3rd Strike Lag/Speed Tests


I wouldn’t call it 1 frame of of lag, I would call it “between 0 and 1 frames of lag.” It’s a limitation of the hardware/technology.

You also technically can’t affect the frame after your button press (due to inherent lag in the game), but I would count that frame as lag, wouldn’t you?

I considered that, so I picked a move that has an obvious 1st frame of animation. I’ll edit my original post to include that reasoning.


it was recorded from versus city cab + CPS3. OE for both consoles was on CRT.


To the first part I think we’re on the same page.

Regarding the bolded: I’m not quite sure I follow.

Here’s what I’m getting at. Have you seen NKI’s ST lag test? From the very moment the red light turns on, he counts that moment as 1 frame of lag. That’s the first mistake that I don’t agree with, because like you said it’s a limit with the current way display/video game technology is that only the next frame displayed from that moment can change.

I’ll agree with you insofar as one day someone might invent a technology that could instantaneously change that moment in time to display the different animation, and that theoretical technology would be the first “0 lag” technology in video game history but I don’t see it that way. I play Super Mario 3 and to me that’s a 0 lag game, even though theoretically speaking it’s a 1 frame lag because of limits of technology. If you want to count that then that’s fine that’s just a difference of philosophy.

But back to NKI’s video. The reason counting moment 1 as a frame is a mistake is because let’s say that such a technology did exist where the moment the red light came on the animation on screen was already changed, under NKI’s video that counts as 1 frame of “lag” when it’s not that’s instantaneous.

He’s off by at least 1, 2 by my definition, and possibly 3. Let me explain.

So the red light comes on. Ok you guys call that 1 I call that 0 whatever. 1 frame goes by and there’s no animation. 2 frames go by and there’s no animation. 3 frames go by and there’s no animation. Frame 4 hits BOOM animation. But that’s not 4 frames of lag it’s 3 (or IMO 2) because why would you count the changed screen as another frame of input lag? The lag is 3 frames from when the button is pushed until the animation changes not 4. And like I said because I don’t count the 0 frame as 1 I say you’re looking at 2 frames of lag.

But I will take it one final step further, sort of. While I can’t confirm this for myself, you probably could, I could be wrong on this and I’m glad you took this into consideration, but am I wrong that Ryu’s C.HK in ST, that during the 4 frames of startup, the first frame shows no animation change at all? I thought that the actual animation itself, which is 3 frames, was him still just crouched? I could be wrong. But if I’m right that puts the game at 1 frame of lag by my definition 2 by yours.

Point of reference: Ryu C.HK is 4 frames startup 4 active 5 recovery. The animation however is just 2 frames (mistake when I said 3).

This doesn’t change anything btw. If everyone is doing tests this same way then that just means IMO games like ST and 3S have 1-2 frames and SFV has 6 frames.


Question: are you aware of the inherent 3-4 frames of lag built into the game?


Isn’t that what we’re discussing now?


I thought we were, but then you said this:


Here let me visualize my argument.

These are from NKI’s ST lag test video description can be seen in these screen caps.

From the moment the light comes on he counts that as 1 frame.

This isn’t wise IMO because the moment the light comes on its impossible for the frame captured at that moment in time to change. You can’t change an already displayed frame. Only the next frame can show a change. It would be like if you were counting seconds and you said OK: 1…2. Can’t do that. That’s only 1 second of time, not 2. It’s a simple mistake people make but it’s true. You start at zero. OK: 0…1. 1 second.

Here’s another flaw. When NKI gets to the forth frame he counts that as 4 frames of lag. No. That’s 3 frames of lag if by the 4th frame it’s changed.

By NKIs measurement if the animation changed on the first frame he counts that as 2 frames. Because he’s mistakenly counting frame zero as 1 and the changed frame as another frame of lag.

Even if you don’t agree with me that frame 0 is frame 0 and you call that frame 1, you’re still making a second mistake by calling the next frame 2 frames of delay. If the next frame displayed the animation than by your definition that’s 1 frame of lag not 2. And by my definition that’s zero frames of lag, not 1 due to limitations of the technology and the fact human beings experience life as moving forward we cannot freeze to my knowledge unless there are some vampires out there who’d challenge me on that.

Either way you look at it, you’re off by 1 frame if you’re using NKI’s method at least.


^The light comes on and NKI puts a 1 above the image. But no delay has occurred yet. Thats starting the count as 1 not 0 as we discussed earlier. You wouldn’t count a second as 1…2 you’d count it as 0…1. Regardless if you guys want to call that 1, that’s fine, as long as other people know this and other testers are doing the same thing.

However I feel my methodology makes more sense. Let’s say that the moment the light came on the animation DID change. That would be instantaneous, 0 frames of lag. Yet you’re calling it one…

Now we get to the first frame. If this displayed an animation id then say the game has no frame delay at all, because the first frame since the button push has changed. He counts this as 2. But how? If the animation changes here, how can that be 2 frames of delay when only 1
Frame has occurred?http://i1278.photobucket.com/albums/y511/Mark_Wyda/Mobile%20Uploads/image_zpscnmuxti0.png
Now frame 2.


And now on the 3rd frame the animation changes:

So if you did it this way and people are doing it this way you’re off by 1-2 frames.

Like I said this doesn’t change anything. It just means old school SF has 1-2.5 frames of lag new school SF has 4-6 frames for some ungodly stupid reason.


Post fixed. Sorry it’s 5:00am here my brain is French fried.

You misinterpreted. All I was saying is that technically nothing changes. This is arguing semantics but I feel my points are valid. All this would mean if I’m correct is that old SF has 1.5 by my definition 2.5 by yours and if the measurements of SFIV and V were done with the same flaws then 4 frames and 7 respectively not 5 and 8 as it is now (or 3 to 6 by my rules). Either way NKI (and you guys if you’re following his method) is off by 1, if not 2 which I would argue personally.

*doesnt take into account the “invisible startup” of Ryu’s C.HK, which if I was right about and that was just going from memory, would bring the game to 1 frame of delay.


I’m also curious if you guys tested out 3S’s 240P modes on Dreamcast. Maybe test out 2nd Impact as well.

NG and 2I, while barebones in features, were far better ports to Dreamcast than 3rd Strike. They displayed the proper resolution and played better. I have no idea why every other Capcom 2D fighting port forced 480P with that ugly shimmer effect and pixelated graphics (I guess because it’s then compatible with the VGA box but still…)

If you could PM me the equipment you guys use I’m curious to test games myself.

Also I’m thinking about TAS folks. How would you TAS a game that had input delay? Yet all SF and pretty much every video game in history has been torn apart via tool assist videos. Do all 2D games have input delay?

Has anyone measured games using emulators and advancing 1 frame at a time instead of filming the screen and lag measurement tools?


This is 5 consecutive frames from my cps3 test. The button gets pressed at the end of the first frame, then there are the inherent full 3 frames of lag, then the move comes out. These images aren’t scanned instantly, they get scanned over the course of 59.583 seconds. The goal of these tests wasn’t to find out how many frames of lag 3rd strike has built into it - it was to find out how much lag the player can expect to experience for each system, because that’s all anybody cares about.

I counted the frame where the button is pressed because, depending on when in that frame the button is pressed, the player can experience up to an additional frame of lag. Given a large enough sample size, this first frame should theoretically average out to half a frame of additional lag (give or take a little depending on how long the input window for each frame is and also for the time it takes for the scanner to reset from the bottom of the CRT to the top).

I didn’t count the frame where the move comes out as part of the lag. I counted everything up to that frame.


you’ll always be behind by a frame because input is processed before the frame is drawn out.
so no matter when during a frame, it’s always the next frame because the frame you’re looking at was already processed and drawn without your input.
this style of game always draws after a game step. no matter how long the step, it always draws out once before taking another step.

of course it’s not 100% consistent either. the step size varies because the amount of time to process what happened that step varies with a minimum of 16.66~ms.
which we’ve all seen with things like yang’s stage, twirling knockback animations slow the game down.

i feel like this entire discussion (which i’m not even sure the point of, no matter how you want to count them, isotopez is correct) hinges on parryall not understanding how games are programmed/run.


I understand completely.

What you just said is exactly what I was saying.

My point is, why would you count it as lag when that’s a given. It’s impossible for anything other than the next possible frame to display a change, so that’s not lag IMO but it’s fine if you disagree.


I see where you guys are coming from. Next possible frame being the only frame that can show a change, that is a “frame of lag” but all I was pointing out was, by that definition all video games ever made have at least 1 frame of lag then. Limits of the technology.

edit. It’s the limits of 60fps gaming. 120 and even 240Fps gaming is possible and 120 and 240Hz monitors exist. Under those frame rates 1 frame becomes 8.33ms and 4.167ms. Of course you’d need that fast of a camera to capture such high speeds. It could go on. You could have 50,000fps and input response as quick as .1ms right. I don’t think that’s necessary I’m simply stating what’s possible in the future who knows. I just put this here as food for thought :slight_smile:

I’m not saying anyone is right or wrong here, I’m simply pointing out what I’ve observed and communicating it in as clear language as I possibly can.


There is something I think you’re missing Tebbo, and that is that under NKIs lag test method, he counts the moment the red light comes on as a frame of lag. He puts a 1 above it. I don’t agree with that. Because let’s just say theoretically, when that moment of time is captured, let’s say the animation **did ** change. He’d count that as 1 frame of “lag”. Yet under that circumstance, no lag has occurred yet. It’s only once you start counting frames from that zero point that there’s lag.

So to take that one step further. Let’s say on the first frame of lag from when the button is pressed the animation changed. NKI would call that 2 frames of lag, but 2 frames of lag haven’t occurred. Even under what you guys are talking that’s 1 frame of lag not 2, and I call it 0 frames of lag because I’m taking into consideration the limits of the technology and not counting the first frame as lag. Either way it’s off by 1.

I’ll make a video sometime it’s on my list of things to do. It doesn’t change anything, but if kept the same way as it is now, all video games ever made will have 2 frames at least under NKIs method. Mario has 2 frames of lag Sonic has 2 frames of lag. But under my idiology 0 frame is possible and 0 lag means the next frame from when the button is pressed shows an animation change.


Is there a thread somewhere for discussion of NKI’s lag tests? I’m thinking it would be a better place for you to discuss NKI’s lag tests. I have no idea who that is, which is part of the reason I haven’t been responding to anything NKI related.

I’m not positive, but I’m willing to bet sonic and Mario games have the same lag as 3s give or take a little for the different refresh rates. Put in your terms, that would be 3 (using the rules of significant figures) frames of lag on average.

Also, using your rules, scientifically, there is no way to have more than 1 frame of precision in your final result, which is another great reason not to use your method i.e. it doesn’t really tell us anything new.


The video is posted. I was wondering how you conducted your tests, I brought up NKIs video to show you the two things I thought he messed up on, and if you did the test the same way as him.

NKI is a community OG and legend who has contributed tons and tons of work to the FGC including translations frame data videos and much more.




Ok. Awesome thank you.

I was always curious why some arcade boards ran at 59.X FPS vs 60. Model 2 and 3, CPS3, Neo Geo…


Do Scanlines effect lag on PS3/360? I know the filters do, according to the OP but it doesn’t mention Scanlines.



@isotopez if you’re still reading the forum, can you test the nesica version too? I tried it yesterday and it didn’t feel bad at all (for being on PC and everything, of course it’s not cps3+crt by any means. But it could be better than the actual nesica cabinets if you have a good monitor maybe)