyond that is game+hardware’s business.

So now, from your tests, the MAME core is confirmed to have only 2 frames of input lag, i’m pretty sure it was 1 frame greater 1-2 years ago (i also remember ppl saying standalone didn’t have that extra frame of input lag), thanks for the update.

Btw, nowaday the MAME core works properly with readahead ?

From my tests i confirmed that both cores have 3 frames of lag in almost all cases. Character sprites react on the 4th frame at best. Bullets appear on the 4th frame. So that’s 3 frames of lag. It is even higher in 30fps games like Metal Slug (6-7 frames of lag in both cores). So it depends on the game but it’s the same in both cores.

The only outlier is the character portrait in SF2 which has zero lag on FB and only 1 frame in MAME. But i’m not going to say FB doesn’t have lag or MAME only has 1 frame of lag. Like i said, both cores have 3 frames of lag in general except for this one exception.

I didn’t test runahead. Also, the latest MAME 0.214 core is broken. I use the previous, 0.212 core that i kept as a backup.

You are mixing input lag induced by emulation and input lag inherent to the game though. There is no way to cut through the input lag inherent to the game, except through readahead, and i’m not sure of MAME’s compatibility with this feature, unlike FBNeo.