RMG-K, a fork of the RMG Nintendo 64 emulator, now runs rollback netcode across its entire library as of a May 14 update. That means every N64 game—from Super Smash Bros. to GoldenEye 007 to Mario Kart 64—can theoretically play online with the same responsiveness you'd expect from a fighting game. Developer CigNus has noted it's currently limited to two-player sessions, so don't expect four-player Mario Party chaos just yet. If you've written off emulator netplay as laggy and unplayable, this is the point where that assumption becomes outdated.
The Hidden Variable: Why "Entire Library" Is a Bigger Deal Than It Sounds
Here's the part most coverage skips. Rollback netcode in emulators isn't new—Fightcade and similar projects have used it for years—but it's almost always been hand-tuned per game. The developer writes prediction logic, hitbox timing, and state reconciliation specifically for Street Fighter II or Third Strike. What CigNus claims to have done is implement rollback at the emulator core level, meaning it works for any N64 game without individual patches.
This is genuinely unusual. Emulator rollback typically requires the emulator to save and restore full machine state dozens of times per second—CPU registers, RAM, GPU framebuffer, the works. For a 3D console like the N64, that state is enormous compared to a 2D arcade board. The "it was honestly not that hard" quote from the developer sounds like modesty, but it also hints at a specific technical bet: RMG-K may be using a lightweight state representation or aggressive compression rather than full snapshots. That trade-off matters to you.
If the state is incomplete or compressed too aggressively, you get "false rollbacks"—the screen stutters backward because the emulator mispredicted, then corrects. Bluesky user Grasluu00 reported Spain-to-Australia play at 4 frames of delay versus 9 previously. That's a massive improvement. But notice what's unsaid: we don't know the packet loss rate, the game being tested, or whether that 4-frame figure holds under sustained play with variable latency. One clean session isn't a guarantee.
The "entire library" promise also masks asymmetry between games. A 2D fighter or GoldenEye deathmatch has predictable player movement—rollback shines here. But try F-Zero X at 60fps with thirty machines on screen, or a game with heavy RNG synchronization like Pokémon Stadium's minigames. The emulator can't predict random number generation accurately, so you'll see more visual corrections. The netcode runs. Whether it runs well depends on what you're actually playing.
For decision-making: if your goal is competitive Smash 64 or GoldenEye multiplayer, this update is transformative. If you want to co-op through Banjo-Kazooie or race in Diddy Kong Racing, the experience may still frustrate. Two-player limits also mean you're not replacing Parsec or local sessions for group play—you're gaining a specific tool for head-to-head matches.
Where to Start Without Wasting an Evening
Don't just download RMG-K and assume it'll "work like Steam." Emulator netcode has friction that platform-native multiplayer doesn't.
First, verify your controller setup before touching netplay. N64 games expect specific deadzones and analog ranges. A modern Xbox or PlayStation controller mapped 1:1 will feel wrong in Super Smash Bros.—the original game's physics were built around the N64 controller's octagonal gate and specific stick resistance. Many competitive players still use original hardware with USB adapters, or controllers like the Brawler64 that replicate the original's mechanical limits. Your online experience is capped by how well your inputs translate.
Second, matchmaking isn't automatic. RMG-K uses direct connections, not a server browser. You'll need your opponent's IP, or a relay service, or a Discord community coordinating sessions. The technical barrier is low but real. Compare this to something like Nintendo Switch Online, where you press a button and wait. The payoff is lower latency; the cost is logistical effort.
Third, test your specific game before committing to a session. Boot it locally, check if your machine maintains full speed with no frame drops. Rollback can't fix a host who's already dropping frames—the emulator has less headroom to rewind and resimulate. If your PC struggles with certain N64 titles at native resolution, netplay will amplify that problem.
Here's a practical priority list:
| Your Situation | Best First Step | Expected Friction |
|---|---|---|
| Competitive Smash 64 player | Join a dedicated Discord, test rollback vs. your usual netplay setup | Finding equally-matched opponents; controller parity |
| Casual with a specific friend | Both install RMG-K, test one game you both own legally | Scheduling; NAT traversal if direct connection fails |
| Curious about GoldenEye multiplayer | Watch Grasluu00's footage first, gauge if latency looks acceptable | Two-player limit; no four-player complex missions |
| Wanting to explore obscure games | Start with games that have deterministic physics (racing, fighting) | Little community; you'll be discovering edge cases |
The "legally own" point matters for emulator ethics and for your own sanity. Dumping your own cartridges ensures you have a known-good ROM. Downloaded ROMs vary in header accuracy, and mismatched ROM versions between players are a classic desync source that rollback doesn't solve.
The Trade-Offs Nobody's Talking About Yet
Rollback netcode shifts burden from network to local hardware. That's the core bargain, and it has consequences.
You gain: Playability across continental distances. Grasluu00's Spain-to-Australia example would be unthinkable with delay-based netcode—you'd be waiting half a second for inputs to register. Rollback lets both players act immediately, with visual correction if the prediction fails.
You lose: Consistent visual smoothness. The screen will occasionally "snap" as the emulator corrects. In a fighting game, you're trained to watch hit confirms and frame data; the snap is distracting but doesn't break strategy. In a 3D platformer or shooter, sudden position changes can disorient. GoldenEye specifically: you round a corner, see an enemy, fire—and then the emulator rewinds two frames and you're already dead. The information you acted on was predictive, not confirmed.
The hidden cost: CPU and memory usage. Saving state every frame for potential rollback is cheap on modern hardware for 2D games. For N64 emulation with accurate graphics plugins, it's nontrivial. If you're on a laptop or older desktop, you may need to drop internal resolution or disable certain enhancements to maintain the headroom rollback demands. The emulator "works" on your system for local play. Netplay is a different threshold.
The community risk: Two-player limits mean this won't replace existing solutions for group play. Parsec, Moonlight, and similar streaming tools still dominate for couch co-op at a distance. RMG-K fills a niche—competitive 1v1 for games that never had online infrastructure—not a universal need. If you're imagining finally playing Conker's Bad Fur Day campaign co-op with a friend, that's not what this update enables.
There's also a sustainability question. CigNus's fork is relatively new. Rollback implementation is complex enough that maintenance matters—network libraries update, bugs emerge with specific games, community expectations grow. The "not that hard" quote is encouraging for adoption, but also suggests the developer may not have built extensive debugging or telemetry tools yet. Early adopters are effectively beta testers.
What to Do Differently Now
Stop assuming emulator netplay is uniformly worse than native online. For N64 specifically, RMG-K's rollback means certain games now play better than Nintendo's own official offerings—compare to Nintendo Switch Online's N64 emulation, which uses delay-based netcode and adds noticeable input lag. The paradox: a fan emulator may now outcompete the platform holder for competitive play. Your decision framework should weight specific games and opponent locations more heavily than "official vs. unofficial." The technical implementation matters more than the brand.
