SH3 Blitter Delay Discussion Thread

[nesrulz]

Deathsmiles MegaBlack Label 1CC with blitter 63%.

http://www.youtube.com/watch?v=XMuixXuWgMk

[blip82]

Just wondering if there have been any updates to these settings?
I find 63% on DSBL is just way too slow when the screen gets busy, but i have have never played the original PCB so can't compare. 55% seem to be a good speed for me.

I am running groovymame 0.156 and using the ROM from the Mame ROM set.

[blip82]

Is 63% still the right values for DSBL using groovymame 0.156? I have never played the real pcb but 63% feel way too slow when the screen gets busy. If I turn off blitter delay I can get full 100% speed.

I'm guessing this is because blitter delay is a bit crap still? I have been running DSBL at 56% blitter delay and it seem pretty good.

[spadgy]

Firstly, apologies to blip82 for taking so long to approve your debut post (you can post etc. freely now that is done). We get a lot to approve I'm afraid you slipped through! I see above in your determination to have your first post appear instantly you've done the post twice (a common happening! ). Usually, I delete the duplicate posts and explain in the thread why one went. However, as your 'duplicates' are a little different here I'm not deleting, and still explaining. So that's what's happened here, and, other users, that's why there's two posts.

Welcome to the forum, and I hope you get some help!

[Xyga]

So still nobody who understands what happened with 0.156 update ?

I don't quite understand the entry... 'detangle the speedups'

http://github.com/mamedev/mame/commit/4 ... bb4f6a39db

[Astraea FGA Mk. I]

The lower the blitter delay the lower the input lag. I have tested various settings extensively. This is how I see it:

Emulation of the SH3 Cave games will never be identical to the actual PCB installed in a cabinet. Spending lots of time trying to get it as close as possible is futile as the end result will be the same; it is not 100% accurate. Since the emulated experience will always be different, why not find a viable purpose? Blitter delay set to 40 or lower makes the response extremely tight, it feels as nice as the 360 ports. As a result there will be less overall slowdown making the game much more difficult. This is a good thing! Practice a game under these conditions. That way when you play the actual PCB you will perform better. Using it as a training tool makes sense to me. Take what you got and find a way to make it something unique that isn't possible anywhere else.

[shmuppyLove]

As a result there will be less overall slowdown making the game much more difficult.

FTFY

[CStarFlare]

Spending lots of time trying to get it as close as possible is futile as the end result will be the same; it is not 100% accurate. Since the emulated experience will always be different, why not find a viable purpose?.

Playing the games is the purpose. Accuracy as close to the board as possible is a valid goal - it makes your scores more comparable to others and is overall a more enjoyable experience. (See Mushi PS2 vs. Espgaluda PS2)

Personally, I'm holding off on SH3 emulation until it's reasonably good without fooling with individual settings - some measure of standardization would be nice. But I'm glad there's people fooling with it and working out the best settings in case curiosity gets the better of me.

[Astraea FGA Mk. I]

As a result there will be less overall slowdown making the game much more difficult.

FTFY

Thank you I fixed it.

Spending lots of time trying to get it as close as possible is futile as the end result will be the same; it is not 100% accurate. Since the emulated experience will always be different, why not find a viable purpose?.

Playing the games is the purpose. Accuracy as close to the board as possible is a valid goal - it makes your scores more comparable to others and is overall a more enjoyable experience. (See Mushi PS2 vs. Espgaluda PS2)

Personally, I'm holding off on SH3 emulation until it's reasonably good without fooling with individual settings - some measure of standardization would be nice. But I'm glad there's people fooling with it and working out the best settings in case curiosity gets the better of me.

Buy a cabinet and the PCB

No matter what your scores in MAME are they are not comparable to the PCB. We are not talking about a degree of equality, what matters is this question;
Is SH3 emulation the exact equivalent of the PCB, enough so that scores are interchangeable?

Things alike have advantages and disadvantages. Instead of muting the exceptional potential of the mediums why not differentiate them by their specific unique advantages?
Futile effort pisses me off. It is or it isn't. It doesn't matter how infinitely close you come, if it isn't then it isn't. Play with 62% blitter delay if you want, the input lag ruins the experience entirely.

[trap15]

It is futile, as no matter what your setting it's still wrong most of the time.

[BPzeBanshee]

So still nobody who understands what happened with 0.156 update ?

I don't quite understand the entry... 'detangle the speedups'

http://github.com/mamedev/mame/commit/4 ... bb4f6a39db

Nothing's different. It's reorganisation of the code.

The real alterations you should be looking at is in the general refactoring of the SH4 cpu driver (I'm no real hardware expert but in MAME the "SH3" inherits from SH4). From the last time I bothered to do any thorough testing performance has actually been slightly *worse* doing benchmarks, but upon asking about this on Bannister Forums Shoutbox I was told such refactorisations are critical to getting a possible dynamic recompiler in the distant future when R.Belmont or someone who actually knows how to make a dynarec for MAME gives enough of a shit to do so.

[liteon]

Trap15 or nimitz PLEASE realese something for everyone (particularly the community) speak same thing and harmonized things.
a kind of tribute to cave (since the arcade game seems dead) and for and to sustain the style.

a lovely gift for new year

[Xyga]

Thanks for the explanation BPzeBanshee, so since nothing changed, comments here and there on the net like "some games are now playing at the right speed" are complete BS out of people's imagination ? Lol excellent.

a possible dynamic recompiler in the distant future when R.Belmont or someone who actually knows how to make a dynarec for MAME gives enough of a shit to do so.

Will nevah happen...neeeevah.

@liteon: don't fall into the trap(15)!

[blip82]

Thanks everyone for your replies. I know emulated will never be the same (or close in some cases) but living in an apartment I just don't have the room or money to buy all the cave shooters I would like.

I do have an astro city and egret 3 and like to get a close as possible experience with what I have. I'll probably hold off on the cave stuff until the drivers are sorted a bit more and keep playing with blitter delay turned off I like the fast speed.

Off to Japan in a month so maybe I can find some cheap PCBs

[Astraea FGA Mk. I]

I realize now that from a valid perspective what I have said is incorrect as someone should set the blitter delay to whatever allows them to enjoy the emulation most. I have stated my opinion and philosophy regarding utilising advantage and minimising disadvantage which is coming from waste-not perfectionist mentality.
Enjoy slowpoke it is brilliant!

[Strikers1945guy]

Off to Japan in a month so maybe I can find some cheap PCBs

Buy the ports man

[Astraea FGA Mk. I]

Something else to consider is emulation speed. When at 100% the emulator is emulating at full speed. This is important. For example, if the emulator is running at 75% the music playback will be 3/4 of the normal tempo in parallel to everything on screen running at 3/4 the normal rate.
You can use FRAPS FPS counter to see specifically how well your system runs these games.
60 = 100%, 0 = a static image. 100 / 60 = (1 2/3), so each value on fraps’ counter = 1.666.. %.
SH-3 emulation requires a highly clocked and/or efficient (modern) processor. While MAME offers a multi-threading option in the config.ini file it still requires and does the majority of the work on one thread.

HLSL (High-level shader language) allows for interesting filters to apply to your emulation in MAME. Someone on this board made an excellent CRT emulation preset that truly completes the experience, search for it on the hardware board. HLSL requires a half decent video card. If you have a gaming quality GPU it will not be an issue. Otherwise look it up, figure out what cost efficient solution is available to you.

Having SH-3 on MAME emulating at a static rock solid 100% with HLSL enabled makes all the difference. Then worry about bitter delay as having this set to a specific value is moot if your computer cannot even emulate it properly in the first place.

Also disable v-sync and triple buffering in the .ini file as it adds massive input lag.

[BPzeBanshee]

You don't need FRAPS for MAME. Press F11 to show FPS on the top-right corner.

[nesrulz]

Blitter OFF here.

[tzakiel]

Just wondering if anyone has done any more comparisons or tried and found other good settings. I am still wondering if mushihimesama and espgaluda II can be made somewhat close to the PCB speeds and what settings are good for those games.

[stacks183]

So, I have been doing a lot of testing in order to get slowdown in SH3 games in MAME as close to PCB as possible. Obviously, it isn't perfect, but I think I got pretty good results.

I'm using MAME version 1.68 with vsync enabled (I don't think it has any effect on these settings) and with cheats enabled in ini file (this is important). When you enable cheats, MAME 1.68 let's you overclock CPU in Slider Settings. For some SH3 games, it's impossible to get good results with just a Blitter slider, you also need to underclock CPU.

Here's the settings I use:

• Mushihimesama - never tested this game properly, since Steam port is superior to MAME in every way. Before it was released on Steam, I was using 59% Blitter delay, but I don't know whether it is accurate.
• Ibara - CPU set to 50%, Blitter delay to 59%.
• Espgaluda 2 - CPU set to 44%, Blitter delay to 59%.
• Ibara Kuro - CPU set to 50%, Blitter delay to 59%.
• Pink Sweets - CPU set to 48%, Blitter delay to 59%.
• Nushihimesama Futari 1.0 - CPU set to 40%, Blitter delay to 59%.
• Mushihimesama Futari 1.5 - CPU set to 40%, Blitter delay to 59%.
• Mushihimesama Futari BL - CPU set to 40%, Blitter delay to 59%.
• Muchi Muchi Pork - set Blitter delay to 59%. No need to underclock CPU for this one.
• Deathsmiles - set Blitter delay to 63%. I genuinely think that Deathsmiles with 63 Blitter Delay in MAME is superior to Steam version.
• Deathsmiles MBL - Blitter delay to 63%. It feels like 63% is too much slowdown when playing earlier stages (I don't think it is, though), but then you play at lv999 and it looks just right.
• DoDonPachi DFK (every version) - again, Steam version is way superior to MAME. Before Steam release, I was playing all versions of this game with 63% Blitter delay, and I think this is fairly accurate, at least for 1.0 and 1.5.

Overall, if you want to find good settings for SH3 games slowdown in MAME yourself, just set Blitter delay to 59% or 63% - these seems to be two magic numbers. If neither setting produces enough slowdown, enable cheats and underclock CPU to 50, and then go down from there, until you hit just enough slowdown. It's also always worth it to set Blitter delay to 59% if you can, since it gives less input lag, but sometimes it's just doesn't cut it. And obviously, if you enable vsync as I did, there will be even more input lag.

Also, you need to set CPU value each time you boot the game. To do this quickly just hold ctrl+arrow key.

I would really love to hear opinion of other people who play SH3 games in MAME. What settings are you using?

[Xyga]

Necro for more on that CPU+blitter tweaking stuff, following a recent discussion here: http://shmups.system11.org/viewtopic.ph ... &start=210

If trap15 is wrong why does underclocking the CPU down to about half in MAME (even somewhere between 30~40% as far as I've experienced) works so much better with some games than blitter delay alone?

After Pink Sweets I gave DFK 1.5 a try yesterday and, although it is much harder to adjust compared to to PS, trying values that approach the original behaviour of the game again it was obvious that it requires a massive CPU underclock as well (maybe a nudge under 40%)
It's really difficult to figure since the game packs tons of slowdowns that you trigger yourself depending on how you play.

Even if it's probably impossible with only these two sliders to achieve a timings behaviour fully identical to the originals, exploring this alternative way of tweaking the slowdowns, bit-by-bit I'm beginning to suspect there really are overall more correct or - at least - acceptable settings waiting to be discovered for each cv1k game in MAME as it is now, making the games playable if not accurate, kind of like a different set of ports after the 360 and PC.

This 'method' of course relies entirely on observation so it's ghetto and a pain (especially if you only have replay videos of the pcb's as reference), plus we can't save the CPU value, indeed. I've tried only two games so far and 'found' approaching values that are different from stacks183's.
When you're near what begins to look like the expected speed/slowdown behaviour, a 1% increment/decrement step influences the whole thing. Also how to guess the 'best' balance of CPU and blitter sliders together, how do the two things influence each-other ? hard to figure where the potential 'sweet spot' is located.

Savestating and going through the levels or particular ingame spots one-by-one is the only way to test this, otherwise it's too much of a pain (since a reset/reboot also resets the CPU slider)
As mentioned I've used GroovyMAME d3d9ex to compensate for the high lag (0.196, autosync, frame_delay 5, portaudio), the downside being that once in a while savestates will crash the emulator, still, for the purpose of testing it's working well-enough.
To be thorough you'd have to adjust the CPU & blitter going through different styles of play, like for instance a low rank run then a high-rank one. Ugh.

[Goompaolo9500]

Try the latest version of MAME. I even tried Akai Katana on there.

[blip82]

Yeah, wondering if any of this is still relevant since 0.191. Pretty much all CV-1000 games run at 100% for me now in Groovymame. Still have issues with getting the correct blitter delay but wondering if the cpu underclock is still the same?

[Xyga]

The 100% when you press F11 is the output refresh rate, which is not our concern here. It's about the core emulation speed and how to adjust it for cv1k since the default isn't right yet (slowdowns not showing to state the obvious)

AFAIK things in MAME 0.200 haven't changed, cv1k emulation is the same, and most games as It seems rather evident now, still require a CPU underclock along setting the blitter delay, if you want to play with a game speed approching that of the orignals.

@blip82: I haven't read about any changes regarding that in the latest mame versions changelogs... in any case the best CPU + blitter values aren't known, there's what stacks183 posted, but with what I've experienced his set values weren't the best, like for instance Pink Sweets working much better @ approximately CPU 40%, blitter 63%, and DFK1.5 probably somewhere below 40% CPU-wise, no idea blitter-wise (yet)
Checking all games one by one to find the best settings with these two sliders is a huge job that will probably never be thoroughly done (in time I'll post some settings if I find worthy ones, but unless at least a handful of 'testers' participate it won't do much knowing about that method of adjusting cv1k)

[Shepardus]

There were supposedly some performance improvements in MAME 0.191 as discussed here, which means you can reach 100% more easily (which, as a reminder, means that the game's running as fast as MAME thinks it should be running, not that there's no slowdown), but the emulation accuracy still remains what it is.

[Xyga]

Yes, though from what I could observe at the time it seemed to mainly attend to the sluggish loadings in places, like when loading a new stage or boss in the ygw games.
Comparing with two different pc's (one a bit too slow for cv1k running it like 80~95%, one more than capable) I couldn't detect an overall performance improvement, though I haven't checked with throttle off, nor running the bench command...
Maybe that shifted the 'right' CPU+blitter ceilings, anyway in my case I've been testing with 0.196 (groovy), and there haven't been any changes to cv1k since then afaik.

The 100% counter still misleads a number of users who don't distinguish from the emulated game speed, it should be clear though that this is just a prerequisite before even attempting to tweak anything related to actual in-game speed, like blitter and cpu.
(this is redundant to your post but it does no harm repeating ^^)

[Bananamatic]

what drives people to do this considering the emulation is completely fubar
why not just play a game that runs well

[Xyga]

I know you only post this to be annoying and try to trigger shit again, because you would never really care to try yourself, nor actually read people's posts paying attention, even less so consider their arguments or participate/contribute in any useful way.

Don't ruin this thread, you've already made your mind on the topic anyway.

[Bananamatic]

the most useful contribution I can give is to ask your boss for extra hours instead of playing around with blitter
then use the money to buy the ports (or even the pcb if there aren't any ports)