The fuck?
So either this isn't common or MS has some next level AI going on in the cloud.
Not really :(
could it be that you're still connected to them via voice, but its not them you're playing against.?I've spoken to every single person I've fought online do far, or heard their home background noise.
So either this isn't common or MS has some next level AI going on in the cloud.
If you've used the GGPO client with any real frequency before, you've probably experienced this a couple of times in the past where the game flat out desyncs and both people on each end are fighting some weird otherworldly version of their opponent that's just straight up not matching up with the actual inputs their opponent is doing. The end result would be two people who both thought they won their match. Except I've never had it resync back up in GGPO. Whenever it would happen, we would just have to kill the window and restart the match.
I've never seen it happen in any retail releases that supported rollback or GGPO, but given that it's happening in KI, I guess it might be possible.
Justin Wong and Max were both streaming a couple days ago with no difference between streams, my theory is that DSP is just a wanker trying to weasel out of an RQ. But then again... I have seen GGPO desync and stranger things have happened.
The AI thing is definitely not true. And if at any points the inputs don't match up it's probably because one of the videos are fraudulent.
It makes sense. Voice and gameplay networking are probably handled by two completely separate systems. You are still connected through voice (most likely through MS built-in API) but the game goes out of sync. As a result, you still hear people but you might not see the same things gameplay-wise.That makes no sense. Or am I missing something?
This is probably an extreme example. It is more likely that it won't be as noticeable in most cases, hence why you don't have a huge number of players reporting weird match results. Surely two players in chat would notice if they don't get the same match outcome?
Is there really a competitive and professional following for the Killer Instinct series?I'm sure all of the competitive and professional fighting game players will have a huge problem with this, but a lot of people who invested into the console and game probably won't want to hear about it and will either say it's not much of a big deal, or deny it happens often enough to be a problem. The point is, the fact that it even happens at all is a problem in and of itself.
I've never considered the series a serious fighting milestone. It just had good graphics on the SNES and some interesting characters, but the fighting always felt weak to me.
Continuing with the beer analogy, this kind of networking error is like a game of telephone where one step in the chain got incredibly drunk. Messages are being passed back and forth, but what is sent and what is received no longer have relevance to each other.
I know one of you has KI! Get it done Parker!
so wtf happened here?
And a reminder to others if you want to feel like you're an online champion, go to a local tournament fighting event and play there bc online will never and should never replace that. People take online way to seriously.
Historically that's how the first warcraft multiplayer ever went :DOriginally Posted by MrCookiepants
So that's the same match between the same people, yet it plays out completely differently on each side?
http://www.codeofhonor.com/blog/the-...arcraft-part-3
----The first-ever multiplayer game of Warcraft was a crushing victory, an abject defeat, and a tie, all at once. Wait, how is that possible? Well, therein lies a tale.
Well no, the whole point of rollback is the match will play out the same in the end because of the constant input revisions. If the matches end up different then that's not a correct rollback implementation. And even GGPO, which is notably permissive regarding desyncs, won't let them happen for more than 20 frames. Whatever they did here, they trid to be clever to increase their lag window beyond what has been tested to be reasonable, thus the results.It's just Desync from rollback netcode. It happens once in a blue moon if the connection is bad between the players. It has happened in 3rd Strike, DS and it will happen again in future titles using rollback netcode. It's not some massive controversy. The video is just an extreme example of it happening.
there you go, me against perfect legendif i play any fighter online and record the match, and the person i'm playing also records the match, we would expect the footage to be near identical.
so wtf happened here?
http://www.twitch.tv/perfectlegend/b/483791655 (01:16:00)
https://skydrive.live.com/redir?resi...C2DC740B%21108
1:1
if people need more i'll record tonight
@but yeah, let's blow stuff out of proportion as usual
This.It's just Desync from rollback netcode. It happens once in a blue moon if the connection is bad between the players. It has happened in 3rd Strike, DS and it will happen again in future titles using rollback netcode. It's not some massive controversy. The video is just an extreme example of it happening.
Have you people played Virtua Fighter 5 FS? Surely you remember that game's stupid rollback mystery.
It's a small bug, that happens rarely.
OMG 11/10. Haha damn this thread has delivered and then some. You should copyright that before they do. While you're at it, copyright "shootars". Imagine the next movie tie-in game to James Cameron's Avatar 2, it has advanced cloud powered AI; they call them "Avataratars".Originally Posted by elektrixx
You're not directly fighting people; you're against their "fightars".
Really? How about playing the game on its own merits before condemning it because of an issue that at the end of the day, seems like a small issue. It would be one thing if this issue broke the game, but it doesn't, and don't act like it does. No fighting game's online play is perfect. None. This is just another example of that.
For example, if the latency was 50ms to the other guy, I could input a PPK combo which would happen immediately on my instance of the match BUT it would happen 50ms later on the other guy's instance of the match as a result of traveling through the internet, and it just so happens that within that 50ms delay that the other guy started blocking. This means on my instance of the game I'm doing a combo on my opponent because I opened him up with my PPK, but on the other guy's instance of the game he blocked my PPK and is now punishing me in his game, and the individual game states will start diverging wildly from then on. This isn't really much of a problem for low-skill games as the speed of inputs and reactions is slow enough that it's very likely both instances will come to the same conclusion of the game state, but for pros where 10 commands are being inputted per second the chances of one slipping into the latency gap is likely.
tldr the two consoles don't have an agreed upon game state, they just assume they are by replicating the inputs each player sends into the instance of the match of the other player, delays and all.
You got all that from this one instance?It looks to me like there's no actual agreed upon game state, each player's Xbone runs their own instance of the match and just sends the inputs to the other guy's instance of the match. That's actually pretty ingenious, but it only works if you can guarantee <5ms of latency (for a fighter). I think this only really happens if you get a hard desync though (no communication whatsoever), instability would probably use GGPO (I'm not sure if Killer Instinct uses GGPO, but I would design it that way).
For example, if the latency was 50ms to the other guy, I could input a PPK combo which would happen immediately on my instance of the match BUT it would happen 50ms later on the other guy's instance of the match as a result of traveling through the internet, and it just so happens that within that 50ms delay that the other guy started blocking. This means on my instance of the game I'm doing a combo on my opponent because I opened him up with my PPK, but on the other guy's instance of the game he blocked my PPK and is now punishing me in his game, and the individual game states will start diverging wildly from then on. This isn't really much of a problem for low-skill games as the speed of inputs and reactions is slow enough that it's very likely both instances will come to the same conclusion of the game state, but for pros where 10 commands are being inputted per second the chances of one slipping into the latency gap is likely.
tldr the two consoles don't have an agreed upon game state, they just assume they are by replicating the inputs each player sends into the instance of the match of the other player, delays and all.
The latency does play a part in KI's GGPO-based netcode based on this video, I can agree with that.It looks to me like there's no actual agreed upon game state, each player's Xbone runs their own instance of the match and just sends the inputs to the other guy's instance of the match. That's actually pretty ingenious, but it only works if you can guarantee <5ms of latency (for a fighter). I think this only really happens if you get a hard desync though (no communication whatsoever), instability would probably use GGPO (I'm not sure if Killer Instinct uses GGPO, but I would design it that way).
For example, if the latency was 50ms to the other guy, I could input a PPK combo which would happen immediately on my instance of the match BUT it would happen 50ms later on the other guy's instance of the match as a result of traveling through the internet, and it just so happens that within that 50ms delay that the other guy started blocking. This means on my instance of the game I'm doing a combo on my opponent because I opened him up with my PPK, but on the other guy's instance of the game he blocked my PPK and is now punishing me in his game, and the individual game states will start diverging wildly from then on. This isn't really much of a problem for low-skill games as the speed of inputs and reactions is slow enough that it's very likely both instances will come to the same conclusion of the game state, but for pros where 10 commands are being inputted per second the chances of one slipping into the latency gap is likely.
tldr the two consoles don't have an agreed upon game state, they just assume they are by replicating the inputs each player sends into the instance of the match of the other player, delays and all.
But, again; this is all blue moon shit. There's too much footage out there that can dispute the "AI" nonsense and summing it up to be just de-sync.
No. Why would anyone want to fake this. It's just a de-sync.Let's not forget the possibility this was flat out faked too.
The closest thing this could be fake to is the replay mode issues littering KI, but even then it'll take a lot of internet magic to fake this on this guy's behalf to make DSP--a professional fighting game player--look like a sore loser. And if it is fake; who /cares/?
That's some messed up shit.
Yes, that's approximately how netplay should work, apart from the comment about latency.It looks to me like there's no actual agreed upon game state, each player's Xbone runs their own instance of the match and just sends the inputs to the other guy's instance of the match. That's actually pretty ingenious, but it only works if you can guarantee <5ms of latency (for a fighter).
We've gone over this in the other thread, but the basic idea is in input delay systems you augment the available latency by running the game late (so you can add the delay to the maximum expected latency) and in rollback systems you augment the input latency by executing the game even if you don't get input and retroactively apply it "in the past" by rewinding and fast-forwarding the game internally (so you can add the rollback window to the maximum expected latency). GGPO combine both delay and rollbacks to get the widest lag window possible (delay+rollback window).
In the ideal model it's not that the game state is not agreed upon, it's that the mechanics guarantee if it was agreed upon once it will not have diverged whenever you've got all the inputs. If you're missing some, it will diverge temporarily and come back to an agreement when you receive the missing inputs.
However, the principle is only sound for a small window (20 frames in GPGO's case, much much less in rollcaster's case) *and* to account for bugs the system should regularly check both sides agree for the span of the fight for which they've received all the inputs. There should also be regular adjustments for wall clock skew.
And when I say "regular" I mean at least every second, not huge desynced sequences like what's happening here. No idea what they are thinking for by letting it drift so much.
How did this manage to get through QA?
| Thread Tools | |