DOCSIS vs Fiber vs 5G vs Starlink: Real-World Performance Profiles
Four last-mile technologies compete for your household in 2026. They are sold with the same number on the box — "1 gigabit" — and they deliver radically different connections. This guide measures each on throughput, upload, latency, jitter, and behavior under load, because the advertised downstream rate is the least useful number on your bill.
The advertised-speed trap
Every one of these technologies can plausibly claim "gigabit" service. That single number compresses four variables that behave completely differently in practice: peak downstream throughput, symmetrical upload capability, idle-network round-trip latency, and round-trip latency when the line is actually being used. The downstream peak is usually the only one anyone quotes, and it is usually the least predictive of how the connection feels.
The four rows below map each access technology to the four variables that actually determine user experience. Numbers are conservative medians drawn from Ookla's 2026 Speedtest Global Index, MLab's NDT7 dataset, and Cloudflare Radar's aggregated quarterly reports.
| Technology | Down (peak tier) | Up (typical) | Idle latency | Latency under load |
|---|---|---|---|---|
| Fiber FTTH | 1–10 Gbps symmetric | 1–10 Gbps | 2–6 ms | 3–10 ms |
| DOCSIS 3.1 | 1.2 Gbps | 35–50 Mbps | 10–18 ms | 40–250 ms |
| DOCSIS 4.0 | up to 10 Gbps | 1–6 Gbps | 6–12 ms | 15–60 ms |
| 5G FWA | 300–1000 Mbps | 20–100 Mbps | 18–35 ms | 40–180 ms |
| Starlink (Gen3) | 100–300 Mbps | 10–40 Mbps | 25–45 ms | 50–250 ms |
Fiber (FTTH): the reference implementation
Fiber-to-the-home sets the ceiling on every metric. A passive optical network runs a dedicated strand to a splitter and from there to each premises, which means the last-mile hop is electrically quiet, full-duplex, and effectively unlimited until you reach the OLT uplink. Residential tiers start at 1 Gbps symmetric in 2026 and 2 Gbps to 10 Gbps tiers are common in metros served by AT&T, Frontier, Ziply, Sonic, and most municipal networks.
The important specification for fiber is not the downstream rate. It is the upload rate. A cable connection with 1 Gbps down and 35 Mbps up will saturate the upstream with a single large photo backup and the entire household's latency will balloon until the upload finishes. A symmetric 1 Gbps fiber connection rarely produces that failure mode because upstream congestion is almost unreachable at home-LAN scale.
Idle latency on fiber typically measures 2 to 6 ms to the nearest major carrier. Under full bidirectional load, well-provisioned FTTH connections hold inside 10 ms of round-trip time because the access network has no scheduling constraint comparable to DOCSIS request-grant or 5G airlink contention.
DOCSIS cable: the pragmatic choice with one structural flaw
DOCSIS 3.1 is the cable standard that ships gigabit tiers to roughly 100 million North American households in 2026. It is pragmatic because the coaxial plant already exists and the upgrade cost is a single CMTS replacement per node. It is flawed because coax is a shared half-duplex medium and the upstream direction uses a request-grant scheduler: a cable modem asks permission to transmit on a TDMA grid, the CMTS grants a slot, the modem transmits, and every one of those exchanges adds round-trip delay.
The practical result is that a DOCSIS 3.1 connection under upload load produces latency spikes the user can feel. Video calls glitch when someone hits upload on a photo backup. Online games desync when a cloud sync fires. This is the single most cited complaint against cable and it is not a cable-modem firmware bug — it is how the physical layer works.
DOCSIS 4.0 is the fix. Full-duplex DOCSIS (FDX) splits the coax frequency plan into dynamically allocated upstream and downstream bands and eliminates the strict scheduling constraint that creates the latency stall. Comcast began DOCSIS 4.0 rollouts in 2024 and mid-2026 coverage is still patchy. When it lands in a neighborhood, DOCSIS 4.0 measurably closes the latency gap with fiber and enables multi-gigabit symmetric tiers. Before it lands, the structural upload constraint defines the experience.
Test your connection with the Y2KDash bufferbloat test to see the DOCSIS scheduling artifact in action: an idle ping of 12 ms that climbs to 180 ms the moment an upload starts is the signature.
5G Fixed Wireless Access: variance is the whole story
T-Mobile Home Internet, Verizon 5G Home, and AT&T Internet Air deliver residential service over the same 5G radio stack that carries phone traffic in the same cell. The household modem is simply a customer-premises 5G receiver with a built-in router. The tower schedules downlink and uplink airtime across every subscriber on the cell, which means throughput, latency, and jitter are all a function of how busy the cell is at the moment you measure.
During off-peak hours (early morning, midday on weekdays), 5G FWA can match DOCSIS 3.1 downstream throughput and deliver 25 to 35 ms of stable latency. During peak hours (weekday evenings), the same connection routinely drops to 50 Mbps with jitter in the 100+ ms range because the tower is saturated. A midweek daytime Ookla test showing 400 Mbps is genuinely how the connection performs at that moment. It is not how it will perform at 8 PM on a Thursday when everyone in the neighborhood is streaming.
5G FWA is the correct choice for addresses where the only wired option is DSL, where cable has a monopoly pricing problem, or as a redundancy link behind fiber. It is the wrong choice for households that need deterministic performance — remote workers on consistent video calls, competitive gamers, or small businesses.
Starlink: the niche that went mainstream
Starlink is a low-earth-orbit satellite constellation with ground terminals that steer a phased array at whichever satellite has the best geometry at any given moment. Because the constellation is low (around 550 km) rather than geostationary (36,000 km), round-trip latency is a viable 25 to 45 ms rather than the 600+ ms of legacy satellite. This is the technical breakthrough that made Starlink compete with terrestrial broadband.
Throughput sits in a narrower band than either cable or fiber. The Gen3 dish delivers a reliable 100 to 300 Mbps down and 10 to 40 Mbps up, and that range has held steady since early 2024 because SpaceX adds constellation capacity roughly in pace with subscriber growth. Bigger downstream numbers appear in off-peak speed tests but the 95th-percentile downstream still lands under 400 Mbps in most cells.
Starlink's failure modes are weather and handoff. Heavy rain, dense snow, and occasional geometric handoffs between satellites produce brief outages — typically a fraction of a second, occasionally several seconds during storms. For interactive applications (gaming, video calls), these show up as momentary drops. For bulk transfers they are invisible because TCP retransmits seamlessly.
The correct framing for Starlink is that it is the best available rural broadband in 2026 and a serviceable urban backup, not a primary connection for anyone with fiber or DOCSIS 4.0 available.
Which should you actually choose?
Rank by the metric your household actually consumes:
- Symmetric upload matters (remote work, content creation, cloud backup, home servers): fiber, then DOCSIS 4.0, then everything else far behind.
- Consistent low latency matters (competitive gaming, live streaming, VoIP): fiber, then DOCSIS 4.0, then cable with proper SQM, then 5G FWA, then Starlink.
- Peak throughput matters (large file transfers, multi-4K-stream households): fiber top tier, DOCSIS 4.0 top tier, DOCSIS 3.1 top tier. 5G and Starlink both cap well below.
- Rural address, no wired option: Starlink first, 5G FWA second, DSL dead last.
For the single most-asked question — "should I pay more for fiber if it's available?" — the answer is yes if any of the following apply: you work from home, anyone in the household plays competitive games, you host files or run any kind of server, or you currently suffer the bufferbloat cable failure mode daily. If none of those apply and your DOCSIS 3.1 connection is stable, the practical upgrade is smaller than the marketing implies.
Frequently asked questions
- Which is fastest overall: DOCSIS, fiber, 5G, or Starlink?
- Fiber wins on every metric that matters: peak throughput, symmetrical upload, latency under load, and jitter. DOCSIS 4.0 closes the throughput gap but inherits coax's half-duplex scheduling. 5G and Starlink sit in a different tier because they share a physical medium with other subscribers.
- Is Starlink good for gaming?
- Starlink is playable for most online games but not competitive. Median latency sits around 25 to 45 ms, jitter is noticeably higher than wired options, and packet loss during rain or snow events will briefly disconnect you from multiplayer sessions.
- Is 5G home internet better than cable?
- It depends entirely on your cell tower load. 5G FWA can match DOCSIS 3.1 during off-peak hours and collapse to DSL-like speeds during evening congestion. Cable throughput is more predictable because DOCSIS channels are not shared with phone traffic.
- Does fiber really have lower latency than cable?
- Yes. Fiber last-mile adds roughly 1 to 3 ms of access-network delay. DOCSIS adds 8 to 15 ms because of upstream request-grant scheduling on shared coax. The difference is measurable on every ping test and is especially visible under upload load.
- Should I switch from cable to fiber if it's available?
- If symmetric upload, video call quality, or cloud backup speed matter to you, yes. If you only browse and stream and your DOCSIS provider gives you stable 300+ Mbps down, the practical difference is smaller than the marketing implies.