What's new

CPS2 SSF2T Speed Test - Real Arcade Hardware vs MiSTer vs FBNeo vs 30th Anniversary

I found in that video some really wild inconsistencies with each of the rounds. I was wondering if could run the same tests again without jump (neutral only,) and with other stages? T.Hawk's Stage can be wildly inconsistent. The only thing that was exact, within a margin of error that was consistent is the transition from to the continue screen, and the countdown to 0 on the continue screen.
 
TechnicalMonkey said:
I found in that video some really wild inconsistencies with each of the rounds. I was wondering if could run the same tests again without jump (neutral only,) and with other stages? T.Hawk's Stage can be wildly inconsistent. The only thing that was exact, within a margin of error that was consistent is the transition from to the continue screen, and the countdown to 0 on the continue screen.
Click to expand...

Each stage in ST is different speed there is an old video that shows this off on YouTube. That's why I make sure to run same stage on all testing.


I'll see if I can do another speed test on Cammy stage since that is one of the most popular in fightcade.
 
Probably won't
Harineto said:
Each stage in ST is different speed there is an old video that shows this off on YouTube. That's why I make sure to run same stage on all testing.


I'll see if I can do another speed test on Cammy stage since that is one of the most popular in fightcade.
Click to expand...
Mexico is one of if not the worst stage as it is inconsistent with the amount of objects on that stage. Take a look at the left side of the stage and count how many bird are feeding on the ground, or which ones fly away at random.
 
LOL that MiSTer is the least accurate of all of them, should wipe the smugness off some faces. I have always thought that FBNeo on my Pi4 with RGB-Pi feels practically identical to my real CPS1/CPS2 boards, any time I think I've been cheated by lag or inaccuracy I immediately try the same on the real hardware and usually find it is the same. In fact extreme run ahead on emulation actually makes lag lower than the real thing.
 
edd_jedi said:
LOL that MiSTer is the least accurate of all of them, should wipe the smugness off some faces. I have always thought that FBNeo on my Pi4 with RGB-Pi feels practically identical to my real CPS1/CPS2 boards, any time I think I've been cheated by lag or inaccuracy I immediately try the same on the real hardware and usually find it is the same. In fact extreme run ahead on emulation actually makes lag lower than the real thing.
Click to expand...
The problem with RunAhead though is that more than one frame ahead causes the character (best noticable in shmups) to "jump" ahead or skip ahead. At 2 frames it gets noticable (especially when playing with high ship speeds) and at 3 it just feels wrong. This is my experience with RA. I guess FBurn's RunAhead behaves the same.
 
Last edited:
Runahead (or preemptive frames) can correctly remove as many input lag frames as they were in OG hardware/software/CRT. Basically it runs the game internally quicker, in "fastforward" mode, assuming the input doesn't change, so processing the input before the real hardware and then waiting the right moment for the frame to be displayed correctly; if the input changes while waiting (and the previous assumption fails), then the emulation backtracks (rewind), processes the right input and then goes on in "fastforward" mode again. So, as said before, this tecnique can cut "just" the original input lag. If it's configured to cut more frames than OG input lag, then frames of animation are cut at each backtrack, resulting in stutters/jumpiness.

Pretty neat technology indeed. Ideally, if you know the OG game input lag, the input lag added by your emulation setup over the OG one and if the latter is lower than the former, then run ahead should be configured to compensate exactly the latter. Basically what we want is OG lag + Emulation lag - RunAhead compensation = OG lag.
 
Last edited:
I think that we are forgetting that this game does skip frames making execution harder that made the game too fast on lots of software emulation. The standard to play 2X is Hard-Turbo 3, and on ST it was Hard-Turbo 2. Neither one of these is played on software emulators, at these speeds online, or offline, without hacking them to lower overall base CPU speeds compared to their hardware counterparts. JTCPSx, while not perfect, has been the closest to emulating the hardware properly than anything out there. I would love to see the perfect version realized, so if anyone want to take a crack at it, or apply the finishing touches, let's make it happen. Btw @Harineto, if you get the chance, could you run that same JTCPS1 comparison, with the turbo mode on please?
 
Pushing rolling updates on games demanding the highest levels of accuracy with zero regression testing says a lot about how seriously a dev is taking the work and how accurate the rest of their cores should be expected to be but there’s no reason someone else can’t come along and fix it. Its not how accurate "Mister" is its how shit the core written is or whats being applied to it. Hard to believe its been used in Tournaments alongside real hardware.
 
TechnicalMonkey said:
I think that we are forgetting that this game does skip frames making execution harder that made the game too fast on lots of software emulation.
Click to expand...

I haven't done any research into what attributes in that game's software programming would cause it to skip frames, but in the end, MiSTer doesn't change anything with the software code of the game that's executing. It's hardware simulation only. If both the OG hardware and FPGA core are running the exact same game code and the MiSTer FPGA core is 100% accurate to the original hardware then frames should also be skipping identically between the two.
 
ShootTheCore said:
I haven't done any research into what attributes in that game's software programming would cause it to skip frames, but in the end, MiSTer doesn't change anything with the software code of the game that's executing. It's hardware simulation only. If both the OG hardware and FPGA core are running the exact same game code and the MiSTer FPGA core is 100% accurate to the original hardware then frames should also be skipping identically between the two.
Click to expand...
That's the case even if a software emulator is 100% accurate. FPGA isn't magical per se.
 
I'm bullying Claude Opus 4.5 to suggest how we can improve the speed of the core.

Why not

It's thinking alright haha

1767908832295.png
 
Last edited:
TechnicalMonkey said:
The standard to play 2X is Hard-Turbo 3, and on ST it was Hard-Turbo 2.
Click to expand...

This is actually a good point, and while it may not necessarily impact the results of the tests... I would be curious to know what speed Harineto was using (I assume it was all default settings).

djsheep said:
lol.
Click to expand...

Very insightful post, thank you for this.

SC-Ultra said:
Hard to believe its been used in Tournaments alongside real hardware.
Click to expand...

@TechnicalMonkey probably knows more than I do, but as I recall people performed blind experiments at several tournaments and nobody could tell the difference between the two at the time. Worth pointing out that this was years ago and there have likely been more than a few regressions or other changes to the core that might impact what those results would look like if they were done today.

Locutus73 said:
That's the case even if a software emulator is 100% accurate. FPGA isn't magical per se.
Click to expand...

AFAIK none of the software emulators are 100% accurate, but people can get by with FB Neo/FBA in the majority of cases. That said, I have never seen Jotego claim the CPS cores are cycle accurate. Maybe he did and I'm wrong, but I've only seen MiSTer weirdos on social media make such claims and they obviously aren't true.

FWIW, someone reported speed discrepancies with the Hyper Fighting CPS1 core, so it sounds like Jotego is prioritizing the BRAM migration work which may help resolve these problems.
 
You must log in or register to reply here.
Back
Top