Tweet
There are already some example USB loopback benchmark results on our FAQ page. These were all done on Windows however and we didn't have any Linux examples.
The current loop back plug only supports the USB3.0 5Gbps singling rate. We are working on a 20G version, but it isn't available yet (as at May 2025).
Data rates will never reach these speeds on a real device however because of 8b/10b encoding, protocol overhead, checksums and host and device latency.
For perfect hardware and software (also never the case) you might in theory get 3.8Gbps, after the mandatory protocol overheads.
So what is possible in real life, in terms of useful data transfer.
Test setup.
Motherboard: MAG-X870E-TOMAHAWK-WIFI
CPU: AMD R9 9950X3D
OS: Ubuntu 24.04.2 LTS & Win11
RAM: 64GB
Ports tested:
Front 5G, Rear 5G, 10G & 40G
Linux Results:
Max read speed in benchmark mode: 3207 Mbit/s
Max write speed in benchmark mode: 3048 Mbit/s
Windows results:
Max read speed in benchmark mode: 3381 Mbit/s
Max write speed in benchmark mode: 3340 Mbit/s
Comments:
There was some variation between the different rear ports and some variation between runs. For this motherboard the 5G port was actually slightly faster than the 10G and 40G ports (but only ~3% faster). This 3% variation is counter intuitive, but likely there is some extra latency. Maybe the higher speed ports on this motherboard aren't directly connected to the CPU and instead are connected to a different controller chip. Windows was slightly faster than Linux, up to 10%. Likley this is due to device driver differences. On Windows we provide our own driver, but on Linux the built in general LibUSB driver is used. But the host controller driver is also important. (Single threaded CPU speed and PCIe bus speeds would also be important, but on this test system, there wasn't a bottleneck)
Screen shots
The current loop back plug only supports the USB3.0 5Gbps singling rate. We are working on a 20G version, but it isn't available yet (as at May 2025).
Data rates will never reach these speeds on a real device however because of 8b/10b encoding, protocol overhead, checksums and host and device latency.
For perfect hardware and software (also never the case) you might in theory get 3.8Gbps, after the mandatory protocol overheads.
So what is possible in real life, in terms of useful data transfer.
Test setup.
Motherboard: MAG-X870E-TOMAHAWK-WIFI
CPU: AMD R9 9950X3D
OS: Ubuntu 24.04.2 LTS & Win11
RAM: 64GB
Ports tested:
Front 5G, Rear 5G, 10G & 40G
Linux Results:
Max read speed in benchmark mode: 3207 Mbit/s
Max write speed in benchmark mode: 3048 Mbit/s
Windows results:
Max read speed in benchmark mode: 3381 Mbit/s
Max write speed in benchmark mode: 3340 Mbit/s
Comments:
There was some variation between the different rear ports and some variation between runs. For this motherboard the 5G port was actually slightly faster than the 10G and 40G ports (but only ~3% faster). This 3% variation is counter intuitive, but likely there is some extra latency. Maybe the higher speed ports on this motherboard aren't directly connected to the CPU and instead are connected to a different controller chip. Windows was slightly faster than Linux, up to 10%. Likley this is due to device driver differences. On Windows we provide our own driver, but on Linux the built in general LibUSB driver is used. But the host controller driver is also important. (Single threaded CPU speed and PCIe bus speeds would also be important, but on this test system, there wasn't a bottleneck)
Screen shots