Halloween at Nexthosting: 10% off game servers with the code – until November 1
Why p99 latency often matters more than raw Mbps for gamers and game server operators. Measurement methods, p99 explained, and steps to cut latency.
Bandwidth indicates how much data your connection can carry per second, measured in Mbps or Gbps. Latency measures how long a single packet takes to reach its destination and come back, given in milliseconds. For streaming and large downloads, bandwidth is what matters most. For gaming, voice chat, and anything that has to respond in real time, latency determines how the game feels, not line capacity.
In short:
- Low latency is critical for real-time applications such as gaming or voice chat, while bandwidth mainly matters for large downloads or streaming.
- Real-world data throughput is usually below the nominal bandwidth because protocol overhead, Wi-Fi interference, and server load limit it.
- To judge the actual gaming or application experience, it is more important to measure p99 latency and jitter than just the average.
- When testing, you should take several measurements at different times to get meaningful real-time values for ping, jitter, and packet loss.
- For latency-critical applications, server proximity and routing quality matter more than simply adding bandwidth.
Bandwidth describes the maximum amount of data a connection can theoretically transfer per second. A fiber connection rated at 1,000 Mbps promises exactly that: one gigabit per second, under ideal conditions. In practice, the full figure never arrives, because protocol overhead, encryption, and network conditions use up part of the capacity.
Real-world throughput, meaning what actually arrives, is therefore almost always below the nominal value. TCP headers, Wi-Fi interference, devices active at the same time in the household, and the load on the server at the other end all take their share. If you book 500 Mbps and measure 420 Mbps in a speed test, you don't have a faulty connection, just a perfectly normal result.
A look at typical scenarios helps in practice:
Bandwidth is therefore a matter of capacity, not of the speed of individual data packets. This very mix-up leads to bad purchases when users reflexively upgrade to the next-higher plan because of stuttering in an online game, even though the problem lies elsewhere.
Latency is the time a data packet needs to get from your device to the server. The ping value that most tools display actually measures the round trip time (RTT): the time for the outbound and return trip combined. A ping of 20 ms therefore means that a packet needs roughly 10 ms to reach the destination and 10 ms to come back, although the split is rarely exactly symmetrical.
The average of a ping measurement often hides the actual problem. If 95 out of 100 packets arrive at 18 ms but 5 outliers sit at 200 ms, the average shows an unremarkable value while the gaming experience still stutters. That is why experienced network engineers work with percentiles instead of averages: p50 is the median, p95 shows how bad the worst 5% are, and p99 exposes the truly critical outliers that delay a headshot in a shooter.
Jitter describes the variation in latency between individual packets. A ping that stays constant at 25 ms feels better than one that fluctuates between 15 and 60 ms, even if the average is identical. Packet loss, in turn, means that packets don't arrive at all and have to be requested again, which leads to stuttering, rubberbanding, or complete disconnections.
| Metric | What it measures | Typical effect of a bad value |
|---|---|---|
| Ping/RTT | Outbound and return trip of a packet | Noticeable delay on inputs |
| Jitter | Variation in latency | Uneven, unpredictable stuttering |
| Packet loss | Share of lost packets | Freezes, rubberbanding, disconnects |
| p99 latency | Latency of the worst 1% | Rare but extremely disruptive spikes |
A stable ping with low jitter often delivers a better gaming feel than a lower but more erratic average, as studies on ping stability also suggest.
More bandwidth does not automatically solve a latency problem. This is the misconception behind most wasted investments: switching from 100 to 500 Mbps noticeably helps when downloading large files, but usually changes nothing about the ping to the game server, as Digi International also points out.
The bandwidth-delay product shows how the two values are actually related. It describes how many bits can be in flight "on the wire" at the same time before an acknowledgment comes back. The formula is simple: bandwidth multiplied by latency. At 100 Mbps and 10 ms of latency, the result is a product of around 1,000,000 bits that are unacknowledged in the network at any given time. If this buffer is too small for the line speed, TCP throttles artificially even though bandwidth would still be free.
It is easy to tell which factor is the limit in each case:
Pro tip: If you're unsure which factor is causing your problem, run a download in parallel while a game is running. If the ping collapses as soon as the line is saturated, traffic prioritization often helps more than a plan upgrade.
Latency is made up of four different delays, and only one of them cannot be optimized away because of physics.
Propagation delay results from the pure signal travel time. Light travels through fiber at around 200,000 kilometers per second, slower than in a vacuum. This limit is absolute: a server on another continent will never deliver the same ping as one that is 50 kilometers away, no matter how much money you put into the connection.
Transmission delay describes how long it takes to push a packet completely onto the line. It depends on packet size and available bandwidth: larger packets take longer to transmit fully, and more bandwidth shortens this time. For small gaming packets of a few hundred bytes, this share hardly matters, but for large data streams it does.
Processing and queueing delay arise inside the network devices themselves:
These four components add up to the total latency shown in the ping tool. If you want to lower latency, you therefore need to think not only about the line but also about server location, hardware, and routing.
Values can only be put in context if you know what is normal for the respective technology. Fiber (FTTH) runs at 11 to 14 ms of latency under good conditions, with cable connections somewhat higher.
| Connection type | Typical latency | Typical bandwidth |
|---|---|---|
| Fiber (FTTH) | 11–14 ms | Up to several Gbps |
| Cable (DOCSIS) | 15–30 ms | 50–1,000 Mbps |
| DSL | 20–30 ms | 16–200 Mbps |
| 5G (sub-6) | low latency range, from single-digit to low double-digit milliseconds | moderate to high bandwidth |
| LEO satellite | 20–60 ms | 50–200 Mbps |
| GEO satellite | 450 ms | 10–100 Mbps |
For smooth gaming, values below 30 ms are considered very good and up to 60 ms are usually still unnoticeable; above that, input delay becomes noticeable. Streaming is much more tolerant of latency; what counts here is mainly constant bandwidth without dropouts.
These values are reference points, not guarantees. At peak times, with overloaded nodes, or in heavily used mobile cells, real-world measurements can be significantly worse than the technical averages for the respective connection type.
Most users rely on a single speed test and draw the wrong conclusions from it. A clean measurement takes more than one click.
ping command or mtr show fluctuations over time.A common mistake is measuring at the wrong time: a test at three in the morning almost always shows better values than one during the evening peak. Your own hardware also skews results; an old Wi-Fi module or an overloaded home network with four devices streaming at once makes any measurement worthless, without the connection itself being to blame. Repeat measurements over several days and at different times before you make a diagnosis.
Reducing latency and increasing throughput are two different jobs that are rarely solved by the same measure.
For latency, what counts most is proximity to the server and the quality of the route to it. If you run a game server with users from all over Europe, you should choose a data center with good connectivity to major internet exchange points rather than the cheapest location. Using a CDN helps with static web content, but does little for real-time game data, since direct server-client connections are required there.
Concrete adjustments for everyday use:
With packet size there is a real trade-off: small packets cause more header overhead per transmitted byte, but reduce the time until a single packet is fully transmitted, which is often more important for real-time applications than maximum bandwidth utilization.
Pro tip: Before you invest in more bandwidth, use traceroute to check whether the delay originates in your home network at all, or only several hops further along at your provider. This saves money and shows where a change will actually make a difference.
With game servers, the difference between bandwidth and latency is especially clear. A server with excellent connectivity but a poor location does little for players in the same region, because signal travel time over long distances can hardly be compensated for. Tick rate and server hardware additionally influence how quickly game states are calculated and sent back to clients, which directly affects the processing component of latency.
For operators who do the work themselves, the following applies:
If you run your own infrastructure as a community admin, Nexthosting offers background material on latency-relevant configuration questions for various game servers.
Latency takes priority when real-time response is what matters; bandwidth counts for large amounts of data. This rule of thumb settles most cases of doubt.
To get started right away:
| Situation | Most important factor |
|---|---|
| Online gaming, voice chat | Latency and jitter |
| Streaming, downloads | Bandwidth |
| Cloud gaming | Both at once |
| Web browsing with many resources | Latency per connection |
Monitoring is not a one-off project. Networks change due to load, new devices, and provider maintenance, so recurring measurement is worthwhile instead of a one-time diagnosis.
If you still reflexively reach for more bandwidth when a game stutters, you will rarely solve the actual problem. p99 latency deserves significantly more attention than the average, because it is these outliers that ruin the gaming experience, not the median. I would advise every operator to switch monitoring to percentiles before thinking about hardware upgrades.
Wi-Fi 6 and 5G bring real latency improvements, but they don't replace a wired connection for competitive gaming. The physical limit set by signal travel time remains, regardless of the wireless standard. For game server communities, server proximity to the player base remains the lever with the greatest effect, greater than any client-side tuning. If you take server location and peering seriously, you solve more latency problems than any router purchase ever will.
— Erik
Server location, NVMe storage, and well-designed DDoS protection are exactly the factors described in this article as decisive for latency, and that is exactly where Nexthosting puts its focus. Instead of pushing bandwidth as a selling point, the emphasis is on proximity to your community, stable hardware, and personal support when a server buckles under load. If you run Rust, take a look at the Rust server plans; for Counter-Strike communities, see the CS2 server options. If you have more flexible requirements and want to run your own VPS for latency-critical applications, the Nexthosting homepage gives an overview of suitable products, and you can configure a server directly.
Low latency is always better because it means shorter response times. In gaming and voice chat, even a difference of 20 to 30 ms noticeably affects how the game feels.
On Wi-Fi, values below 30 ms are considered good, and up to about 60 ms is usually still acceptable for gaming. A wired connection generally delivers more stable and lower values than Wi-Fi.
Latency is the time a data packet needs to travel from the sender to the receiver, measured in milliseconds. The commonly used ping value measures the round trip time, meaning the outbound and return trip combined.
Bandwidth is the theoretical maximum capacity of a line, while throughput describes the amount of data that actually arrives under real-world conditions. Protocol overhead, encryption, and network load mean that throughput is almost always below the nominal bandwidth.