You ran a speed test, it came back at several hundred Mbps, and the game is still rubber-banding. That result was never evidence about your game — it measured a different quantity, over a different path, to a different server. Here is what your game actually reacts to, and which numbers replace the one you just ran.
First, find your case
| What you are seeing | Most likely cause | What to do next |
|---|---|---|
| Great Mbps, lag on every server region | Local: Wi-Fi, a busy uplink, or your ISP's own line | Go wired, pause uploads, retest all regions |
| Great Mbps, only some regions play badly | The route to those regions, not your connection | Play on a region that tests clean |
| Fine alone, bad when someone streams | Your uplink is full and game packets are queueing | Test once idle and once under load, compare |
| Low average ping, still feels choppy | Latency swings that an average hides | Read the latency range, not the median alone |
| Teleporting, swallowed inputs | Packet loss | Rule out Wi-Fi first, then compare regions |
| Every network number clean, still stuttering | Not a network problem | Check frame time, device load, the game's own status |
A speed test answers a question your game never asks
Mbps is how much data your line can move per second. What a game needs is how long one small packet takes to get there and come back, measured in milliseconds. They are different units because they are different problems.
| What a speed test reports | What a game actually depends on |
|---|---|
| Peak throughput to a nearby test server | Round-trip time to the game's server region |
| A few seconds of bulk transfer | Consistency across a 30-minute match |
| Ping measured on an otherwise idle line | Latency while the line is also doing other things |
| Whether the pipe is wide | Whether every small packet arrives, and arrives on schedule |
A download fills a pipe: the wider it is, the sooner a large file finishes. A match sends a thin stream of tiny packets, tens of times per second, and every one of them is urgent. Widening the pipe does not make a packet arrive sooner, the same way adding lanes to a road does not shorten it. That is why a connection can be measured at 500 Mbps and still deliver a late shot — the capacity was never the part that was failing.
The speed test picked the closest server. Your game did not.
This is the part that catches most people out, and it is structural rather than bad luck.
A speed test picks a server for you, and it optimizes for proximity — typically the nearest one, often inside your own ISP's network. That is the correct design for measuring a line's capacity, and it means the test finishes without ever leaving your city or your provider.
Your game server is not chosen that way. It sits in a fixed region, and your packets have to cross whatever transit and peering lies between you and it. So the stretch of path that decides whether you can play — cross-region transit, congested peering points, a provider routing you the long way round — is exactly the stretch the speed test never touched. A clean score tells you the first mile is healthy. Lag usually lives after it.
The way to see this is to measure several regions at once and put the results next to each other. One number on its own has nothing to be compared against; eight of them show you the shape of your routing immediately.
Averages hide the two things that actually lose you a round
A speed test reports an average over a few seconds of bulk transfer. Averages are the right tool for throughput and the wrong tool for a game, because what breaks a fight is never the average — it is one moment inside it.
- A dropped packet. Your line can be averaging a comfortable 30 ms and still lose the packet carrying your shot. In game that reads as a rubber-band, a swallowed input, or a hit that registers a beat late.
- A spike. Sitting at 30 ms and jumping to 150 ms for a fraction of a second feels worse than sitting at 60 ms the whole time. Games are built around consistent timing, and an average smooths the inconsistency out of sight.
This is also why single-shot tests disagree with each other, and why the measurement here is built to catch the moments rather than smooth them. Each region gets 3 warm-up packets that are thrown away, then a round of 20 sent on a fixed 50 ms schedule — fixed, because waiting for each reply before sending the next one barely fills a queue and systematically under-reports loss. Drop even one packet, or come back with a jumpy round trip, and that region is automatically re-measured for 2 more rounds and scored on all 60 samples. Twenty packets can only resolve loss in steps of 5 percent; sixty brings that down to 1.67 percent, which is the difference between a verdict and a guess.
A fast line still queues when it is full
High bandwidth can even work against you here. When an upload, a cloud backup or another person's stream saturates the uplink, your game packets wait in a queue behind that traffic — and the bigger the buffer in the equipment, the longer the wait. Ping climbs, sometimes by a lot, while a speed test run afterwards on an idle line comes back looking perfect. It is a common reason a household reports lag that nobody can reproduce an hour later.
You can settle it in two minutes: test one region while the line is idle, then start a large upload and test the same region again. If latency and loss jump under load, the fix is queue management on your router or scheduling the heavy traffic elsewhere, not a bigger plan.
What to measure instead
Four things, in this order:
- Compare regions, do not test one. Run every region in one pass. All of them bad points at your own network; some of them bad points at routing, and switching regions fixes that today.
- Read the range, not a single number. A region that sits between 28 and 34 ms is a different proposition from one that swings between 25 and 140 ms, even when both average out the same.
- Read packet loss next to it. Low latency with loss still plays badly. Any loss at all is worth a second look, which is why a single dropped packet triggers a re-measure here.
- Test in the hours you actually play. Peak-hour congestion is the whole point of the question, and a Sunday-morning result will not show it.
Three numbers do the work, and they are kept separate rather than mashed into one score. For a general real-time match, a median under 25 ms is excellent and under 50 ms is comfortable, with the experience degrading past roughly 90 ms — competitive shooters are graded more tightly than relaxed genres, and these bands are our engineering interpretation rather than a rating published by any studio. The P10–P90 range shows how far the samples spread, and it is the number that best explains why a ping that looks fine can still feel choppy. Packet loss should be 0%, and anything at or above 5% pulls the rating down sharply.
There is a fifth check worth keeping in reserve: if every network number comes back clean, stop spending time on the network. Frame-time spikes, shader compilation, a thermally throttled GPU or a struggling game service all produce stutter that feels exactly like lag. A clean multi-region result is genuinely useful here — it tells you where not to keep looking.
And if the whole squad is affected rather than just you, the question changes shape: you need the region that stays acceptable for the slowest member, not the fastest. That logic is in How to Pick the Best Server Region for a Whole Group.
What this test measures
It measures the round trip from your browser to our probes in 8 regions, over UDP, in the browser, with nothing to install. Use those numbers to compare directions: which regions your provider routes well, which ones are worth avoiding tonight, and whether the problem follows you everywhere or only points one way. The number your game client displays is the final word on a given match, because a client reaches a publisher's servers by its own route — what you get here is the comparison that tells you which route is worth being on. Are Ping Test Results Accurate? goes through the measurement itself in detail.
Stop asking a speed test about your game. Run every region once and the answer is in front of you in under a minute.