Melee on HDTVs

[Zankoku]

Original Facebook Thread

So, apparently the AverMedia LiveGamer Portable is able to passthrough a 480p component signal from a Wii (or a Gamecube, if you somehow have a Gamecube component cable) into an HDMI cable, and the Asus VH236H HDTV is able to display it without any input delay. Below are the reference photos that I took for verifying.

Given, the VH238H is $130 and the AverMedia LGP is $150, so at $280 per monitor not including the setup and game themselves, this probably isn't an affordable option for tournaments just yet, but it's still something to keep in mind for the future, as well as present streaming/recording purposes.

 

[TaFoKiNtS]

Thank you Ankoku!

Also, to piggy back, here's a database of good low lag monitors
http://www.displaylag.com/display-database/

 

[Bones0]

You said "display" instead of "delay".

So if we get the portable thing, we can play lagless on a monitor? Sounds sick. Is the quality on the monitor actually better than on a CRT, or does it just look nicer simply because it's a monitor?

 

[DRGN]

There are much cheaper solutions. You can buy Component to VGA (which most flat screens use, including the one mentioned here) converters for $70. There are probably cheaper ones that too, but I haven't looked too much. Also, if you simply change the video output mode on a Wii to progressive scan, and then use Component cables (which are like $5 for Wiis. Buy it online.) you should be able to get lagless display on your tv. Kinda depends on your specific tv though; different tvs have different quality circuits for their upscaling and other video processing, so some may actually still have a little lag. For $5 though I think it's worth a shot to test!

 

[Zankoku]

The official EVO and MLG monitors do not have component inputs, which is why I needed to convert component to HDMI. If you can confirm the component to VGA converter is lagless on the given monitors (Asus VH236H and BenQ GL2450HM), I'd be all for that too.

 

[Kadano]

1. Those photos don’t prove too much, latency difference can still be at up to roughly 15 ms even though the visible frames are the same. The only way to find out the amount of latency with <1ms accuracy is to use an oscilloscope. Because it’s highly unlikely someone here has access to that, here is an article about PRAD’s findings: http://www.prad.de/en/monitore/specials/inputlag/inputlag.html
In short, LCD’s have a latency of at least 5 ms (10 are more likely to get if you’re lucky). While that is less than one frame, it’s still 7000 times as much as the latency of a CRT (those have about 700 nanoseconds). These values might be off because the digital progressive output of the Gamecube might be a bit faster (I heavily doubt this is more than some microseconds though).
Still, considering top human visual reaction times are about 130 ms, a difference of 10 ms might be noticeable for some players. It’s not much, but I wouldn’t start treating LCDs to be “without any input delay” just yet.

2. The Gamecube’s digital port uses YCbCr natively, the same image format as HDMI. In theory, it might be possible to build a cable directly to HDMI with only a little bit of tweaking involved (as far as I know, some resistance values aren’t compatible, I’m a noob at electronics though). Thus we could bypass buying expensive devices that perform a somewhat ridiculous task—the component cable uses a chip that converts YCbCr to YPbPr, which then has to be re-encoded to YCbCr by the converter. For the Wii I think this chip was built in internally so “stupid” cables are sufficient.
So what I’m trying to say is that we could get a very low-level connection between the Gamecube and a TV set by either custom-building cables to plug into the digital port or directly wiring the internal RGB signal to a VGA cable that can be used with monitors. The latter would obviously require opening the Gamecube and soldering a bit. We would also need a connector for the digital port to make the Gamecube switch into progressive scan.
While pretty work-intensive, these methods should be cheap and yield the best image quality due to little processing.

Something similar has already been done for the N64: http://forums.benheck.com/viewtopic.php?f=58&t=43034

 

[Zankoku]

It's confirmed that the Asus VH236H has an average of 10ms of input delay. It's still used as the tournament standard at EVO and many other tournaments. In this case, with CRTs being a category of TVs no longer seeing manufacture, and coming off the major success Melee saw at EVO, I don't think it's worth shunning the 1/2 frame of delay at the expense of major tournament organizers. Yeah, our game has 3, 2, and even single-frame-startup attacks, but it still feels rather entitled to insist on true lagless when the communities of pretty much every modern HD-compatible tournament game accept the delay associated with VH236H/GL2450HM.

I'm also still somewhat bitter that one of the CRT TVs I provided for EVO was stolen.

 

[Kadano]

It's confirmed that the Asus VH236H has an average of 10ms of input delay. It's still used as the tournament standard at EVO and many other tournaments. In this case, with CRTs being a category of TVs no longer seeing manufacture, and coming off the major success Melee saw at EVO, I don't think it's worth shunning the 1/2 frame of delay at the expense of major tournament organizers. Yeah, our game has 3, 2, and even single-frame-startup attacks, but it still feels rather entitled to insist on true lagless when the communities of pretty much every modern HD-compatible tournament game accept the delay associated with VH236H/GL2450HM.

As long as I can get good CRTs through local listings for 0-15€, I prefer to stick to those. But I understand that for large events like Evo, LCDs would be much easier to handle. I hope internal video → HDMI will be figured out.

I'm also still somewhat bitter that one of the CRT TVs I provided for EVO was stolen.

Of all the things you could steal at a tournament, a CRT would definitely be the last thing to come to mind. Like … why? It makes no sense. Or was it a special flat HD CRT?

Nice all clear controller by the way! Where’d you get that?

 

[Zankoku]

I like collecting things. The all-clear one I bought from mikeHAZE, not sure where he originally found it.

 

[Massive]

I'm also still somewhat bitter that one of the CRT TVs I provided for EVO was stolen.

I had my really nice 480p TV stolen at SMYM a few years back.
I'm still salty about it (so much that I refuse to ever bring a setup OoS again, ever).

In other news, if you need a definitive test of your user experience between the two setups, load up melee using the input delay tester code IE wrote. It's still not perfect, but averaging the values it returns should give you a pretty accurate estimate.

 

[The Mofo]

It would be great if that code, Massive, could indicate sub-frame lag by showing decimal places.

 

[Bones0]

Is there no way to make a Melee mod that runs natively (I think that says what I mean) on a higher resolution so the monitor won't have to convert it?

 

[Zankoku]

The hardware itself is unable to support such a thing.

 

[KrIsP!]

Dolphin emu...I think that would make it possible but it'd take a lot of people doing hard work to produce an illegal dolphin emu. And then we'd need computers/good laptops instead of cubes. I know people got mario galaxy in 720p. Does that make it native? Either way that's a stretch if even possible.

 

[Massive]

It would be great if that code, Massive, could indicate sub-frame lag by showing decimal places.

This is actually not possible, since the game cannot accurately determine input delay faster than it polls (once per frame). The game only gets pad data at intervals of ~16.667 ms and will process your input for the next frame as it renders, so you could say you always have one frame of input delay. If the combined input delay of your display medium is less than that, it will not even be measurable by any in-game method, not reliably anyway.

The only way to accurately capture this information would be to use a high speed camera setup or poll input separately from the game processing logic (which would be just as/more complicated than the cameras).

 

[The Mofo]

If it averages your inputs, you should be able to determine sub-frame lag. I don't feel like giving a full explanation right now, but it's a little bit like Buffon's Needle Trick.

 

[Massive]

Yeah, an average very easily could show those values over a large enough sample size, however the code to do so would be substantially more involved than what is added now.

I'll look into adding it, but its not super high on the priority list since the code released now could technically already be used to generated the average manually.

 

[rjgbadger]

.

I'm also still somewhat bitter that one of the CRT TVs I provided for EVO was stolen.

i brought 3 CRT TVs for the EVO tournament; when i went to get my TVs on monday, ALL 3 of my TVs were taken, but other TVs were left. Employees at the casino said they'd have to get rid of them soon, so i took 3 different TVs. Perhaps one of the ones i took was one of yours?

 

[Zankoku]

Mine was the one used for top 8 on stream. Afterward someone other than me claimed it was his and carried it off.

 

[rjgbadger]

does it have component? one of the tvs i picked up looks kinda like the one on top 8 stream. i can take pictures if you want?

 

[Fizzi]

Just received all the stuff required to test. Here is one image:

Monitor is a BenQ RL2455HM (MLG Monitor used for RTS I think)

 

[jerflip]

Tricky stuff

When you play them side by side, do both displays feel the same? I'm still sorta skeptical

 

[kiw1]

can anyone do photos with regular HDTVS/monitors? If there is any difference/how big the difference is?

because i might will get one of the cards and try out how it works with my regular monitor/TV

 

[himynameisruss]

we definitely need to go the way of emulated melee on PC in the future. 1080p widescreen melee with no input delay on my xl2420t is lovely.

 

[Fizzi]

Tricky stuff

When you play them side by side, do both displays feel the same? I'm still sorta skeptical

Yes, to me it feels flawless. I might even say it feels better simply because the image is nicer. No interlacing means you see the entirety of every frame.

can anyone do photos with regular HDTVS/monitors? If there is any difference/how big the difference is?

because i might will get one of the cards and try out how it works with my regular monitor/TV

I'll do this.

 

[kiw1]

Yes, to me it feels flawless. I might even say it feels better simply because the image is nicer. No interlacing means you see the entirety of every frame.



I'll do this.

you're my hero, what kind of setup do you use anyway? i'm kind of interested in checking out how much delay my TV has, so i can find out how much improvement i will get... though it can't get worse lol

 

[Fizzi]

I just recorded a 240 fps video of the same two set ups.

Can be found here: http://www.youtube.com/watch?v=yCaGMJTUGJY

This frame in particular is very interesting. The GO has begun to appear on the RL2455HM but not on the CRT. This may be due to lag on this particular CRT (I will test different ones), or lag induced by the downscaler (I've sent an e-mail to the company that makes these asking about the conversion time).

Here is a GIF of the same game frame. My assumption is that the Go is one frame and since I am recording at 240 fps we get to see this one frame appear in 4 separate frames. As you can see... it appears earlier on the monitor than the CRT.

 

[SaggyG]

good **** fizzi

 

[Skiller]

Here is a GIF of the same game frame. My assumption is that the Go is one frame

the "go" comes out the first frame after the timer is done and lasts 39 frames, disappearing completely on frame 40.

 

[Kadano]

Here is a GIF of the same game frame. My assumption is that the Go is one frame and since I am recording at 240 fps we get to see this one frame appear in 4 separate frames. As you can see... it appears earlier on the monitor than the CRT.

What devices did you use to convert component to composite? If that thing has lag, that would explain why the CRT was slower.

Edit: Shame on me for not seeing that line in your post.

 

[kiw1]

everything would be so much easier with a wavebird and 2 recievers

 

[Fizzi]

What devices did you use to convert component to composite? If that thing has lag, that would explain why the CRT was slower.

Yep. I mentioned that somewhere in my post.

everything would be so much easier with a wavebird and 2 recievers

I actually did this test as it was suggested by Mofo. The CRT appeared to be about 1.25 - 1.5 frames faster. That said, there are still unknowns with this test.

After discussing with Mofo last night we've come up with a plan for a test that is just about full-proof (I think). It requires me making a small device probably so stay tuned. It would be nice if I could contact Asus/BenQ and get them to send me monitors so I could put them through the test with the promise of advertising the results. I would recommend the crap out of the winner if one is far better... I'll try to do that.

 

[Fizzi]

the "go" comes out the first frame after the timer is done and lasts 39 frames, disappearing completely on frame 40.

Thanks for this. But the frame I'm actually interested in is the really huge Go. The first frame of the Go appears to be very large where the second frame of the Go is much smaller. It makes it very easy to determine the frame appearing because it covers the whole screen.

I've also been using the pause details on the corner of the screen because those appear and dissappear in one frame as well and they cover the whole screen.

 

[kiw1]

Yep. I mentioned that somewhere in my post.



I actually did this test as it was suggested by Mofo. The CRT appeared to be about 1.25 - 1.5 frames faster. That said, there are still unknowns with this test.

After discussing with Mofo last night we've come up with a plan for a test that is just about full-proof (I think). It requires me making a small device probably so stay tuned. It would be nice if I could contact Asus/BenQ and get them to send me monitors so I could put them through the test with the promise of advertising the results. I would recommend the crap out of the winner if one is far better... I'll try to do that.

I guess the only problem with this test would be, that the disc loading time could be different, though if u do it with the start button ingame the result should be bulletproof. or just use the shine, should as well be a real good method

 

[Fizzi]

Using shine is actually harder to determine. The first frame of shine is actually pretty small and kinda blurry looking. Pause seems to be much better.

With the wavebird there's two potential things:
1) Both consoles don't receive the wavebird input at the same time.
2) The consoles are out of sync.

Like I said though, we've thought of a method using just one console that will let us compare set ups very effectively. I'll begin working on setting it up.

 

[Massive]

Honestly, I think this is convincing enough evidence to start using LCDs for tourneys.

Even with the (what appears to be incredibly minor) lag included, this removes an enormous entry barrier from us being included in basically every modern fighting game tournament.

 

[Fizzi]

Honestly, I think this is convincing enough evidence to start using LCDs for tourneys.

Even with the (what appears to be incredibly minor) lag included, this removes an enormous entry barrier from us being included in basically every modern fighting game tournament.

I did a test with a wavebird and the results were not nearly as good. The LCD appeared to be about 1.5 frames behind the CRT.

On another note, Sewell, the manufacturer of the downscaler did respond to me. The e-mail was as follows:

Alright I understand, unfortuantly we dont have a way to split this without converting to HDMI (which is what all our splitters are) so I cant test that latency. So I can't say for certain what that one is, but one tech that I just asked this questions to mentioned that all our converters should be under 4ms, so virtually no delay.
So once again it is no longer than 4ms.
Hope that helps.

If the wavebird test was not flawed... I would expect the delay of the downscaler to be more around 40 ms.

The new concept for a test is as follows:

Single set up test (one display, one console, no splitter)

We have a device that lights an LED when it hears noise beyond some threshold (still in the process of acquiring this, parts are ordered). The video output of the console is plugged into the TV while the audio output is plugged into external speakers. We use the high speed camera to begin recording and then pause/unpause melee. With music off in game, the pause/unpause sound should trigger the LED when the input is read by the console, then, at some later time (or previous, it doesn't really matter), we should see the pause frame appear on screen. We count the amount of video frames between the LED lighting up and the pause frame appearing/disappearing.

We can run this test on multiple set ups using any kind of display and compare the results. I may even be able to get decent data recording faster than 240 fps with this test. The image quality becomes pretty terrible at the faster speeds but it might be good enough to determine the appearance of a frame.

So we can initially do this test on our control (CRT), and then on our test (LCD) compare the deltas from LED on to pause frame appearing and compare them to determine the difference in lag.

Credit to Mofo for the original idea.

The beauty of this test is that it bypasses the problem of the controller state only registering on a game frame because our trigger is provided when the game recognizes the input not when we actually pressed it. It also bypasses the need for a downscaler going to CRT and will allow us to use a completely raw composite input (which is currently viewed as being completely lagless).

 

[Fizzi]

Continuation of the conversation with the Sewell representative:

Me

Okay I appreciate the response.

We have recently tried to come up with ideas on how to compare displays without the need for a downscaler.

We tried one of these tests already and it appeared that the downscaler added about 34 ms of delay compared to the test without it. This would contradict the number you gave me but there may have been some confounding factors in this test.

We have an idea for a better test it just requires some acquiring of extra materials.

Thank you,
Jas

Sewell

I recently tested one of our PC to TV converters and it was at 31ms, which also contradicts what I said yesterday (which came from my co workers). I am almost positive he meant 40ms is the slowest they get.

 

[The Mofo]

 

[Zankoku]

Have you already run the test again on the LGP? I'm interested in hearing the exact delay for that one.