The standard advice when cloud gaming feels sluggish is to get faster internet. It is usually wrong, and it is wrong in an instructive way: bandwidth and latency are unrelated quantities, and cloud gaming is almost entirely a latency problem.
A 1 Gbps connection with a 60 ms round trip will feel worse than a 50 Mbps connection with a 12 ms round trip. Understanding why means following a single button press all the way there and back.
The full chain
Press the jump button. Here is everything that has to happen before you see the character leave the ground.
| Stage | Typical | What it is |
|---|---|---|
| Controller to device | 1-10 ms | Wireless polling interval; wired is lowest |
| Client processing | 1-4 ms | Reading input, packaging it |
| Network to server | 5-40 ms | One way, mostly distance and routing |
| Server input handling | 1-8 ms | Waiting for the next simulation tick |
| Game renders a frame | 8-16 ms | The frame you asked for |
| Video encoding | 4-12 ms | Compressing that frame |
| Network back to you | 5-40 ms | The return trip |
| Decode and display | 8-25 ms | Decompress, then wait for the panel to refresh |
Add the middle of those ranges and you land somewhere near 60 to 90 ms. Local play on the same hardware would be roughly 25 to 50 ms. The extra 30 to 50 ms is the entire cloud gaming penalty, and notice how little of it is the internet.
Bandwidth determines whether the picture looks good. Latency determines whether the game feels good. They are different measurements, and paying for more of the first does nothing for the second.
Why distance is the hard floor
Data in fibre travels at roughly two-thirds the speed of light. That gives you about 1 ms of one-way delay per 200 km of cable, so 2 ms round trip per 200 km at the absolute theoretical best.
Real routing is never a straight line, and every router adds a little. A practical rule is 1 ms round trip per 50 to 70 km of geographic separation.
- Server 100 km away: roughly 4-8 ms of unavoidable network delay
- Server 800 km away: roughly 20-30 ms
- Server on another continent: 120 ms and up, and no amount of money fixes it
This is why the providers build regional data centres rather than a few large ones, and why coverage maps matter far more than advertised bandwidth. If the nearest server is 2,000 km away, the service will not feel good on any connection.
The delays you can actually remove
Most people attack the network, which is the part they control least. Here is the same list ordered by how much you can improve it.
1. Get off wireless, both ends
Wi-Fi adds latency and, worse, adds variable latency. Jitter is more damaging than a slightly higher average, because the client has to buffer for the worst case.
Ethernet to the console or PC is the single biggest improvement available to most people — frequently 10 to 30 ms, and a large reduction in the variance that produces those intermittent lurches.
A wired controller helps too, typically a few milliseconds. Small, but free.
2. Fix your display
This is the most overlooked item, and often the largest single contributor.
Many televisions add 20 to 60 ms of processing latency for motion smoothing and image enhancement. Game mode disables most of that. On a lot of sets, enabling game mode improves the feel more than any network change could.
Refresh rate matters as well. A 60 Hz panel adds up to 16 ms of waiting for the next refresh; 120 Hz halves it.
3. Reduce buffering under your own control
Anything else saturating your connection forces the stream to buffer more defensively. A large download, a backup, or someone streaming video in 4K in the next room all show up as input lag. If your router supports it, enable Smart Queue Management or fq_codel, which specifically targets the buffer bloat that causes latency spikes under load.
4. Check which region you are on
Providers usually pick automatically and do not always pick well. If the service exposes the server region, verify it is the nearest one. Being routed to a data centre two countries away is more common than you would expect.
5. Turn the resolution down
Counter-intuitive but effective. Lower resolution means less to encode and less to decode, which cuts both ends of the chain. Dropping from 4K to 1080p can save 10 to 20 ms. On a marginal connection this is a straight trade of sharpness for responsiveness, and for anything competitive it is worth taking.
Which games work and which do not
The tolerance varies enormously by genre, and averages hide it.
Works well. Turn-based games, strategy, city builders, role-playing games, adventure games, most single-player third-person action. Anything where a 70 ms delay is inside your own reaction time anyway.
Works acceptably. Racing, single-player shooters, action games with generous timing windows. Noticeably different from local play, entirely playable after ten minutes of adjustment.
Does not work. Competitive shooters, fighting games, rhythm games, anything with frame-precise inputs. Not because it is unplayable, but because you are at a structural disadvantage against local players, and in a fighting game a 50 ms handicap decides the match.
What providers do to hide it
The engineering effort is impressive even when the physics is uncooperative.
Regional edge deployment. The only real fix for distance. Put the compute physically closer.
Hardware encoders. Dedicated silicon encoding a frame in 3 to 5 ms instead of tens of milliseconds in software.
Negative-latency prediction. Speculatively act on the input the player is most likely to make next, then correct. Works better than it sounds for continuous inputs like steering; less well for discrete ones.
Adaptive bitrate. Degrade image quality instantly rather than buffering, because a brief soft-looking frame is far less disruptive than a stutter.
Frame pacing. Delivering frames at even intervals rather than as fast as possible. Slightly higher average latency, much lower jitter, and it feels better.
How to test your own setup
Two numbers are worth knowing.
Round trip to the nearest data centre. Ping the service's region endpoint if you can find it, or use the provider's built-in network test. Under 20 ms is excellent, 20-40 ms is good, 40-80 ms is workable for the right genres, above 80 ms is going to be frustrating.
Latency under load. Run a speed test that reports latency while saturating the line, or start a large download and ping during it. If your idle ping is 15 ms and your loaded ping is 200 ms, you have buffer bloat, and fixing it will transform the experience more than any upgrade.
The realistic verdict
Cloud gaming has moved from a curiosity to something genuinely good for a large fraction of games, and the reason is edge deployment rather than faster home connections. If you live near a data centre, use a cable, and enable game mode on your display, the experience for most single-player titles is close enough to local that you will stop noticing.
If you are far from any region, none of the other advice will save it, and no upgrade you can buy will either. Distance is the one part of the chain that engineering cannot argue with.