You pay for 300 Mbps and your speed test says 310. Then your Zoom call freezes the moment someone else in the house starts uploading vacation videos. The line isn't slow. Your packets are stuck in a queue, and the speed test never measured the queue.
Our take: buying a faster plan is the wrong first fix for lag. Bufferbloat comes from how the queue is managed, not from a lack of bandwidth. A free router setting (SQM with fq_codel or CAKE) can often fix what paying extra each month for a faster plan won't, as of 2026.
What is bufferbloat, and why is it a delay problem instead of a speed problem?
Every modem and router has a buffer, a waiting room for packets that arrive faster than the line can send them. A small buffer absorbs short bursts. A huge one holds packets for a long time instead of dropping them, so the line keeps getting longer and every packet that joins it waits.
Think of a highway. Bandwidth is the number of lanes. Latency is how long you sit on the on-ramp. Adding lanes does nothing if the on-ramp is backed up for a mile.
TCP makes this worse. TCP is the protocol behind uploads, downloads and backups, and it keeps speeding up until it sees a dropped packet. A bloated buffer hides those drops for a long time, so TCP fills the buffer to the brim.
Here's the math. Say your upload speed is 20 Mbps and your modem buffers 1 MB. One megabyte is 8 megabits, and at 20 Mbps that takes 400 ms to drain.
How it plays out with a bloated queue:
- The upload starts. A cloud backup sends as fast as TCP allows.
- The modem buffer fills. A full 1 MB buffer on a 20 Mbps uplink holds about 400 ms of data.
- A game packet joins the back of the line. It's tiny, but it waits behind everything already queued.
- It leaves about 400 ms late. Your shot lands after your opponent has already moved.
The same upload with a managed queue (fq_codel):
- The upload starts. Same backup, same speed.
- The traffic is split into flows. The backup goes in one flow and the game in another, and they take turns.
- Packets get dropped early. When queue delay stays above the 5 ms target for 100 ms, CoDel drops a packet and TCP slows down.
- The game packet waits a few milliseconds. That's single digits, not hundreds.
Speed tests miss all of this. They report throughput, and they usually measure it with nothing else competing for the line. Your connection only bloats when it's busy, so a speed test run on a quiet line can't catch it.
Why does my 300 Mbps internet still lag on Zoom and in online games?
Calls and games barely use bandwidth. They send a steady stream of small packets, and each one has to arrive on time. A Zoom HD call needs a few Mbps. What it can't handle is a voice packet arriving a third of a second late.
The weak spot is almost always your upload. As of 2025, a cable plan sold as 300 Mbps often uploads at only about 10 to 35 Mbps, depending on the provider. The smaller the pipe, the faster one big upload fills its queue.
There's a long-standing standard for how much delay a conversation can take. ITU-T G.114, One-way transmission time, was approved in May 2003 and is still in force. As of 2026, it is generally read as recommending one-way delay of about 150 ms or less for most interactive uses. Its Amendment 1, from September 2003, added Appendix II with guidance on one-way delay for Voice over IP, which is exactly the kind of traffic a Zoom call is. As of 2026, Zoom's published network guidance suggests latency of about 150 ms or less.
Now do the math. In this example, 228 ms of upload bloat would exceed the roughly 150 ms one-way budget that G.114 is generally read as recommending, before your voice has even left your house.
How many milliseconds of extra latency under load counts as bufferbloat?
The number that matters is loaded latency minus idle latency, in milliseconds. Idle latency is your ping on a quiet line. Loaded latency is your ping while the line is busy downloading or uploading. The gap between them is the bloat.
The grades below are this site's rules of thumb, set against the G.114 voice delay budget. They aren't an official standard. As a rough rule of thumb as of 2026: under about 30 ms extra is fine for almost everything, 30 to 100 ms is noticeable, and over 100 ms is a problem.
Say your idle ping is 12 ms and it jumps to 240 ms during an upload. That's 228 ms of bloat. Grade: broken.
Each use handles bloat differently:
- Zoom and video calls are hurt first. Delay shows up as people talking over each other, then as frozen video.
- Competitive gaming is the least forgiving.
- 4K streaming is the most forgiving, because the player buffers ahead. As of 2025, Netflix recommends about 15 Mbps for 4K streaming. Heavy bloat still slows startup and causes rebuffering, because the player's small requests get stuck behind bulk traffic.
Grade your extra latency under load
| Extra latency under load | Grade | Zoom and video calls | Competitive gaming | 4K streaming |
|---|---|---|---|---|
| Under 5 ms | Fine | Clean | Clean | Clean |
| 5 to 30 ms | Fine in practice | Clean | Slight jitter, rarely noticed | Clean |
| 30 to 100 ms | Noticeable | Occasional talk-over and freezes | Visible lag spikes | Slow start, rare stalls |
| Over 100 ms | Broken | Frozen video, choppy audio | Unplayable while others upload | Frequent rebuffering when others are active |
Site grading, framed against the ITU-T G.114 one-way delay recommendation (May 2003).
How do I test for bufferbloat in 15 seconds?
Use a browser test that measures latency three ways: idle, while downloading, and while uploading. As of 2026, the Waveform bufferbloat test does this and gives a letter grade in well under a minute. Cloudflare's speed test shows the same split under "latency during download" and "latency during upload".
- Plug in with an Ethernet cable. A Wi-Fi test mixes radio interference in with queue delay.
- Pause anything else that's using the line: TVs, consoles, phone backups.
- Run the test and write down three numbers: idle, download-loaded and upload-loaded.
- Subtract idle from the higher of the two loaded numbers, then find your row in the grade table above.
Your result: idle ___ ms, loaded ___ ms, extra ___ ms.
As a rough guide as of 2026, a healthy connection gets close to your plan speed and stays within about 30 ms of idle latency under load. If the speed is fine and the latency jumps by 150 ms, you've found your problem, and it isn't your plan.
There's also a newer metric. The IETF's IP Performance Measurement working group has been developing "Responsiveness under Working Conditions", which reports results in RPM (round trips per minute). Higher is better: 60,000 divided by the loaded round-trip time in milliseconds. As of 2026, on macOS 12 Monterey and later, typing networkQuality in Terminal runs this test.
Reading a loaded-latency result: the worked example
- 12 msIdle latency in the example, line quiet
- 240 msUpload-loaded latency in the example
- 228 msExtra latency under load: graded broken
- 30 msRoughly the most extra latency a healthy line adds
Why does my internet lag only when someone else is uploading or streaming?
Everyone in your home shares one uplink and one queue. When a single device starts uploading, it fills that queue for everyone else.
Upload bloat is usually worse than download bloat. A buffer the same size drains much more slowly on a 20 Mbps uplink than on a 300 Mbps downlink.
Common culprits:
- Phone photo backups to iCloud or Google Photos
- Console and PC game updates (the TCP acknowledgments going back up add to the queue)
- Screen sharing on a video call
- A Plex server or a security camera streaming out
This is also why video can take ages to start on a fast plan. The player's small requests, and the acknowledgments its download depends on, are stuck in line behind someone else's bulk data.
Is 42% packet loss bad, and is it the same thing as bufferbloat?
Yes, 42% is severe. As of 2026, loss above roughly 1% can start to show on calls and games, and Zoom's guidance suggests staying under about 2%. But heavy packet loss is a different fault from bufferbloat. Bloat is mostly delay with little loss. Losing packets even on a quiet line points to a bad line, a failing modem, damaged coax or phone wiring, or Wi-Fi interference.
Work from your symptom:
- Lag only on calls, and idle ping looks fine: probably upload bloat. Next step: run the loaded test and check the upload-loaded number.
- Lag whenever someone uploads or streams: bufferbloat. Next step: turn on SQM (see the fix section below).
- Constant packet loss, even when the line is quiet: a line or Wi-Fi fault. Next step: test over Ethernet. If the loss is still there, call your ISP and ask for a line check.
- Low speed test result, but latency stays flat: your plan or network congestion. Next step: compare against your plan's speed at different times of day.
Jitter, the variation in delay, is its own problem. ITU-T added Appendix III to G.114 in Amendment 2 (November 2009), covering delay variation on unshared access lines. Bloat causes jitter too, because the queue keeps filling and draining. A spiky ping on its own can't tell you which fault you have. The loaded test can.
Does a faster plan fix bufferbloat, and does the connection type matter?
A faster plan drains a full buffer more quickly, so it can soften the symptom. It doesn't remove the queue, and plenty of gigabit connections still fail the loaded test because their upload is a fraction of their download.
Every connection type can bloat. Older DSL and cable with a low upload speed tend to be the worst. Fiber with matching upload and download speeds hides it best. The FCC's Measuring Broadband America reports have tracked latency under load by technology and keep finding the same pattern: fiber's latency rises least under load, while cable and DSL rise more. Cable has a built-in improvement. CableLabs' DOCSIS 3.1 specifications, first released in 2013, call for cable modems to support active queue management on the upstream (a variant called DOCSIS-PIE), as of 2026. How well it works still depends on your modem and your ISP's settings.
The trap: the bloated buffer usually sits in the modem or router you rent, not deep inside your ISP's network. Changing your equipment often helps more than changing your plan.
Our view: pay for the speed you actually use, and fix delay in your settings.
How do I fix bufferbloat with SQM, fq_codel or CAKE on my router?
The trick is to move the queue. Your devices send traffic through your router, then your modem, then out to your ISP. Without a shaper, the bottleneck is the modem's link to the ISP, so the queue builds in the modem's unmanaged buffer. Set your router's shaper slightly below your measured line rate, and the router becomes the bottleneck instead. The queue now forms inside the router, where fq_codel or CAKE keeps it short.
Say you measured 300 Mbps down and 20 Mbps up. Set the shaper to 90%: 270 Mbps down and 18 Mbps up. As a rough estimate as of 2026, you give up about 5 to 15% of your headline speed. Few people notice that loss. Everyone notices when the lag goes away.
Where to find the setting:
OpenWrt: the SQM package (luci-app-sqm)
OPNsense: shapers in the traffic shaper menu
Ubiquiti: Smart Queues
Other routers: look for "SQM", "Smart Queue" or "QoS" in the settings
fq_codel (described in RFC 8290, published in 2018) gives each traffic flow its own queue and serves them in turn, so one big upload can't starve a game. CoDel (RFC 8289) drops packets early once queue delay stays above its target, which tells TCP to slow down before the queue grows. CAKE, part of Linux since kernel 4.19 (released in 2018) as of 2026, bundles the shaper, fair queuing and overhead handling into one setting. On OpenWrt it's the simpler choice.
Common mistakes:
- Setting the shaper at your advertised speed rather than your measured speed
- Testing over Wi-Fi after the change
- Setting only the download and forgetting the upload, which is where most bloat lives
Fixes ranked by effort
| Step | What to do | Effort |
|---|---|---|
| 1 | Turn on the router's built-in SQM or Smart Queue feature | Minutes |
| 2 | Choose CAKE or fq_codel as the queue algorithm | Minutes |
| 3 | Set download and upload shapers to about 85 to 95% of measured line rate | 10 minutes |
| 4 | Retest loaded latency and nudge the percentage | 10 minutes |
| 5 | Replace a rental modem or router that lacks SQM | Hours and cost |
Site guidance.
Frequently asked questions
Does Wi-Fi cause bufferbloat?
Wi-Fi can bloat too, because access points and drivers have their own queues. But most home bufferbloat sits on the upload link to the ISP. Test over Ethernet first to see which one you have.
Will a VPN fix bufferbloat?
No. A VPN wraps your traffic, but it still goes through the same queue in your router and modem. Expect the same loaded latency, sometimes slightly more.
Do I need to buy a gaming router?
Only if your current router has no SQM or Smart Queue setting. What fixes bloat is fair queuing with a shaper (fq_codel or CAKE), not gaming branding. Many mid-range routers already include it.