How long to download 5 GB at 500 Mbps

Five gigabytes at 500 Mbps is 80 seconds, or 1 minute 20 seconds

One division, then a readout in bytes

Five gigabytes hold 5,000,000,000 bytes, or 40,000,000,000 bits. Divide by 500,000,000 bits per second and the answer is 80 seconds, or 1 minute 20 seconds. The byte readout is 500 / 8 = 62.5 MB/s.

Per minute the line moves 62.5 x 60 = 3,750 MB, or 3.75 GB. Per hour it moves 225 GB. Those rates explain why a 5 GB file feels instant on this tier and why a 1 TB drive image still takes most of an afternoon.

The same division covers any file: 1 GB is 8,000 megabits, so 8,000 / 500 = 16 seconds. Only the payload changes.

At 62.5 MB/s the slowest part is often the drive

Storage paths in the same unit

A 500 Mbps line writes 62.5 MB every second, and a modern laptop can absorb that without pausing. The table converts common write paths to megabits per second so they can be compared with the plan.

Only the USB 2.0 stick is slower than the line: 5,000 MB / 30 MB/s = 166.7 seconds, or 2 minutes 47 seconds. A spinning laptop drive at 100 MB/s finishes its share in 50 seconds and waits for the network.

A network drive changes the picture. A two bay NAS on a 2.5GbE port tops out near 280 MB/s, which is 2,240 Mbps, so it keeps pace; the same enclosure on a gigabit port is limited to 1,000 Mbps and stretches the file to 40 seconds. Check which port the cable landed in before blaming the ISP.

A RAID array is not automatically faster. Parity RAID pays a rebuild and parity write cost, and a two drive mirror writes the same bytes twice. SMR drives, common in budget enclosures, slow down once their cache fills, which shows up as a copy that starts strong and sags after a minute. Watch the drive activity light and the queue depth, not just the network graph.

Game installs are the exception. The client decrypts and decompresses while it downloads, so the CPU and the drive work at the same time, and the network rate can dip while the install catches up. That pause is local work, not a slow line.

Overhead, the binary size and the handshake

80 seconds becomes 85, then 91

Six percent of protocol and framing overhead turns 80 seconds into 80 x 1.06 = 84.8 seconds, which rounds to 1 minute 25 seconds. A 5 GiB file, the size Windows shows for a 5 GB download, carries 7.37 percent more data, so the same line needs 85.9 seconds, which rounds to 1 minute 26 seconds.

Stack the two and 5 GiB with 6 percent overhead needs 91.05 seconds, or 1 minute 31 seconds. The gap between 80 and 91 seconds is 14 percent, which is the realistic band for a single copy.

Connection setup is not the problem. A TCP and TLS handshake costs one or two round trips, and at 20 to 40 ms per trip that is 0.04 to 0.08 seconds. DNS adds one lookup before the first byte. Neither shows up in a timer measured in minutes.

What 500 Mbps moves in an hour, a day and a month

Volume, not just speed

Multiply the rate out and the tier becomes a volume product. At 62.5 MB/s the line moves 225 GB in an hour, 5,400 GB in a day, and 162 TB in a 30 day month if it never idles. The 5 GB file fits 45 times into one hour and 1,080 times into one day.

A 1.2 TB quota is exhausted in 1,200 / 225 = 5.33 hours of full rate transfer, or 5 hours 20 minutes. That is the number that matters for a capped family: the plan is fast enough to spend a month of data in a single evening.

Uploads matter too, and they depend on the plan shape rather than the download number. A symmetric 500/500 line sends the same file in 80 seconds. A 500/20 cable plan needs 40,000 / 20 = 2,000 seconds, or 33 minutes 20 seconds, which is 25 times the download wait.

The upload half of the plan decides the schedule

Symmetric fiber against cable upstream

Download throughput gets the advertising, but the upload number decides how long a backup, a video export or a cloud sync takes. A symmetric 500/500 fiber line sends the same 5 GB file in 80 seconds, matching the download. The cable version of the same tier, at 500/20, needs 33 minutes 20 seconds for that file, 25 times the download wait.

The gap is physical rather than a setting. Cable upstream traffic shares a narrow slice of spectrum with the rest of the neighborhood, while fiber carries the same capacity in both directions. That is why the same provider can sell 500 Mbps down on coax and still cap upload near 20 or 35 Mbps.

Scale the job up and the asymmetry dominates. A 100 GB cloud backup is 800,000 megabits. At 20 Mbps upstream it needs 40,000 seconds, or 11 hours 6 minutes 40 seconds; at 500 Mbps symmetric it needs 1,600 seconds, or 26 minutes 40 seconds. The FCC's Thirteenth Measuring Broadband America report, published on 9 August 2024 from September and October 2022 data, found eight of twelve provider and technology groups at or above advertised peak download speed and four between 86 and 90 percent, which is a reminder that the label is a ceiling rather than a typical evening.

A single connection rarely fills the pipe

Why a speed test and a download disagree

A throughput probe opens several connections and reports the sum. A single download often cannot, because the server, the CDN edge or the route caps one socket.

At 150 Mbps on one connection the file takes 40,000 / 150 = 266.7 seconds, or 4 minutes 27 seconds, on a line that can do it in 80. At 250 Mbps it takes 160 seconds, or 2 minutes 40 seconds. Parallel connections, which browsers and Steam use by default, recover most of the gap, but only when the server allows them.

Test with more than one tool. A browser download, a command line client and a multithreaded utility can give three different rates on the same wire, because each opens a different number of sockets and hits a different cache. Compare a single stream with eight parallel streams: if the sum climbs, the per connection cap is the limit; if both sit at the same number, the line or the server is saturated. Encryption and checksum offload can also cost throughput on an older CPU.

Distance and packet loss set a second ceiling. A TCP sender can only keep a certain amount of data in flight, set by the receive window and the round trip time, so a link with 40 ms of latency and a few ms of jitter needs a larger window than one with 5 ms. Loss makes it worse: congestion control backs off, retransmits, then climbs again, which is why a clean probe can beat a download from a distant server.

The client display confuses the issue further. Steam's own support page notes that its download rate is shown in bytes per second while connections are sold in bits per second. A Steam window reading 62.5 MB/s is a saturated 500 Mbps line, and the number 500 will never appear in it.

Put the three ceilings side by side and the arithmetic of a per connection cap is plain:

The last row is the common case on a busy CDN, and no plan upgrade fixes it. Test eight parallel streams before buying more bandwidth.

A patch, a game and a drive image in one table

Every size is the same division

Convert the file to megabits and divide by 500. The table doubles as a check, because 100 GB is 20 times 5 GB and 1,600 seconds is 20 times 80.

The 60 GB game is the row most people care about. Sixteen minutes is short enough that a download can happen between dinner and the first match, and long enough that starting it at 11 p.m. still means waiting.

When 500 Mbps is wasted: the utilization test

Headroom bought for bursts

Take a home that moves 1.2 TB in a month. At full rate that traffic takes 1,200 GB x 8 = 9,600,000 megabits, and 9,600,000 / 500 = 19,200 seconds, or 5 hours 20 minutes. Spread across a 30 day month of 720 hours, the line is busy 5.33 / 720 = 0.74 percent of the time.

Even 4 TB a month reaches only 64,000 seconds, or 17 hours 46 minutes 40 seconds, which is 2.47 percent of the month. The purchase is not average throughput; it is the size of the burst. At 500 Mbps a 60 GB game blocks the evening for 16 minutes and a 100 GB backup for 26 minutes 40 seconds. At 100 Mbps the same two jobs take 1 hour 20 minutes and 2 hours 13 minutes 20 seconds.

Provisioned rate, burst size and headroom are three different numbers, and only the first appears on the bill. That is the honest case for the tier. A family that never moves more than 300 GB a month pays for speed it will not use, while one that installs titles weekly and backs up a laptop nightly spends the same money on time it gets back every week.

Sales pressure pushes the highest tier, and the upsell is rarely dishonest, just poorly matched to a workload. A quote for 1 Gbps often costs little more than 500 Mbps, and the difference only shows up on large transfers. Work out the minutes saved per week, multiply by your hourly value of waiting, and compare that with the monthly difference. For a family that downloads one game a month the saving is minutes; for one that ships video files daily it is hours.