Gbps to GB/s Converter for PCIe and NVMe Bus Speeds

Reading a Bus Rate in Gbps as Drive Throughput in GB/s

Buses are specified in gigabits per second and drives are sold in gigabytes per second, so anyone comparing an NVMe drive against the slot it will sit in is holding two numbers that cannot be compared yet. A motherboard block diagram says the M.2 slot is PCIe 4.0 x4; the drive box says 7,000 MB/s. Whether those agree comes down to a division by eight, plus a small correction for how the link encodes its bits.

Factor: GB/s = Gbps ÷ 8, because a byte is eight bits. PCIe 4.0 signals at 16 GT/s per lane, so an x4 link is 4 × 16 = 64 Gbps raw and 64 ÷ 8 = 8 GB/s. After the 128b/130b line code, about 7.88 GB/s remains for payload.

128b/130b Costs About 1.5%

PCIe generations 3 and up wrap every 128 payload bits in a 130-bit block. That 1.54% tax is why a bus dividing to a clean 8 GB/s delivers a shade under it, before protocol headers are even counted.

SATA Pays a Much Older Tax

SATA III runs at 6 Gbps, dividing to 0.75 GB/s — but 8b/10b encoding spends a fifth of the link on framing, leaving 600 MB/s. Real drives settle near 550 MB/s, and no firmware update moves that ceiling.

The Bus Is Rarely the Limit

A converted bus figure is a ceiling, not a promise. Controller queue depth, DRAM cache, SLC write buffers and thermal throttling decide how much of it a drive holds — and none appear on the interface spec.

Checking an Interface Rate Before You Buy

Enter the Link Rate in Gbps

Take the signalling rate from the specification, multiplying lanes by the per-lane rate if the slot is quoted that way, and type it into the left field. 2.5 and 2,5 behave identically.

Hold It Against the Drive's Claim

Compare the GB/s result with the sequential read figure printed on the drive. If the drive's number is higher than the slot's ceiling, the slot decides the outcome whichever drive you buy.

Reverse It for a Review Figure

Benchmark articles quote GB/s. Press swap and enter that figure to see what sustained Gbps the link was carrying, which shows how much of the bus a drive really uses.

Copy It Into Your Build Notes

Each field has its own copy button giving the bare figure with no unit — convenient while assembling a comparison of slots, enclosures and candidate drives.

Both fields accept input, so the reverse direction needs no separate mode: type a GB/s value on the right and the Gbps equivalent appears as you go. Each side also carries a searchable unit list, useful when a spec sheet mixes rate units within one table.

Storage Interfaces From Gbps to GB/s

These are the interfaces a desktop drive is likely to sit behind. The third column is the plain division by eight; the fourth applies the link's own encoding, which is the closest a specification gets to what the drive can actually push.

Interface	Signalling rate	÷ 8 (raw GB/s)	After line encoding
SATA III	6 Gbps	0.75 GB/s	0.6 GB/s (8b/10b)
USB 3.2 Gen 2	10 Gbps	1.25 GB/s	≈1.21 GB/s (128b/132b)
USB 3.2 Gen 2x2	20 Gbps	2.5 GB/s	≈2.42 GB/s (128b/132b)
PCIe 3.0 x4	32 Gbps	4 GB/s	≈3.94 GB/s (128b/130b)
Thunderbolt 4 / USB4	40 Gbps	5 GB/s	≈4.85 GB/s (64b/66b)
PCIe 4.0 x4	64 Gbps	8 GB/s	≈7.88 GB/s (128b/130b)
PCIe 5.0 x4	128 Gbps	16 GB/s	≈15.75 GB/s (128b/130b)

Two rows need a caveat. Thunderbolt and USB4 carry display and USB traffic on the same 40 Gbps link, so a tunnelled NVMe drive sees only part of it and lands nearer 3 GB/s. And an M.2 slot wired to the chipset rather than the CPU shares an uplink with everything else on that chipset, so its usable share drops as soon as another device gets busy.

Precise Enough for Encoding Math

Results carry up to eight decimals, so an encoding-adjusted rate such as 63.015 Gbps reads back fine-grained enough to set against a benchmark chart.

Built for Spec-Sheet Comparison

Work through a list of slots and enclosures one after another, copying each bare result into your own table with no unit suffix to strip.

Reviews Read Backwards Too

A measured GB/s result turns straight back into the Gbps the link was carrying — how you tell a drive that saturated its bus from one that merely looked fast.

No Round Trip to a Server

Conversion happens on your own machine as you type, so a long parts list stays as quick at the end as at the start, even on a phone in a shop.

Bus Throughput Questions

Why does a PCIe 4.0 x4 slot give about 7.88 GB/s instead of a flat 8?

The 8 GB/s figure is 64 Gbps divided by eight and nothing more. On the wire, PCIe 3.0 and later transmit 130 bits for every 128 bits of payload so the receiver can recover its clock and stay aligned. That leaves 128/130 of the raw rate — about 98.5%, or roughly 7.88 GB/s. Transaction-layer headers take a further slice, which is why benchmarks top out slightly below even that.

What holds a SATA SSD to around 550 MB/s on a 6 Gbps port?

Two layers of overhead stacked on the division. Six gigabits per second is 0.75 GB/s, but SATA uses 8b/10b encoding, spending two bits in every ten on framing and DC balance. That drops the payload ceiling to 600 MB/s, and command overhead removes the rest — which is why a modern SATA drive lands near 550 MB/s whatever NAND is inside it. The interface, not the flash, is the bottleneck.

My drive is rated 7 GB/s but small-file work feels slow. Where did the speed go?

Advertised throughput is a sequential figure: one long stream of large blocks, queued deep, measured on a cool drive. Copying thousands of small files is a random workload, where the meaningful metric is operations per second rather than gigabytes per second. A million random 4 KiB reads per second is about 4 GB/s of payload, but at a queue depth of one — how a desktop actually behaves — the same drive delivers a fraction of that, with the bus nowhere near saturated either way.

Why do drive makers quote GB/s when the slot is specified in Gbps?

The two numbers describe different things. A slot specification is an electrical property of the link and belongs in bits per second, the unit signalling engineers work in. The drive figure is a measured result at the file level, where data is counted in bytes. Quoting a drive in bits would also make it look eight times more impressive on the shelf, so the byte figure is both the honest unit and the modest one.

Will a 40 Gbps Thunderbolt enclosure run an NVMe drive at full speed?

No, and the arithmetic shows why before you buy. Forty gigabits per second divides to 5 GB/s, but that link also carries display and USB traffic, and the PCIe tunnel inside it gets only a portion. In practice an external NVMe drive on Thunderbolt lands closer to 3 GB/s, so a drive rated 7 GB/s internally feels much like one rated 3.5 GB/s in that enclosure — the enclosure sets the pace, not the drive.