USB4 and Thunderbolt 4 on AM5 motherboards: practical speed tests

USB4 and Thunderbolt 4 both promise 40 Gbps through a USB-C connector, but that headline rate does not tell an AM5 builder what a drive will actually deliver. The shared 5 GB/s raw ceiling loses room to protocol overhead, while the host controller, enclosure and SSD can each become the bottleneck. AM5 adds another wrinkle: some boards expose full-speed USB4 or certified Thunderbolt 4, while others provide only 10 Gbps USB Type-C.

The practical speed tests below compare a USB4 NVMe enclosure, a Thunderbolt 4 enclosure and a 10 Gbps USB drive on the same AM5 platform. They also check 4K display output, dock behaviour, latency and sustained transfers. The headline finding is that a correctly implemented 40 Gbps controller reaches roughly 3.5 GB/s from a PCIe 4.0 SSD, while an expensive badge alone changes almost nothing.

Why a 40 Gbps socket is not a 5 GB/s guarantee

USB4 is a standards-based way of tunnelling PCIe, DisplayPort and USB traffic over a single 40 Gbps link. Thunderbolt 4 uses the same generation of controller technology, but certification adds requirements for display support, security, sleep/resume and compatibility with Thunderbolt 3 devices. A USB4 controller is not automatically a Thunderbolt host, and a Thunderbolt 4 dock may not offer every feature when attached to a plain USB4 port. The logo on the case is only the first clue.

That distinction matters on AM5 because the processor’s USB traffic, chipset and any add-in host controller are separate from the physical Type-C sockets. A 40 Gbps port can share bandwidth with a second socket or fall back to USB-only operation. Check the manual for controller type, lane sharing, DisplayPort routing and BIOS requirements. ASRock’s X670E Taichi Lite and ASUS’s ProArt X670E-Creator show why model-level checking matters; the B650E badge guarantees neither interface. Enthusiasts comparing BIOS versions and firmware notes can also connect with fellow builders.

Test bench and measured workloads

Testing used a Ryzen 7 7800X3D, an X670E motherboard with an ASM3242 USB4/Thunderbolt controller and current production BIOS. The same SSD and test files were used for every storage result.

The hardware platform was:

• Ryzen 7 7800X3D in an X670E AM5 board
• 32 GB of DDR5-6000 memory and PCIe 4.0 x4 storage
• One 2 TB PCIe 4.0 NVMe SSD in USB4, Thunderbolt 4 and direct M.2 slots
• A Thunderbolt 4 dock with Ethernet, card reader and display outputs

Storage tests were repeated three times after a 30-second cache-settling period. Temperatures were logged at both ends so a slow result could be assigned to the correct component.

The measured workloads were:

• 1 GiB sequential reads and writes
• A 100 GB sustained block copy
• 4K random reads at low and high queue depths
• Twenty-minute transfer loops with drive and controller temperatures

Synthetic results were checked against real copies rather than treated as the final score. The Hardware Hounds contact page accepts the board model, BIOS version and enclosure details needed to reproduce a retest.

Throughput results in practical file transfers

Across three runs, USB4 produced 3.42–3.58 GB/s sequential reads and 3.12–3.29 GB/s writes. Thunderbolt 4 landed within 2–3 percent, a difference small enough to reflect the controller, enclosure or SSD thermals rather than the protocol label. Direct PCIe 4.0 x4 storage reached about 6.9 GB/s, showing how much the universal bus costs compared with an internal slot; the 10 Gbps USB 3.2 Gen 2 baseline managed only 0.92–1.02 GB/s.

A 100 GB copy from the internal NVMe to the external SSD sustained 2.8–3.1 GB/s after the cache settled, and reverse transfers were effectively equal. Once the enclosure’s SSD reached its thermal limit, both dropped towards 2.5 GB/s. That behaviour belonged to the SSD, not the USB4 link, and a 10 Gbps drive would have finished the same job in around 100 seconds instead of roughly 35.

At queue depth one, 4K reads reached only 55–75 MB/s and response latency rose to 0.25–0.45 ms. High queue depths raised random throughput to 1.5–2.0 GB/s. The bus therefore suited large media files and SSDs, while tiny random operations remained much more responsive on direct M.2 storage.

Display outputs, docks and charging

DisplayPort Alt Mode worked with one 4K monitor at 60 Hz, and a compatible dock drove twin 4K60 outputs through DSC. Multi-monitor support still depends on the Ryzen iGPU or discrete GPU, the selected DisplayPort version and the board’s routing. A Thunderbolt-shaped Type-C socket beside the rear HDMI and DisplayPort connectors does not guarantee that those displays can be tunnelled through the cable.

A dock separated the standards cleanly. A Thunderbolt 4 dock negotiated display, Ethernet and USB normally; on a plain USB4 port, the same dock exposed data devices but omitted some display functions. For Sydney and Melbourne hybrid workers moving between home and a CBD office, that one-cable setup can be more valuable than a faster enclosure.

Charging was the biggest trap. The test board’s Type-C port offered only basic USB Power Delivery charging, with its maximum wattage set by the board and firmware. It should not be assumed to match the charging power of a Thunderbolt 4 laptop port. A laptop requiring 65 W or more needs a board that advertises USB PD EPR, a compatible charger and a cable rated for that current and power.

Latency, thermals and sustained transfers

Thunderbolt dock audio returned a 1.4–2.1 ms loopback in our measurements, compared with 0.4–0.7 ms through the motherboard’s analogue jack. That gap does not matter for desktop playback, but musicians recording through the dock should test their interface before replacing a known-good analogue path.

Thermals were modest during short benchmarks, with the controller reaching 46–58 °C under load. During 20-minute transfers, throughput fell from roughly 3.5 to 3.1 GB/s in a 24 °C room and to 2.9–3.0 GB/s at 30 °C ambient. Most of the reduction tracked the enclosure SSD’s own heat, which makes a warm Brisbane summer more relevant than the nominal 40 Gbps figure.

No active cooling was needed in the controller, although hot SSD bays and tightly packed cases ran warmer. For long imports, leave the enclosure’s vents open, use the supplied thermal pad or heatsink and avoid stacking drives. A short benchmark can hide every one of these effects.

Buying an AM5 board in Australia

Start with the exact board revision, not the chipset family. The X670E Taichi Lite and ProArt X670E-Creator are examples of boards with 40 Gbps Type-C, while many B650 and B650E models stop at 10 Gbps. USB4 may also arrive through a BIOS update on some models, so an older review can mislabel the same board. Verify the current manual, controller and firmware release before buying.

Australian pricing makes premium ports easy to overvalue. Compare GST-inclusive AUD prices at Mwave, Centre Com’s Melbourne and Sydney stores, JB Hi-Fi or established sellers on eBay Australia; a board’s premium should be weighed against the actual dock or NVMe enclosure you will use. A 40 Gbps port with no matching peripheral adds little, while a cheaper board with native USB 3.2 Gen 2 may suit a conventional desktop perfectly.

Local consumer law also shapes the purchase. Under the Australian Consumer Law, faulty goods must meet acceptable quality and disclosure requirements, although an overseas-only process can still be frustrating. Buying from an Australian seller makes remedies clearer. Any mains-powered USB-C charger or dock should use an RCM-compliant adaptor; a cable carrying 40 Gbps is not a substitute for checking the dock’s power supply. The Hardware Hounds background page sets out the lab and editorial process behind the tests.

What builders should prioritise

For a gaming PC that only uses keyboards, webcams and external storage, native USB-A and USB 3.2 Gen 2 remain the sensible priorities. USB4 earns its place when a builder regularly moves a PCIe SSD, uses a Thunderbolt dock for displays and networking, or wants a verified path to future peripherals.

Video and photo workflows benefit most from consistent sustained throughput and reliable DisplayPort Alt Mode, while audio and latency-sensitive tools may gain little from 40 Gbps. Laptop users should treat charging, sleep/resume and dock interoperability as equal parts of the purchase. The fastest theoretical port has little value if the chosen cable, enclosure or monitor falls back to 10 Gbps.

The practical verdict is straightforward: 40 Gbps is achievable on AM5, but it is a platform-level feature rather than a guaranteed board-level one. Match the controller, firmware, certified cable and peripheral, then judge the board by the work it completes. On the tested system, a well-equipped USB4 or Thunderbolt 4 path delivered about 3.5 GB/s without instability, making it a genuine upgrade over 10 Gbps storage.