How long to download 100 GB at 200 Mbps

Four thousand seconds, and the three numbers behind it

One hundred decimal gigabytes is 800,000 megabits, because each gigabyte holds 8,000. A 200 Mbps line moves 200 megabits a second, so 800,000 divided by 200 is 4,000 seconds, which is 1 hour 6 minutes 40 seconds. The meter reads 25 MB/s, and the wait is short enough to sit through but long enough that a dropped connection costs real time.

The prefix carries the whole calculation. A gigabyte is 10⁹ bytes, so 100 GB is 100,000,000,000 bytes and 800,000,000,000 bits, which is the decimal reading the NIST prefix table defines for giga. Two hundred megabits a second is 25 megabytes a second, since eight bits make a byte, and 25 MB/s clears 90 GB in an hour. Each gigabyte costs 40 seconds, and ten gigabytes cost 400.

A file manager that reports binary units shows the same file as 93.13 GiB, since 100,000,000,000 divided by 1,073,741,824 is 93.13, and its meter ticks at 23.84 MiB/s. Both readings describe one transfer. The gap between them is 7 percent in each direction, and it is the reason a download can look smaller and slower than the number on the bill.

Two hundred megabits is twice the 2024 federal floor

The FCC raised its fixed broadband benchmark on March 14, 2024, in the report numbered FCC 24-27. The old standard of 25 Mbps down and 3 Mbps up, set in 2015, became 100 Mbps down and 20 Mbps up, and the same report set a long term goal of 1 Gbps down and 500 Mbps up. The Commission published the change in its March 2024 announcement and the reasoning in the report itself.

Two hundred megabits is twice the download floor, which sounds generous until the upload half is compared. The floor pairs 100 down with 20 up, a five to one split, and most 200 Mbps cable plans keep a similar asymmetry. Sending the same 100 GB archive the other way on a 20 Mbps uplink takes 40,000 seconds, or 11 hours 6 minutes 40 seconds, which is a different job on a different day.

The FCC household guide rates 100 Mbps as enough for several devices at once, and it says in plain text that its numbers are rough guidelines rather than measurements. That caveat matters here: the guide is describing browsing, video and calls added together, not one process pulling a single file at full rate. A 200 Mbps plan is comfortable for a family and still takes over an hour for this file.

Where the tier sits in the market is the useful part. A 200 Mbps plan is the first step above the federal floor, it is often the cheapest tier a cable operator sells after a promotional year, and it is fast enough that the line stops being the obvious suspect when a download crawls. From here the interesting failures are downstream of the modem.

Where Wi-Fi 5 runs out on a 200 Mbps line

A 200 Mbps plan sits close to the point where an older wireless link stops keeping up. The 802.11ac rate table puts a 2x2 link on an 80 MHz channel at 866.7 Mbps at MCS9 with a short guard interval, a 2x2 link on 40 MHz at 400 Mbps, and a 1x1 link on 40 MHz at 200 Mbps. Those are PHY rates, measured at the antenna. A file sees roughly half to 60 percent of them once beacons, acknowledgements, retries and neighbours on the same channel take their slots.

Run the arithmetic on the weakest common case. A laptop with one antenna on a 40 MHz channel negotiates 200 Mbps of PHY rate, delivers about 110 Mbps of payload at 55 percent efficiency, and the file then needs 800,000 divided by 110, which is 7,273 seconds, or 2 hours 1 minute 13 seconds. The plan did not slow down. The air did.

The first row is the frame tax. A 1,500 byte IP packet carries 1,460 bytes of TCP payload, and it leaves the card as 1,538 bytes on the wire, so 94.9 percent of the line reaches the file. That is 189.8 Mbps from a 200 Mbps plan, and it adds 215 seconds to the job. Anyone who runs a speed test and sees 190 has not been shortchanged.

The jump from the third row to the fifth is the one that surprises people. Dropping from two spatial streams to one, and from 80 MHz to 40 MHz, cuts the ceiling by a factor of four, and the clock more than doubles. Buying a faster plan does nothing about it. Moving the laptop, or spending the difference on a mesh node with a wired backhaul, does.

Airtime, streams and the mesh tax

Every device on the access point competes for the same airtime, so the useful question is not how much bandwidth is left but how many transmission slots remain. A 4K stream asks for about 15 Mbps according to the published Netflix speed table, and a group video call about 6 Mbps each way. Subtract both from the rated line before dividing.

Three 4K streams take 45 Mbps, which leaves 155 Mbps, and the file then needs 800,000 divided by 155, which is 5,161 seconds, or 1 hour 26 minutes 1 second. One stream plus two calls takes 27 Mbps, leaves 173, and gives 4,624 seconds, or 1 hour 17 minutes 4 seconds. The arithmetic is linear until the line is full, and on a 200 Mbps plan a normal evening rarely fills it.

A mesh node changes the picture in a way the plan cannot. When the satellite unit talks to the base over the same radio it uses for clients, the backhaul and the client share one channel, so the usable rate at the far node is roughly half of what the base delivers. At 100 Mbps of payload the file needs 8,000 seconds, and a 200 Mbps plan has become a 100 Mbps plan without anyone touching a setting.

Wired backhaul removes that tax, and so does a tri-band node with a dedicated radio for the link between units. Neither is a bandwidth upgrade. Both are ways of not throwing away the bandwidth that is already paid for, which is why they show up before a plan change in any sensible order of fixes.

The disk is idle at 25 MB/s

Twenty-five megabytes a second is slow enough that storage never limits this transfer. A 5,400 rpm laptop drive writes sequentially at 100 MB/s or more, a SATA SSD at 500, and an NVMe drive in the thousands. The download lands on disk long before the progress bar fills, and the CPU has nothing to do beyond checksumming a stream that arrives at a fifth of a gigabit.

The steps after the last byte are where time goes. Copying 100 GB to a NAS over gigabit Ethernet at 110 MB/s costs 909 seconds, or 15 minutes 9 seconds. A SHA-256 read back at 500 MB/s costs 200 seconds. A second copy to an external drive at 150 MB/s costs 667 seconds. None of these overlap with the download unless the client is built to stream and copy at once, and most are not.

The 4 GB wall, and why the copy comes first

One hard limit sits outside all of this. FAT32 refuses any single file above 4 GB, so an old USB stick fails a 100 GB archive near the 4 GB mark even with 200 GB free. At 25 MB/s the refusal lands 160 seconds into the download, with four of every hundred gigabytes written and the remaining ninety-six still on the wire. exFAT, NTFS, APFS and ext4 all take the size without complaint, and the stick still writes at whatever speed its controller allows.

The practical order is copy first, verify second, delete last. A power cut during a copy costs the copy, not the archive, and a digest taken before the delete is the only evidence that the file that survived is the file that arrived. On a 66 minute transfer that order matters more than the line speed, because the last byte is not the last task.

Five minutes of checks before a 66 minute wait

A counter reading taken 300 seconds in tells you more than any speed test run before the transfer started. At the full 200 Mbps the counter should show 7.5 GB after five minutes, since 25 MB/s multiplied by 300 seconds is 7,500 MB. Two thirds of that reading means the path is delivering 133 Mbps and the job will take 6,015 seconds.

Run iperf3 against a nearby host with one stream and then with four. One stream at 190 Mbps means the path is clean and the window is large enough. One stream at 60 Mbps with four streams at 190 means the receive window is the ceiling, which is a server or stack problem rather than a line problem. Four streams that also stall near 60 means the line or the access point is saturated by something else.

Watch the round trip during the run, not before it. A line that idles at 15 ms and sits at 250 ms while saturated is bloated, and every call in the house suffers even though the throughput number looks healthy. Smart queue management on the router fixes that, and it costs nothing at this rate. Set the shaper to 95 percent of the measured rate so the queue forms on the router rather than at the modem.

Check the channel width last. A laptop that reports 40 MHz on a 5 GHz network near the access point is being steered by a driver or a router setting, and the fix is one configuration change rather than a new plan. The wireless section above puts a number on what that setting is worth: 4,211 seconds at 40 MHz in the same room against 4,444 through two walls at 80 MHz.