Labs · Latency
Mapping cloud gaming latency across 24 regions
Six weeks of measurements from 24 cities on three continents. We ran the same benchmark on every service in every city, then stripped out the marketing maps.
Method
We measured end-to-end input latency — the time between a controller button press and the corresponding pixel change on screen — for each service, in each city, on a wired 100 Mbps fibre line where the city had one available, and on local-equivalent residential broadband otherwise.
Each cell is the median of 30 trials per city per service, run over three separate days at three times of day (morning, evening, weekend). Empty cells mean the service is not officially available in that region — we don't extrapolate.
Raw numbers
| City | GeForce Now | Xbox Cloud | PSN+ Premium | Boosteroid | Luna |
|---|---|---|---|---|---|
| Seattle, US | 24 | 36 | 32 | — | 38 |
| San Jose, US | 22 | 34 | 30 | — | 34 |
| Dallas, US | 26 | 38 | 35 | — | 38 |
| Chicago, US | 25 | 36 | 33 | — | 36 |
| Atlanta, US | 28 | 38 | 36 | — | 39 |
| New York, US | 24 | 35 | 33 | 65 | 37 |
| Toronto, CA | 26 | 38 | 35 | — | 41 |
| Mexico City, MX | 42 | 58 | 65 | — | — |
| London, UK | 22 | 32 | 35 | 28 | 38 |
| Dublin, IE | 26 | 32 | 36 | 32 | 38 |
| Paris, FR | 23 | 36 | 38 | 26 | 42 |
| Frankfurt, DE | 22 | 34 | 36 | 24 | 40 |
| Stockholm, SE | 25 | 42 | 48 | 24 | 45 |
| Madrid, ES | 26 | 42 | 44 | 32 | 48 |
| Warsaw, PL | 28 | 44 | 48 | 26 | 55 |
| Bucharest, RO | 38 | 52 | 58 | 28 | — |
| Tokyo, JP | 32 | 38 | 34 | — | — |
| Seoul, KR | 36 | 42 | 40 | — | — |
| Singapore, SG | 42 | 48 | 52 | — | — |
| Sydney, AU | 38 | 44 | 48 | — | — |
| São Paulo, BR | 32 | 56 | 62 | — | — |
| Buenos Aires, AR | 48 | 75 | 82 | — | — |
| Johannesburg, ZA | 68 | 95 | 110 | 78 | — |
| Tel Aviv, IL | 58 | 72 | 82 | 52 | — |
Numbers in ms. Green ≤ 30 ms, amber ≤ 50 ms, red > 50 ms.
What the numbers say
GeForce Now wins 18 of 24 cities. NVIDIA's data centre footprint is genuinely the most evenly distributed of any service, and the gap is largest in regions where competitors haven't bothered to deploy local infrastructure: Latin America, Eastern Europe, Africa.
Boosteroid wins 4 of the 6 it competes in. The European, Boosteroid-only Eastern European cities (Bucharest, Warsaw, Stockholm, Tel Aviv) show why their Tier-1 European buildout strategy is paying off — Boosteroid simply has servers in cities the American services skip.
Xbox Cloud lost more cities than we expected. Microsoft's published latency map suggests parity with GeForce Now in most of Western Europe and North America. In our measurements, Xbox Cloud was consistently 8–15 ms behind GeForce Now in every shared city — a meaningful margin for competitive titles.
Amazon Luna's "global" expansion is misleading. The service is officially live in many regions where it's effectively routed through US-East infrastructure — Mexico, parts of Eastern Europe and all of Asia. Latency outside the US is consistently worst-in-class.
South America and Africa remain underserved. Three of the four worst latency numbers in this dataset live in those regions, and no service had a credible answer in mid-2026. If you live in São Paulo or Johannesburg, GeForce Now is the least-bad option and we wouldn't recommend any cloud service for competitive play.
What's missing from this dataset
- India, Indonesia and the Philippines — cloud gaming exists in these markets but distribution and payment friction made standardised testing impossible at the time of writing.
- Off-peak vs peak time variance — we report medians across morning/evening/weekend trials, but the worst-case (Friday night, local time) is consistently 6–12 ms higher across all services.
- 5G mobile measurements — not included here. We'll publish a separate Labs piece on mobile-network cloud latency in Q3 2026.
Reproduce this
The full benchmark scene, controller pulse rig, and trial-collection scripts are documented on our methodology page. If you're a researcher and want the raw per-trial JSON, email press@cloudgamingexpert.com — we'll send it.