Gbps vs GB/s: Reading eGPU Throughput Numbers Correctly

Two eGPU connections quote almost the same number on the box, yet one is meaningfully faster in real use. That gap comes down to a unit confusion that costs buyers money: Gbps versus GB/s. They look alike, differ by a factor of eight, and marketing leans on the bigger-sounding one. OCuLink's real-world throughput of roughly 6.6 to 6.7 GB/s actually beats Thunderbolt 5's roughly 5.6 to 5.8 GB/s, even though both get advertised somewhere near 64 to 80 Gbps. Reading the numbers correctly tells you which link will feed your graphics card better.

Quick Answer

Gbps (gigabits per second) and GB/s (gigabytes per second) are not the same: divide the Gbps figure by 8 to get bytes, then subtract protocol overhead for the usable rate. By that maths, OCuLink delivers roughly 6.6 GB/s in practice against Thunderbolt 5's roughly 5.6 GB/s, despite similar box specs. For an eGPU, OCuLink is the faster real-world link.

The Eight-Times Difference Nobody Explains

A bit is a single one or zero; a byte is eight bits. Network and interface speeds are quoted in bits per second (Gbps) because the number looks bigger, while file sizes and drive speeds are measured in bytes (GB/s). So a connection rated at 64 Gbps is not 64 GB/s; it is 64 divided by 8, or 8 GB/s, before any losses. Miss that and you overestimate a link's data rate eightfold.

That single conversion is where most eGPU confusion starts. The box shouts a headline Gbps figure, the buyer mentally treats it like GB/s, and two products that look far apart on paper turn out close, or even reversed, once you do the division. Always convert to the same unit before comparing anything.

Why the Box Number Is Never What You Get

Conversion is only half the story. The advertised Gbps is the raw signalling rate, and real data never gets all of it. Protocols add overhead, headers, encoding, handshaking, error correction, that consume a slice of the bandwidth. So the usable GB/s is always lower than the raw figure divided by eight. A link rated 80 Gbps does not deliver 10 GB/s of clean data; the overhead trims it well below that.

This is why two interfaces with similar headline Gbps can perform differently. OCuLink is a lean, PCIe-style link with relatively low overhead, so a large fraction of its raw rate survives as usable throughput, landing around 6.6 to 6.7 GB/s. Thunderbolt 5 carries more protocol overhead and shares its lane with other traffic, so its real eGPU throughput sits lower, around 5.6 to 5.8 GB/s, even where its box figure looks competitive. The headline number flatters the busier protocol.

How To Read eGPU Throughput Correctly

Use a simple routine on every spec sheet. First, take the Gbps figure and divide by 8 to get the theoretical bytes per second. Second, knock off protocol overhead to estimate the usable rate; a rough rule is that real throughput lands meaningfully below the divided figure, more so on heavier protocols. Third, compare like with like in GB/s, never one product's Gbps against another's GB/s.

For an eGPU specifically, this matters because the link feeds a graphics card that wants bandwidth. More usable GB/s means less bottleneck between the GPU and the rest of the system, which shows up as smoother frame delivery in bandwidth-sensitive scenarios. When you weigh up an external GPU connection, the docking and connectivity options that external GPU users rely on make it easier to see which interface a given setup actually uses.

OCuLink vs Thunderbolt 5 in Plain Terms

Read past the marketing and the picture is clear: OCuLink's lower overhead gives it the higher real-world throughput for an eGPU, roughly a gigabyte per second more usable bandwidth than Thunderbolt 5 in practice. Thunderbolt 5 wins on flexibility, since it carries video, data, charging and daisy-chaining over one cable, which OCuLink does not. If raw eGPU bandwidth is the priority, OCuLink leads; if you want one cable doing everything, Thunderbolt 5 earns its place. Either way, judge them in GB/s, not the Gbps on the sticker. To pair the right card with whichever link you choose, the graphics cards buyers reach for most show where the sensible performance tiers sit.

Frequently Asked Questions

How do I convert Gbps to GB/s?

Divide the Gbps figure by 8, since there are eight bits in a byte. A 64 Gbps link is therefore 8 GB/s in theory. Then subtract protocol overhead to get the usable real-world rate, which is always lower than the converted figure.

Why is OCuLink faster than Thunderbolt 5 for an eGPU?

OCuLink is a lean PCIe-style link with low overhead, so more of its raw bandwidth survives as usable throughput, around 6.6 to 6.7 GB/s. Thunderbolt 5 carries more protocol overhead, leaving roughly 5.6 to 5.8 GB/s for eGPU data despite similar box specs.

Does the box number tell me the real speed?

No. The advertised Gbps is the raw signalling rate before conversion and overhead. The usable GB/s is always lower, and how much lower depends on the protocol. Always convert and account for overhead rather than trusting the headline figure.

Should I pick OCuLink or Thunderbolt 5?

Pick OCuLink if maximum eGPU bandwidth is your priority, since it delivers more usable throughput. Pick Thunderbolt 5 if you want one cable that also carries video, charging and other devices. The right choice depends on whether you value raw speed or all-in-one flexibility.