stage speeds


Is this only apparent in the cps2 game? or does it exist in ports as well (DC, hdr remix?)

I’m referring to certain stages like ussr playing faster


Every stage in ST/SFIIX is different on actual CPSII arcade and on emulated CPSII. So if you play GGPO or Supercade online then you will be able to see that the stages are indeed running at different speeds. If you play using something like MAME or FBA offline then you will encounter stage speed differences as well. Just like an actual arcade board. The order of fastest to slowest stage speeds is consistent from what I remember for each regional version of arcade ST and for all of the various emulators. The thing is that most, if not all, of the emulators run ST overall faster compared to the original arcade board. So the same stage on arcade ST is much slower than the same stage on MAME or FBA or GGPO (providing the speeds are the same setting).

T0 is slowest speed and T3 is fastest (T1-T4 on the USA/World boards). Also T0 is not actually zero speed compared to something like CE or Super, it’s actually a ‘turbo’ speed that skips over frames of animations. T0 is kind of misleading if you’ve only played ST and nothing else. Some of the console versions of ST actually have a zero speed setting that is the same speed as CE or Super and does not drop frames (HDR has this for speed zero for example).

As far as console ports go, most of them have one constant speed for all stages per speed setting, so T1 is the same on every stage, T2 is the same on every stage, and so on. This is how the DC version of ST runs, PS2 AE, and how HDR Remix and Classic Run, amongst others. CCC2 is one of the few versions however which has different stage speeds, as that version was emulated CPSII on the PS2, based off of an ST rom file. CCC2 has many known issues that make the game unplayable and stage speeds and other speed settings happen to be one of them. So the one console version where arcade speeds were attempted to be replicated ended up having that goal cause tons of problems.

Stage speeds are important at competitive levels. Getting a stage like Dhalsim’s or Zangief’s versus Blanka’s or Ryu’s can noticeably change the timing for lots of things. You need to constantly be adjusting on the fly in ST because every stage has different timing. It’s yet another dynamic that arcade ST has that many other games do not share. Grand Master Challenge indeed.


I’m planning to do some stage speed tests once I get GMC.

I’ll record a video for each stage on GMC T3 and US T2.

And then count how many seconds elapsed from “Round 1 Fight” to time over.


I’ve mentioned this before but take note that T.Akiba’s listing is of distance time. Stages have different speeds -and- lengths. Performing a push test can’t separate the 2 variables. So even though his listing has USSR as one of the “fastest” stages, USSR is actually very slow on X T3.

The only way to get an accurate list of stage speeds is to time a full round from a common start and end point (like the moment the intro text disappears to when the victory message appears) like papasi is planning on doing.

The only way to get an accurate list of stage distance is to paste screen captures together, although the foreground angling and different levels of parallax make things annoying. Anyone with free time could shed light on some new data capturing this information.


nope I guess I can’t do it now since I can’t get the gmc from mikeidge for the time being.

hopefully this will happen in the future.


it’s stage speeds, not stage distances, learn to read what the thread is about.


Since it apparently wasn’t clear to everyone, I was referencing VF4’s discussion of T.Akiba’s stage speed/length listing, as translated by NKI. For years, it was taken as the gospel for stage speed (at least in English-writing communities) until I debunked it with the USSR example. AFAIK, there’s no definitive list of stage speed because the only one currently available factors in both speed and distance into an amalgam ranking instead of properly testing for only speed.

I strongly suggest thinking twice and taking your own advice to thoroughly read before posting next time. It doesn’t do anyone any good making callous comments, esp. when they turn out to be based on incorrect or ignorant assumptions.

On another unrelated note, I’ll point out that X uses the T1-4 speed names and everywhere else uses T0-3 (which doesn’t make sense since T0 still has frame skipping and is one of the minor reasons that X makes more slightly sense overall). HSF2 does have a true no-frameskip T0 speed.


This is kinda off topic, but what do the speeds in HSF2 correspond to?

Here’s my guess:

T0 - CE/WW
T1 - HF
T2 - ST US Speed 2/JP Speed 3
T3 - ST US Speed 3/JP Speed 4

Is this correct?


Emulator stage speed data can easily be obtained and achieve more accuracy than video tests.

Run the game, start a match and pause before the round begins. Frame advance through the stage and count the amount of frames played to whatever benchmark you’re setting (i.e. round over) for each stage. You’ll have accurate speed down to 1/60th of a second.

As far as actual arcades and consoles, unless you have really good recording equipment…meh.


Ah, good point, Pasky. It would require a considerable amount of additional testing time (or probably a macro or two) but that would be ideal. Kawaks has reliable frame-by-frame if anyone is interested. I’d imagine FBA and other emulators would do just as well.


No need for a macro, and Kawaks is a terrible choice, Use FBA-RR.

Set the frame advance hotkey, and display frame input, both exclusive to fba-rr.

No need for special math or anything, the frame counter and patience is all you need.

The findings from the emulator should show which stages are faster, regardless of version, although console/arcade versions may differ in speed slightly from emulation, the emulator findings should still be accurate as far as finding normal to fastest as far as stage speeds go.

You could also easily find stage width by finding the X coordinate of a player and just placing him at one end of the screen, marking the coordinate value, and then moving him to the other end of the screen and subtract the two. I know the coordinate value exists, I found it once in doing some other testing, it was actually easy to find, I don’t remember the value range it was measured in though, not sure if the middle of the stage is 0 and one side is negative or positive or not, been a while since I’ve looked at it.

Either way that should provide stage widths, although it wouldn’t be in any measurable form (pixels etc…) but would still provide width data.


Ha, I was just pointed to this thread. Coincidentally I did this earlier.

For those interested, the seconds on the in-game clock are not the same frame differences. Frame count from 99 to 98 differs from 98 to 97 which differs from 97 to 96 and so on.


Care to explain the reason for it?

No point of bring it up if you are not going to capture such data.


nice. i am really interested to see comparison of JP T3 vs US T2 on various stages. Hopefully it’s not too much too ask :smiley: (if you have time and energy to do that)


Because I believe it’s always a better idea to bring up possible solutions than just complain about 2 missing pieces of information that a previous test tried to determine. Assuming FBA’s frame counter works well and an absolute horizontal position value exists, Pasky’s suggestions offer meaningful info on capturing both of those solutions.

I’ve actually tried to capture stage distance in the past based on connecting screen captures but it became too imprecise because I don’t have any imaging software on my PC. Sorwah and I were even going to compare whether USSR’s speed was different between the JP and US version (as has been questioned recently) as a starting point but it turned out my supergun isn’t compatible with his Hauppauge. When I have time in the future, I’ll try these new proposed tests via emulator. Or if you have time and access, you’re certainly free to preempt me and try the tests yourself.

Switching to another point, Sorwah’s test here shows free select and fixed speeds in ST are essentially identical for all intents and purposes. Honestly though, zass already came upon the same results years back in a link mentioned above. So I hope that once and for all, folks running ST tourneys will quit wasting time with free select.


Actually, I take what I said about fba-rr back, I thought it had the frame counter when it’s mame-rr that does, mame also has a frame skip counter (Not shown when paused) letting you know if the emulator had to skip a frame (not whether st does it’s own frame skipping however).

mame-rr - Project Hosting on Google Code

I had a little time and this is what I meant, I began the count the very first frame after “FIGHT!” disappears. I ended the count the very first frame the two dummies move to their draw game pose animations:

USA: 5173 - 2118 = (3055 / 60) = 50.9166 Seconds

China: 12388 - 9407 = (2981 / 60) = 49.6833 Seconds

USSR: 15373 - 12468 = (2905 / 60) = 48.4166 Seconds

So ya, proof of concept. USSR is definitely faster than USA and China. Haven’t done any other stages.


Ha Pasky, we basically did the same thing timing wise. :stuck_out_tongue:


i hate sim, honda and chun’s stages so badly, i really hate these stages, the music and the speed, so annoying.


Stage speed never bothers me.

Get good.


Alright, I’ve gone and did comparisons of each stage with Ryu & Ken doing nothing but jumping straight up (incase I need to make visual display of speeds) in both GMC & USA versions of the game. The only exception is Dictator’s stage - where I use two Dictators. This was done because you can’t have a non-dictator on Dictator’s stage due to the game’s coding of not letting you win against the final boss by plugging in Player 2. Instead, you would get the continuing character’s stage if neither player is Dictator.

Anyways, on to the data!

First frame was registered as the first frame that “FIGHT!” did NOT appear on, and the last frame in each timing was the end of the first round on the frame the characters first displayed their Draw Loss Poses.

In terms of Speed differences between each setting, I just compared Guile’s stage:

The ##:## is Total Seconds:Frame Remainder (60 frames in every second)
So 30:45 = 30 seconds and 45 additional frames

For those who have trouble reading the data, here’s the speed order from fastest to slowest:

  1. USA Turbo 3, GMC Turbo 4, USA Free 3 (47:56, 47:56, 48:02)
  2. GMC Free 3, USA Turbo 2, GMC Turbo 3, USA Free 2 (49:54, 49:49, 49:58, 49:50)
  3. GMC Free 2, USA Turbo 1, GMC Turbo 2, USA Free 1 (51:23, 51:29, 51:25, 51:22)
  4. GMC Free 1, USA Turbo 0, GMC Turbo 1 (52:40, 52:47, 52:49)

Then I went and compared USA T2 & GMC T3 stages from each other as those are the accepted settings for Tournament Play:

For those having trouble reading the data, here’s the speed order from fastest to slowest:


  1. Dhalsim
  2. Zangief
  3. Sagat
  4. Claw
  5. Dictator
  6. Guile
  7. Cammy
  8. Fei Long
  9. Chun Li
  10. E. Honda
  11. Dee Jay
  12. Ken
  13. T. Hawk
  14. Blanka
  15. Boxer
  16. Ryu


  1. Dhalsim
  2. Sagat
  3. Zangief
  4. Claw
  5. Guile
  6. Dictator
  7. Cammy
  8. Chun Li
  9. Fei Long
  10. Boxer
  11. E. Honda
  12. Ken
  13. Dee Jay
  14. T. Hawk
  15. Blanka
  16. Ryu

If you were to group them so the ones in the same second are listed:


  1. Dhalsim, Zangief, Sagat, Claw (48:11, 48:13, 48:33, 48:45)
  2. Dictator, Guile (49:39, 49:58)
  3. Cammy, Fei Long, Chun Li (50:06, 50:48, 50:54)
  4. E. Honda, Dee Jay, Ken, T. Hawk, Blanka, Boxer, Ryu (51:03, 51:18, 51:19, 51:23, 51:26, 51:40, 51:51)


  1. Dhalsim, Sagat, Zangief, Claw (48:08, 48:25, 48:28, 48:37)
  2. Guile, Dictator (49:49, 49:54)
  3. Cammy, Chun Li, Fei Long (50:05, 50:45, 50:56)
  4. Boxer, E. Honda, Ken, Dee Jay, T. Hawk, Blanka, Ryu (51:03, 51:05, 51:11, 51:19, 51:23, 51:32, 51:48)

You could even argue that Tiers 2 and 3 could be either merged or have Cammy jump to Tier 2 as she is closer to Tier 2 than Tier 3’s other numbers.

Keep in mind, that each of these time differences reported are the TOTAL time difference lost from one version to the other over the course of an ENTIRE round. It’s not like each version is running equally then bam one is hanging for an additional 9 frames. That’s 9 frames over the course of thousands. For example, Chun Li’s stage is a difference of 9 frames over the course of 3000 being rendered. That’s a 0.3% loss/gain of frames over the course of 50 seconds. I personally don’t feel that’s noticeable when actively playing.


Super Turbo Tournament at Evolution: Las Vegas, NV July 30, 2011 1pm