Guilty Gear Strive — Sol Badguy clashes with Ky Kiske, showcasing rollback netcode in action

If you’ve ever played an online fighting game and felt like your inputs were delayed — like there was a layer of mud between your controller and the character — you’ve experienced delay-based netcode. If you’ve played a game where online felt almost like offline, that was rollback.

The difference between these two networking approaches is the single biggest factor determining whether an online PVP game feels good. It’s more important than resolution, framerate, or server location. A game with rollback at 150ms ping feels better than a game with delay at 50ms ping. That’s how much it matters.

This is the explanation of rollback netcode that I wish existed when I started playing competitive PVP games.

The Problem: Latency Exists

When two players connect over the internet, there’s a delay between when Player A presses a button and when Player B’s game receives that input. This delay (latency, or “ping”) depends on physical distance, network routing, and connection quality. It typically ranges from 20ms (same city, good connection) to 200ms+ (cross-continent or poor connection).

The fundamental problem of online PVP is: what does the game do during those milliseconds of delay?

Your game needs to display something on screen right now. But it doesn’t have your opponent’s input yet — it won’t arrive for another 50ms, 100ms, or longer. The game has to make a decision: wait for the input, or keep going without it.

That decision is the difference between delay-based and rollback netcode.

Delay-Based Netcode: The Old Way

Delay-based netcode solves the latency problem by adding artificial input delay. If the round-trip latency between two players is 100ms (50ms each way), the game adds 3-4 frames of delay to both players’ inputs. This ensures that by the time your input is processed locally, your opponent’s input for the same frame has arrived over the network.

In practical terms: you press a button, and your character acts 3-4 frames later. At 60fps, that’s about 50-67ms of added input lag.

Why it feels bad:

The delay feels like your character is underwater. Reactions that should be possible become impossible because the input delay eats into your reaction window. If a move has a 20-frame startup (333ms) and you have 4 frames of input delay, you only have 16 frames (267ms) of effective reaction time. That 67ms difference is the gap between blocking a mixup and getting hit.

Worse, the delay fluctuates. If the connection quality varies (and internet connections always vary), the game has to adjust the delay frame-by-frame to stay synchronized. This creates “spikes” where the game suddenly feels more sluggish, which destroys your timing for everything.

And worst of all: if the connection can’t keep up, the game literally pauses. Both players freeze while the netcode waits for the data to arrive. In a competitive game, a mid-combo freeze that breaks your execution is unacceptable. But in delay-based games, it happens constantly against opponents with unstable connections.

Guilty Gear Strive was one of the first major Japanese fighting games to ship with rollback netcode — the difference in online player retention was immediate

Games that used delay-based netcode: Most Japanese fighting games before 2020 (Tekken 7, Dragon Ball FighterZ at launch, older Street Fighter games), many older arena PVP games.

Rollback Netcode: The Right Way

Rollback netcode takes a fundamentally different approach. Instead of waiting for your opponent’s input, the game predicts what they’ll do and simulates the result immediately. If the prediction is wrong, it “rolls back” to the correct state and replays the frames.

Here’s how it works step by step:

Frame 1: You press Forward. The game predicts your opponent will continue holding Back (the most common prediction: “they’ll keep doing what they were doing”). The game renders Frame 1 based on your real input and the predicted opponent input. You see your character move forward immediately, zero delay.

Frame 4: Your opponent’s actual input for Frame 1 arrives over the network. It was Forward, not Back — the prediction was wrong. The game internally “rolls back” to Frame 1, replays it with the correct inputs, then fast-forwards through Frames 2-4 to catch up to the present.

What you see: Your opponent’s character snaps to the corrected position. This “teleport” is the visible artifact of a rollback correction.

Why it feels good:

Your own inputs are never delayed. Zero frames of added input lag. Your character responds exactly as fast online as they do offline. This alone makes rollback feel dramatically better than delay.

The prediction is right most of the time. Players don’t change inputs every single frame. When the prediction is correct (which is the majority of frames), the game looks and feels identical to offline play.

When the prediction is wrong, the correction is usually subtle. A small position adjustment, a slight animation skip. At low latency (under 80ms), rollback corrections are nearly invisible. At medium latency (80-150ms), they’re noticeable but don’t affect gameplay. Only at very high latency (200ms+) do the corrections become disruptive.

The trade-off: Occasional visual artifacts (teleporting, animation skips) instead of constant input delay. For competitive play, this is an overwhelmingly better trade-off. You’d rather see your opponent teleport slightly than have your own inputs feel delayed.

GGPO: Where It Came From

The most common rollback implementation is GGPO (Good Game Peace Out), created by fighting game community member Tony Cannon in 2006. GGPO was designed specifically for fighting games, where frame-precise inputs make delay-based netcode especially damaging.

GGPO was open-sourced in 2019, which was the inflection point for rollback adoption. Before that, implementing rollback required either licensing GGPO or building a custom solution from scratch. After open-sourcing, any developer could implement it.

The fighting game community had been demanding rollback for years before it became standard. The difference between a game with rollback and a game without was so stark that players would refuse to buy fighting games without it. This consumer pressure, combined with GGPO’s open-source availability, finally forced the industry to adopt rollback as the standard.

The visual fidelity of modern fighting games makes smooth online play even more important — rollback lets you appreciate every frame

Which Games Have Rollback?

Fighting Games (Where It Matters Most)

Good rollback: Guilty Gear Strive, Street Fighter 6, Mortal Kombat 1, Tekken 8 (added post-launch), Rivals of Aether II, Dragon Ball FighterZ (added post-launch), Under Night In-Birth II

Notable holdouts: Very few major fighters lack rollback now. The FGC’s decade-long campaign for rollback has largely succeeded.

Hero Shooters / Team Games

Most modern hero shooters use server-authoritative netcode with client-side prediction, which is conceptually similar to rollback but works differently in practice. The server is the source of truth, your client predicts movement and abilities, and corrections happen when your prediction doesn’t match the server’s state.

Games using this approach: Overwatch 2, Valorant, Marvel Rivals, Counter-Strike 2

The key difference from fighting game rollback: there’s a dedicated server resolving conflicts, rather than peer-to-peer resolution. This is more appropriate for 5v5+ games where peer-to-peer becomes impractical.

Arena Brawlers

Battlerite used a server-authoritative model. Most upcoming arena brawlers (including Alea and Nebulagon) are implementing either rollback or server-authoritative prediction with rollback-like client prediction.

For arena brawlers specifically, the networking model matters enormously. These games have tight ability timings, counter windows measured in fractions of a second, and combat where a 50ms input delay can make the difference between landing a combo and whiffing it. Any arena brawler releasing without rollback or competitive-quality netcode in 2026 is dead on arrival.

What “Good Netcode” Actually Means

When people say a game has “good netcode,” they usually mean one of these things:

Low input latency. Your inputs feel responsive regardless of your opponent’s connection quality. Rollback achieves this. Delay-based does not.

Graceful degradation. As connection quality decreases, the experience degrades smoothly rather than catastrophically. Rollback degrades through visual artifacts (teleporting). Delay-based degrades through input lag and freezing.

Connection quality transparency. The game tells you about the connection quality before and during the match. Showing ping, rollback frames, or connection stability helps players understand what they’re experiencing.

Region-appropriate matchmaking. The best netcode in the world can’t fix a 300ms connection. Good matchmaking considers geographic distance and connection quality when forming matches.

Why You Should Care

If you play any competitive PVP game online — which is most PVP players in 2026 — netcode quality directly affects your experience more than almost any other technical factor.

A game with rollback and 100ms ping gives you a better competitive experience than a game with delay and 30ms ping. The elimination of input delay is that significant. When you’re trying to react to a mix-up, confirm a combo, or counter an ability within a specific timing window, those extra frames of delay are the difference between execution and frustration.

This is why the FGC fought for rollback for a decade. This is why “does it have rollback?” is the first question asked about every new fighting game announcement. And this is why, when you’re evaluating which PVP games to invest your time in, the networking model deserves as much attention as the gameplay itself.

The games that get netcode right earn your trust. The games that don’t waste your time. In 2026, there’s no excuse for the latter.