You fire a clean shot in Valorant, the enemy drops, then a half second later they are alive again behind a wall and you are dead. That single moment is client-server networking working exactly as designed, and it explains most of what people call “lag.” Your PC and a machine in a datacenter are running two slightly different copies of the same match, and the server gets the final vote on which one was real.
Understanding that split is the difference between blaming the game and fixing the parts you actually control. Some of your latency is pure physics that no setting will beat. A surprising amount of it is routing, buffering, and a handful of local mistakes that are genuinely worth correcting. This walks through how the whole exchange works, then where the milliseconds go, and what you can change today.
The two machines running your match
Every online match splits the work. Your client captures input, predicts what is about to happen, and draws the world. The server holds the one authoritative version of events that every player has to agree with. Between them, small packets fly back and forth dozens of times per second.
Here is the loop in practice. You move your crosshair and click. Your client shows the shot instantly using client-side prediction, so the game never feels like it is waiting on a reply. At the same time it sends that input to the server. The server takes inputs from everyone, works out the true game state, and broadcasts corrections back. If your client guessed wrong, the server’s version overwrites yours. That is why you sometimes see a hit marker but deal no damage. Your machine predicted a hit; the authoritative machine disagreed.
Your client is responsible for a specific set of jobs:
- Input capture: reading mouse, keyboard, and controller in real time.
- Prediction: estimating where other players are between updates using their last position and velocity.
- Rendering: drawing the world, effects, and interface from the latest server data.
- Interpolation: smoothing motion between server packets so movement looks fluid rather than stuttery.
The server owns the decisions that have to be consistent for everyone:
- Player positions: the definitive location of every entity at any moment.
- Hit detection: whether a shot connected, judged against server-side positions.
- Rule enforcement: movement speed, cooldowns, damage, and physics.
- Anti-cheat checks: rejecting actions that are not physically possible.
If you want the deeper mechanics of how these updates are packaged and reconciled, our breakdown of game netcode covers delay-based versus rollback models in detail. For this piece, the takeaway is simpler: your screen is a prediction, and the server is the referee.
Where your milliseconds actually go
Ping is not one number. It is a stack of delays added together, and knowing which layer you can touch is what separates a real fix from wasted effort.
The floor is physics. Data moves through fiber at roughly 200,000 kilometers per second, which feels instant across a city and painfully slow across an ocean. A player in New York hitting a Virginia server sees something like 15ms to 25ms. The same player reaching a London server jumps to 80ms to 100ms, and a Tokyo server pushes past 150ms. No router setting beats the speed of light, so server selection is the single biggest lever most people ignore.
On top of distance sits routing. Your packet does not travel in a straight line. It hops through a chain of routers owned by different networks, and a bad handoff inflates everything. A West Coast player whose ISP drags traffic east before sending it back to a nearby server can watch a 40ms connection balloon to 80ms for no good reason. That specific failure, bad routing rather than bad distance, is the one third-party routing services are built to attack.
Different genres tolerate different amounts of this delay:
| Ping | What it feels like |
|---|---|
| Under 20ms | Ideal for competitive shooters and fighting games |
| 20ms to 50ms | Smooth, no meaningful disadvantage |
| 50ms to 80ms | Playable, a slight edge lost in fast duels |
| 80ms to 120ms | Noticeable delay, real competitive cost |
| Over 120ms | Severe handicap in anything reaction-based |
Those numbers assume your ping is steady. Inconsistent latency, the swing gamers call jitter, usually feels worse than a higher but stable ping. A rock-steady 60ms plays better than a connection bouncing between 30ms and 90ms, because prediction and interpolation can adapt to a constant delay far more easily than a moving one. If your ping graph looks like a heartbeat monitor, our guide to fixing jitter is the better starting point than chasing raw ping.
Tick rate, and why CS2 stopped bragging about it
Server tick rate is how often the server updates the world. A 64-tick server recalculates 64 times per second, sampling every 15.6ms. A 128-tick server doubles that to a sample every 7.8ms, which gives finer resolution for tracking fast movement and shots.
For years this was the headline spec, and it is where a lot of older guides get stuck. Counter-Strike 2 changed the conversation. CS2 does not simply run 128-tick. Valve built a sub-tick system that timestamps every input down to the millisecond, so the server knows the exact moment you clicked, jumped, or fired even when it happened between two ticks. The server still ticks at 64, but it reconstructs the ordering of events inside each tick from those timestamps rather than snapping your action to the next update. Shots register based on when you actually pulled the trigger, not when the next tick arrived. Any guide still describing CS2 as a plain “64-tick versus 128-tick” choice is describing the retired CS:GO model.
Valorant took the opposite route and committed to true 128-tick servers globally. Riot’s own netcode writeup explains why: at 128-tick the server framerate is 7.8125ms, which they needed to keep hit registration tight and to shrink the head start attackers get when peeking a corner. Riot measured that “peeker’s advantage” at roughly 141ms under baseline conditions, and by combining 128-tick servers with their private Riot Direct backbone at 35ms latency they trimmed about 40ms, a 28% reduction. For players on 144Hz monitors the advantage drops to around 71ms, nearly cut in half. Those are small numbers that decide gunfights.
Apex Legends sits at the other extreme, and Respawn has been candid about it. Apex servers run at roughly 20Hz, updating the full world once every 50ms, and in their servers and netcode deep dive the team explained they are in no rush to change it. Tripling to 60Hz would roughly triple bandwidth and CPU cost while saving, in their words, about two frames of latency in the best case. That trade is why a battle royale with 60 players stays at 20Hz while a 10-player tactical shooter runs 128. Our server tickrate explainer has the full comparison across a dozen 2026 titles if you want to see where your game lands.
Tried every setting and still stuck at 90ms?
When the bottleneck is your ISP’s routing to the game server, no local tweak reaches it. WTFast tests alternate paths through its gaming network and often trims 30% to 50% off ping for players stuck on bad routes.
Lag compensation, or why you die behind cover
Without help, every duel would be won by whoever has the lower ping, because their view of the world is fresher. Lag compensation exists to level that, and it is the direct cause of the most frustrating deaths in online shooters.
The common approach is called “favor the shooter.” When your shot arrives, the server rewinds its own history to reconstruct what your screen showed when you fired, accounting for your ping. Then it checks whether your crosshair was on the enemy at that rewound moment.
Walk through a 40ms connection in Valorant:
- You fire at an enemy at your local time of 1000ms.
- Your shot reaches the server at 1040ms.
- The server rewinds its state to roughly when you saw the target.
- It checks whether the enemy sat under your crosshair at that reconstructed moment.
- If yes, the hit lands, even though the enemy has already stepped behind cover on their own screen.
Both players experienced something true. You landed a fair shot on the target you saw. They reached safety before, from their timeline, any bullet could have touched them. The server honored the shooter, and the person in cover ate a bullet that, to them, came from nowhere. Games tune how much of this they allow. Valorant compensates aggressively up to a ping ceiling; CS2’s sub-tick timing narrows the window differently; Apex’s 20Hz cadence produces wider rewind windows that players feel as trades and “around the corner” kills.
The practical lesson is to play to your latency. On a high-ping connection you win by peeking first and pre-aiming known angles, because the shooter is favored and you want to be the one initiating. On a low-ping connection you can afford to hold angles and react, since your view of the server is close to current.
What Frame Warp changes about the last few milliseconds
Everything above is network latency. There is a second, quieter delay between your mouse moving and the screen reacting, and in 2025 Nvidia went after it directly. Reflex 2, announced at CES, adds a technique called Frame Warp on top of the existing Reflex low-latency mode.
The idea is clever. While the GPU finishes rendering a frame, the CPU already knows your very latest mouse position. Just before the frame hits your display, Frame Warp shifts the rendered image to match that newer camera angle, then uses a predictive algorithm to in-paint the small gaps the shift exposes. Nvidia claims up to a 75% cut in PC latency, and in a Valorant demo on an RTX 5090 running past 800 FPS they measured end-to-end latency under 3ms. Frame Warp shipped first for titles like Valorant and The Finals.
It is worth being precise about what this does and does not solve. Frame Warp attacks the local render-to-display gap on your own machine. It does nothing about the 80ms your packet spends crossing the country to a distant server. If your problem is a far or badly routed server, Reflex 2 is not the answer; the fixes in the next section are.
What you can actually change on your end
Physics and server design are fixed. Here is the part that is yours, ordered from biggest impact to smallest.
Go wired. A single Ethernet cable removes the most common source of jitter and packet loss in one move. Wi-Fi adds a variable delay that no in-game setting can undo. If you cannot run a cable, our Wi-Fi gaming guide covers the settings that help most, but wired remains the correct answer.
Test for bufferbloat. This is the fix almost nobody checks. When someone in your house starts a large upload or download, your ping can spike from 30ms to 300ms because cheap router buffers hold packets in a queue. Run the free Waveform Bufferbloat Test while the line is busy. If it grades poorly, enabling SQM or a “smart queue” or “adaptive QoS” feature on a capable router flattens those spikes far more effectively than any legacy tweak.
Pick the right server. Because distance is the biggest fixed cost, manually selecting the closest datacenter in games that let you is often a larger win than every local setting combined. Check your game’s server or region menu before you touch anything else.
Set your DNS and update firmware. Point your router or PC at a fast resolver like Cloudflare’s 1.1.1.1, and keep router firmware current. DNS will not lower in-game ping, but it speeds up matchmaking and connection setup. Our best DNS for gaming comparison has the current benchmarks.
Trim the local adapter, carefully. Disabling power management on your network adapter so Windows cannot idle it is safe and worthwhile. Be skeptical of copy-pasted command lists, though. Old guides still tell you to run netsh int tcp set global chimney=enabled or netdma=enabled; those features were removed from modern Windows and the commands do nothing. Likewise, the CS:GO era cl_cmdrate and cl_updaterate console commands no longer apply in CS2, which manages rates automatically. Our network adapter settings guide sticks to the tweaks that still exist.
Enable the modern low-latency paths. On Windows 11, the newer latency features do more than the decade-old registry hacks. We covered what actually changes in the Windows 11 low latency profile writeup. Pair that with Reflex where the game supports it.
Still watching your ping bounce around?
If the fixes above did not settle your connection, the trouble may live in the hops between you and the server, where you have no control. Route your traffic through WTFast’s gaming network for a few days and see whether a cleaner path steadies it.
The short version
Your game is two machines negotiating a shared reality several dozen times a second, and the server always wins the argument. Some of the delay in that negotiation is physics you cannot beat, some is tick rate and lag compensation the developer chose for you, and some is routing, bufferbloat, and a wired connection sitting entirely in your hands. Start with the parts you own. Go wired, kill bufferbloat, pick the nearest server, and skip the dead console commands. When your ping is still high after all of that and the traceroute shows an ugly path to the server, the problem has left your house, and rerouting is the only lever left. Everything else is you fighting the speed of light, and that one you will not win.