Your internet is fast. Your Zoom calls still freeze. Here is the number nobody checked.

Video calls are the one thing most people do online that their internet plan was never designed for. A plan is sold on download speed. A call lives or dies on upload, on how your router behaves when the line is busy, and on how steady the packets arrive. None of those appear on the sticker.

This guide walks through the five numbers that actually decide a call, how to measure each one on this site, how to compare your readings with the diagnostics built into Zoom, Teams, and Google Meet, and what to change in order of cost. Most fixes are free. One is a cable. The expensive options are covered honestly at the end, including which ones do nothing.

Why a gigabit plan does not help

A 1080p Zoom call in a group meeting needs about 3.8 megabits per second up and 3.0 down; 720p needs 2.6 up and 1.8 down (Zoom system requirements). Microsoft says Teams can deliver HD video in under 1.5 megabits and recommends 2.5 up and 4.0 down for meetings (Microsoft Learn). Google Meet tops out around 3.6 megabits each way for 1080p (Google Workspace Help). Audio alone is 60 to 80 kilobits, less than a tenth of a megabit (Zoom Technical Library).

Those are tiny numbers. Almost every broadband plan clears them with room to spare. So when a call freezes on a connection that just tested at 500 megabits, the speed was never the problem. Something else is.

There is one catch in those requirements that explains a lot of failures: the upload figure is the bigger one. Calls are the rare everyday application that sends as much as it receives. Cable and DSL plans typically allocate five to ten times more capacity to download than upload (FCC Measuring Broadband America). In the FCC's most recent national measurements, the average advertised upload speed on cable was 16.6 megabits, against 528 megabits on fiber. A 1,200-megabit cable plan with a 35-megabit upload has less headroom for a household on three simultaneous calls than a 300-megabit fiber plan with 300 up. The sticker number was never the one that mattered.

The five numbers that decide a call

Upload speed. The direction your face and voice travel. You need a few megabits per call with headroom for whatever else is leaving the house at the same time: cloud photo backup, security cameras, a kid's game stream, another call. On a 10-megabit upload, two HD calls and an iCloud sync are already a fight.

Ping under load (bufferbloat). How much your latency rises when the line is busy. On many home routers, a single large upload or download fills a queue and everything else waits behind it. This is why a call turns into a slideshow the moment someone starts a backup. The site's Latency (Loaded) card grades this A to F. Because calls push upload, you also want to see what happens when the upload side specifically fills, which the audit below covers.

Jitter. The wobble in arrival time from one packet to the next. Voice and video decoders buffer a little to absorb it; when the wobble exceeds the buffer, you get robotic audio and frozen frames. Microsoft's threshold for a healthy Teams call is under 30 milliseconds (Microsoft Learn); Zoom recommends 40 or less (Zoom Support). The site's Jitter card uses the same scale.

Packet loss. Packets that never arrive. Calls cannot wait for a resend, so a lost packet is a missing syllable. Microsoft's Teams target from your device to its network is under 1 percent (Microsoft Learn); Zoom's guidance is 2 percent or less (Zoom Support).

Round-trip latency. The least important of the five for most people, but it has a ceiling. Google says Meet video quality is highest when the round trip to Google is under 100 milliseconds and degrades at 300 or more (Google Workspace Admin Help). Teams wants under 100 from your device to Microsoft's network edge. The telecom standard that underlies all of this, ITU-T G.114, says one-way delay under 150 milliseconds gives "essentially transparent interactivity" and above 400 is unacceptable (ITU-T G.114). Most wired home connections in the United States sit far below the ceiling: median idle latency in the FCC's measurements was 7 to 14 milliseconds on fiber, 12 to 24 on cable, and 23 to 34 on DSL. If your unloaded ping is fine and the call still stutters, the problem is one of the other four.

There is also a sixth factor that is not a network number at all: your own computer. More on that in Step 5.

Why this matters more than it seems

Two findings from the research literature explain why a mediocre connection costs more than a little annoyance.

The first is about delay. In a 2014 experiment, researchers at TU Berlin added artificial transmission delay to conversations and asked participants to rate each other. With 1.2 seconds of delay, people rated their partner as less attentive, less friendly, and less self-disciplined, even though the partner had done nothing differently, and most participants did not consciously notice the delay at all (Schoenenberg, Raake, and Koeppe). People do not blame the network. They blame you.

The second is about audio. In 2018, researchers at USC and the Australian National University played identical conference talks and NPR interviews in high and low audio quality. Listeners who heard the poor audio rated the speaker as less intelligent and less likeable and the work as less important, despite identical content (Newman and Schwarz). Later work found the same effect for witnesses in simulated court proceedings (Bild et al.).

Put together: a connection that drops syllables and adds half a second of lag is quietly changing how colleagues and clients judge you, and they will not know it is the connection. That is worth an hour of testing.

Run the video call audit

Each step takes a few minutes. Do them in order; each rules something in or out.

Step 1: the quiet baseline, with upload in the frame

Pick a time when nobody else is using the connection. Plug the computer you take calls on into the router by cable if you can, even if you normally use Wi-Fi, so you know what the line itself delivers before adding the wireless variable.

Open Settings on the Speed and Network page and enter both your plan's download and upload speeds. Most people leave the upload box blank. Fill it in; it is the one that matters here. Choose the "Internet speed and my ISP" focus. Run the test and write down:

• Upload speed (not download)
• Latency (Unloaded): the Best figure
• Latency (Loaded): the letter grade
• Jitter in milliseconds
• Packet loss percentage

What good looks like for calls: upload comfortably above 10 megabits (the test description notes that below 3 megabits video meetings start freezing), a Loaded Latency grade of A or B, jitter under 20 milliseconds, and zero packet loss.

Step 2: the busy-house test, upload edition

Now make the upload side busy on purpose. Start uploading a large video to Google Drive, Dropbox, or iCloud from another device, or kick off a cloud backup. While that runs, retest on your call computer.

Watch what happens to Jitter and Latency (Unloaded). On a connection with upload bufferbloat, unloaded latency that was 15 milliseconds will jump into the hundreds and jitter will spike, while the download number barely moves. This is the exact mechanism behind "my call freezes whenever someone in the house uploads photos," and it is specific to the upload direction because that is where the queue fills on an asymmetric plan.

If latency and jitter stayed flat, your router is handling bulk traffic well. If they spiked, skip ahead to Fix 2; it is the single most common cause of freezing calls in homes with fast internet.

Step 3: the Wi-Fi tax

If you take calls on Wi-Fi, repeat Step 1 twice at your desk: once wired, once wireless. Use the Watch monitor's Spot label so the two readings sit side by side ("Office wired" versus "Office Wi-Fi"). Compare jitter and packet loss, not speed.

RTINGS tested eleven routers, from budget to premium, and found that Ethernet produced virtually no added latency or jitter while every Wi-Fi configuration added periodic spikes regardless of router model (RTINGS). Microsoft's own quality targets for Teams concede the same point from the other side: the acceptable rate of poor audio streams on 2.4 GHz Wi-Fi is set at double the rate for wired or 5 GHz connections (Microsoft Learn). If your wireless jitter is within a few milliseconds of wired, Wi-Fi is fine for calls. If it is in double digits, the fix is a cable or a 5 GHz band, not a new router.

Step 4: the 9 a.m. problem

Calls fail at the times everyone else is also on calls. Open the Watch monitor on the Speed and Network page, set the interval to 1 minute, label the Spot "Desk, work hours," and leave the tab open through a normal morning of meetings. (The tab must stay open; the monitor runs in your browser and uploads nothing.)

You are looking for a pattern: steady readings that degrade at specific times point to neighborhood congestion or an ISP capacity problem. Random spikes with no time pattern point to something in your house, usually Wi-Fi interference or a device starting a backup. If the pattern is time-of-day and the wired test in Step 1 was clean, export the Evidence Pack and use the ISP call script. A timestamped PDF showing jitter and packet loss climbing every weekday at 9:15 is a very different conversation than "my Zoom is bad."

Step 5: your computer, and your browser

Not every frozen call is the network. Video encoding, background blur, and gallery view all cost processor time, and a laptop that is already warm will drop frames before the network drops a packet.

Open the Device report and note the CPU core count and memory your browser reports. Zoom's recommended configuration is a quad-core processor and 16 GB of memory, and it warns that dual-core laptops share screens at around 5 frames per second (Zoom system requirements). Google's "optimal" tier for Meet is an 11th-generation Intel i5 or newer, a Ryzen 5000, or Apple silicon (Google Meet Help). If the Device report shows two or four cores and 8 GB, that is a reason your calls stutter when you turn on blur, and no router will fix it.

While you are there, open the Privacy report and look at the WebRTC card. Meet and the browser versions of Zoom and Teams run on WebRTC. If you have disabled it for privacy reasons, that card will say so, and it explains why browser-based calls fail or fall back to audio only. The fix is to allow WebRTC in the browser you use for calls and keep it disabled elsewhere, or use the desktop apps.

Check your reading against the app's own numbers

All three major platforms show the same statistics this site measures, live, during a call. Comparing them tells you whether the problem is your connection or theirs.

• Zoom: during a meeting, click the arrow next to Start Video, then Video Settings, then Statistics. The Audio and Video tabs show latency, jitter, and packet loss. Zoom's guidance is latency at or under 150 milliseconds, jitter at or under 40, and loss at or under 2 percent (Zoom Support).
• Teams: click More actions, then Settings, then Call health. Round-trip time should be under 200 milliseconds, received packet loss under 2 percent, and audio jitter under 30 (Microsoft Q&A).
• Google Meet: click More options, then Troubleshooting and help, and read the Network stability graph and CPU indicator (Google Meet Help).

If this site shows clean numbers and the app shows bad ones, the trouble is between your ISP and that platform's servers, or it is your computer (check the CPU figure in Zoom's Statistics or Meet's panel). If both show bad numbers, the trouble is in your house or your line, and the fixes below apply.

Fixes, in order of cost

Fix 1 (free): stop the upload competition

Before every important call, pause what is leaving the house: cloud photo sync on phones, backup clients, security camera uploads if they can be scheduled, torrents, game streams. Google's own troubleshooting advice is to close streaming services, cloud backup clients, and file syncs for the duration of a meeting (Google Meet Help). On an asymmetric plan this is the single highest-value habit, because a phone backing up 2,000 photos will take every megabit of a 10-megabit upload.

Fix 2 (free if your router has it): Smart Queue Management

If Step 2 showed latency spiking under upload load, your router needs to manage its queue. Look in its settings for SQM, Smart Queue, CAKE, FQ-CoDel, or Adaptive QoS, and set the limits to about 90 percent of your measured download and upload speeds. The site's bufferbloat tutorial walks through the setting by brand. One warning repeated there: router modes labeled "Gaming," "Media Priority," or similar can make bufferbloat worse, because they prioritize by application rather than keeping the queue short. Retest Step 2 after enabling; the jitter spike should be gone.

Fix 3 (free): audio first, video second

Turn on your camera only when you need to, and turn it off when the connection degrades. Google's guidance is blunt about it: turning off video "significantly reduces the amount of bandwidth your connection uses to stabilize your audio" (Google Meet Help). Audio is 60 to 80 kilobits; it will survive almost anything if video is not competing with it. Given the credibility research above, clean audio with the camera off beats pixelated video with robotic sound every time.

Fix 4 (free): lower the load on your computer

In Zoom, turn off HD video in Settings, Video and effects, if your CPU is marginal; Zoom notes HD uses more processor and about 2 megabits more each way (Zoom Support). Blur and virtual backgrounds without a green screen have their own processor requirements and are hidden entirely on machines that do not meet them (Zoom Support). In Meet, set send and receive resolution to 360p, switch to the Spotlight layout, and, surprisingly, disconnect a 4K or QHD external monitor if you can; Google lists all three as ways to cut CPU and GPU load (Google Meet Help). Close the other 40 tabs.

Fix 5 (free): take the call off the VPN

If you run a VPN for privacy or for work, your call is taking a detour. Zoom's own guidance to IT departments is to split-tunnel Zoom traffic around the VPN, because sending real-time media through one "can cause significant audio and video degradation" (Zoom Technical Library). Google's troubleshooting steps include disconnecting the VPN and retrying. If your employer requires the VPN, ask IT whether Zoom and Teams are already excluded; most corporate configurations do this. If you use a personal VPN, turn it off for the call or use its split-tunneling feature for the meeting app.

Fix 6 (one cable): wire the call computer

If Step 3 showed a Wi-Fi tax, a flat Ethernet cable to the desk fixes it permanently. Laptops without an Ethernet port take a USB-C adapter. If a cable is impossible, move to the 5 GHz band, get line of sight to the router, and keep the 2.4 GHz band for devices that do not need low jitter.

Fix 7 (modest cost): a wired headset

A USB headset with a boom microphone does three things at once. It eliminates echo, which Zoom says is usually caused by speakers feeding back into a microphone (Zoom Support). It avoids the Bluetooth microphone penalty: Zoom warns that other participants "may experience lower-quality audio from your microphone" on Bluetooth headphones (Zoom Support). And it puts the microphone two inches from your mouth instead of two feet, which is most of what separates "studio" from "speakerphone" in the credibility research. This is the best value hardware purchase in this guide, ahead of any webcam.

The expensive options, honestly

A faster plan. More download does nothing for calls. More upload can help, but only if your audit showed upload saturating. Before upgrading, ask your ISP specifically for a tier with higher upload, which on cable is often a different product than the headline speed. Some cable providers now offer symmetric tiers on upgraded DOCSIS 4.0 networks, with upload speeds that match download (CNET). If fiber is available at your address, it solves the upload problem outright: fiber tiers are generally symmetric.

Low-latency cable (L4S). Comcast has deployed low-latency DOCSIS to more than ten million homes, which brings loaded latency to about 18 milliseconds down and 20 up, close to idle, for applications that mark their traffic (IETF draft, Livingood). The catch for this guide: the application has to participate. Apple FaceTime does (Apple Developer); Zoom, Teams, and Meet were not among the launch partners (Comcast). For FaceTime users on Xfinity with a recent gateway, it is a free upgrade worth asking about. For everyone else, SQM on your own router does the same job today.

A new router. Only if your current one cannot do SQM. RTINGS found no latency advantage for premium routers over budget ones on an uncongested network; the feature that mattered was queue management under load (RTINGS). Shop for the SQM checkbox, not the antenna count.

A better webcam. Fine, after the headset. Google notes that bad video quality can be due to poor lighting as much as the network (Google Workspace Admin Help); a window in front of you or a cheap desk lamp improves any camera more than a camera upgrade improves a dark room. A 1080p camera is in Google's "optimal" tier and is plenty.

A gaming VPN or "network optimizer." No. A VPN adds a hop; see Fix 5.

Quick reference

Sources

• Zoom system requirements and bandwidth. support.zoom.com
• Zoom, accessing meeting statistics (latency, jitter, loss thresholds). support.zoom.com
• Microsoft, Teams network requirements (client-to-edge and edge-to-edge). learn.microsoft.com
• Microsoft, real-time telemetry thresholds. learn.microsoft.com
• Google, prepare your network for Meet. support.google.com
• Google, troubleshoot connection issues with Meet. support.google.com
• ITU-T Recommendation G.114, one-way transmission time. itu.int
• FCC, Measuring Fixed Broadband, Thirteenth Report. fcc.gov
• Schoenenberg, Raake, and Koeppe, "Why are you so slow?" 2014. sciencedirect.com
• Newman and Schwarz, "Good Sound, Good Research." 2018. sagepub.com
• Bild et al., "Sound and Credibility in the Virtual Court." usc.edu
• RTINGS, router latency research. rtings.com
• Livingood, ISP Dual Queue Networking (IETF draft 13). ietf.org
• Comcast, ultra-low lag press release. comcast.com
• Apple Developer, L4S support. developer.apple.com
• CNET, fiber vs cable vs 5G. cnet.com

The guide above is enough to find and fix almost any call problem. These notes are for readers who want the mechanics and the edge cases, including small-office administrators.

Advanced notes

• Why upload bufferbloat is worse than download bufferbloat on cable. The upstream channel on DOCSIS is both narrower and shared across a node, and legacy equipment queues were sized for throughput, not delay. The FCC's working-latency measurement runs probes at 10 packets per second during both the downstream and upstream saturation phases and reports the two separately; the charts show loaded latency significantly higher than idle, with the largest gap on DSL (FCC). Comcast's data show downstream 99th-percentile loaded latency falling from about 65 to 33 milliseconds from network-side active queue management alone, before any L4S queue (IETF draft, Livingood). If your ISP has deployed AQM, your Loaded Latency grade may already be decent on the download side while the upload side is untouched; that is why Step 2 loads upload specifically.
• Microsoft's stricter network-edge targets. From a corporate network edge to Microsoft's edge, Teams wants round trip under 60 milliseconds, jitter under 15, and packet loss under 0.1 percent over any 15-second window (Microsoft Learn). The looser client-to-edge figures quoted in the body (under 100, under 30, under 1 percent) are the right ones for a home user. If you administer a small office, the stricter set is your target for the WAN link.
• Jitter measurement differences. This site reports mean absolute deviation over 12 samples to public anycast endpoints. Zoom, Teams, and Meet report inter-arrival jitter on the actual media stream to their own relays. The numbers will not match exactly; look for agreement in kind (both fine, or both bad), not in digits.
• UDP matters. All three platforms carry media over UDP and fall back to TCP only when UDP is blocked, with worse results (Zoom Support). A firewall or "security" router feature that forces media onto TCP 443 will produce exactly the stutter this guide describes while every speed test passes. If calls are bad only on one network and that network has aggressive filtering, check this before anything else.
• Enhanced media tiers. Zoom's 1080p60 and high-bitrate modes push 8 to 12 megabits per participant and up to 100 megabits download in high-bandwidth mode (Zoom Technical Library). If your organization has enabled these, the bandwidth arithmetic in this guide changes, and upload headroom becomes a real constraint even on good connections.

The full source list is at the end of the Simple version of this guide.