VoIP Bandwidth Requirements: How Much Do You Really Need?

2026-07-22 · 8 min read · by the fvoip team

Here is the most common misconception about VoIP call quality: that it is about how much bandwidth you have. In 2026, almost any business internet connection has enough raw bandwidth for far more calls than the business will ever make at once. A single call needs roughly 100 kbps; even a modest 100 Mbps connection could theoretically carry hundreds of them. Yet plenty of offices with fast internet still get choppy audio. That is because voice quality depends less on how big your pipe is and more on how consistently packets move through it. This guide covers both: the real per-call numbers, and the network settings that actually determine whether calls sound good.

Bandwidth per call: the real numbers

How much bandwidth a call uses depends on the codec, the algorithm that turns voice into data. The three you will actually encounter:

G.711: the classic uncompressed standard, same audio quality as a traditional phone line. The voice payload is 64 kbps, but packet overhead matters: voice is sent as small packets (typically every 20 ms), and each one carries IP, UDP, and RTP headers plus ethernet framing. Loaded onto the wire, a G.711 call consumes roughly 85 to 90 kbps in each direction. Budget ~90 kbps and you will not be surprised.
Opus: the modern standard, and what current systems (including fvoip apps and phones that support it) prefer. Opus is variable-bitrate and spans roughly 6 to 510 kbps across all uses; for voice it typically runs 16 to 40 kbps of payload, delivering wideband HD audio that sounds noticeably better than G.711 at a fraction of the bitrate. With overhead, plan on roughly 30 to 60 kbps per call.
G.729: a legacy low-bandwidth codec at 8 kbps payload, roughly 31 to 32 kbps on the wire with overhead. It was the answer when bandwidth was scarce; it costs some audio quality, and with Opus available there is little reason to choose it today outside of constrained links.

You may also see G.722, an older HD codec at 64 kbps payload (same on-wire cost as G.711, around 87 kbps, but with wideband audio); many desk phones support it as their HD option.

Sizing your connection: the arithmetic

The planning rule of thumb: reserve 100 kbps per concurrent call, in each direction. That covers G.711 worst-case with margin, and leaves headroom when Opus is negotiated. The two words that matter in that sentence are concurrent and each direction:

Concurrent, not total. A 20-person office does not need bandwidth for 20 calls; it needs bandwidth for as many as happen at once, which for general office use is commonly 25 to 50 percent of staff at peak. Ten simultaneous calls is 1 Mbps. That is the entire voice load of a busy 20-to-30-person office.
Each direction, and upload is the constraint. Voice is symmetrical: you send as much as you receive. Business cable and DSL plans are usually asymmetrical (say, 300 Mbps down but 20 Mbps up), and every other service in your office (cloud backups, video calls, file sync, security-camera uploads) competes for that thinner upload lane. When VoIP degrades, the upload side is almost always where it happened. Fiber, with symmetrical upload, sidesteps this entirely.

The metrics that actually determine quality

Once minimum bandwidth is met, three network characteristics decide how calls sound. Good targets:

Latency (one-way delay): under 150 ms. The long-standing ITU guidance for conversational quality. Below it, conversation feels natural; well above it, people start talking over each other. Most US domestic paths are comfortably under this unless something on your network is adding delay.
Jitter: under 30 ms. Jitter is the variation in packet arrival timing. Voice packets leave every 20 ms; if they arrive erratically, the receiving end's jitter buffer must either wait (adding delay) or give up on late packets (causing artifacts). Jitter is the classic symptom of a congested router.
Packet loss: under 1 percent. Lost voice packets are gone; there is no retransmission in real-time audio. Modern codecs conceal an occasional lost packet, but sustained loss above about 1 percent becomes audible as choppy or robotic sound.

The most common villain behind bad jitter and loss on an otherwise fast connection is bufferbloat: a router with oversized buffers lets a big upload (a cloud backup, a large email attachment) fill the queue, and voice packets sit in line behind bulk data for hundreds of milliseconds. The connection is fast; the queue is fat. This is exactly what QoS exists to fix.

Router QoS: the settings that matter

Quality of Service tells your router that voice packets jump the queue. On a business router, the checklist looks like this:

Turn on QoS and set your real bandwidth. QoS engines shape traffic to a configured line speed. Run a speed test, then set the router's up/down figures to roughly 85 to 95 percent of measured throughput so the router, not your ISP's equipment, is the choke point doing the queueing.
Prioritize voice traffic. The clean method is DSCP tagging: VoIP audio (RTP) is conventionally marked EF (Expedited Forwarding, DSCP 46), and call signaling is commonly marked CS3. fvoip phones and apps tag their traffic; configure the router to honor DSCP marks and place EF in its highest-priority queue. If your router cannot match on DSCP, prioritize by the device IPs of your desk phones or by your provider's traffic instead.
Use smart queue management if offered. Many modern routers include SQM / fq_codel / CAKE queueing, which fights bufferbloat directly and often fixes call quality even without manual per-service rules. If your router offers it, enable it.
Disable SIP ALG. Not a QoS setting, but the single most notorious router feature in VoIP. SIP ALG (Application Layer Gateway) rewrites VoIP packets in ways that frequently break modern services: one-way audio, dropped registrations, calls that ring but cannot connect. It ships enabled on many routers. Turn it off.
Put desk phones on ethernet, and on their own VLAN if you can. Wi-Fi adds jitter by nature (retransmissions, roaming, interference); wired phones remove a whole category of problems. A voice VLAN additionally keeps phone traffic separated and consistently prioritized. Supported Yealink, Poly, and Grandstream models all have built-in ethernet, most with a pass-through port so the phone and PC share one wall drop.

A worked example

A 15-person office with 300/20 Mbps cable internet expects at most 6 concurrent calls. Voice load: 6 x 100 kbps = 0.6 Mbps each way, which is 3 percent of the upload. Bandwidth is a non-issue. The configuration that makes it flawless: QoS enabled and shaped to ~18 Mbps up, EF-tagged voice in the priority queue, SQM on, SIP ALG off, phones wired. That office will have better call quality than one with gigabit fiber and an unconfigured router.

What fvoip handles for you

Every fvoip plan includes HD calling with modern codec support, and our phones arrive pre-provisioned with sensible defaults, including DSCP tagging. During onboarding, our US-based 24/7 support team runs a connection check with you and walks through the router settings above for your specific hardware; that help is part of every plan, not an add-on. If you are moving off a legacy system, see what is included on the features page, and bring your numbers with you: porting is free, with local numbers typically completing in 3 to 10 business days via fvoip.com/port.

The short version to remember: 100 kbps per concurrent call, upload matters more than download, jitter under 30 ms and loss under 1 percent, QoS on, SIP ALG off, phones wired. Get those right and VoIP does not just match your old landlines; it sounds better than they ever did.