Although the performance of Thunderbolt 3 and 4 is generally more consistent and reliable than USB 3.x, what you get in terms of real-world transfer speed for the fastest SSDs is around 3 GB/s maximum. That’s significantly less than that achieved by Apple’s better internal SSDs, because only 32 Gb/s of Thunderbolt’s total 40 is available for general data, as four lanes of PCI Express.
I’m in the process of testing a brand new Thunderbolt 4 hub, one of the first of this year’s second generation, and have been measuring its performance in use, something that appears unusual. This article summarises some of my findings that have more general importance.
By a Thunderbolt hub, I mean a peripheral device that connects to a single Thunderbolt port on your Mac, and provides multiple Thunderbolt ports to which you can connect other peripherals. The first computers to offer Thunderbolt 3 ports started shipping at the end of 2015, and Apple’s first Mac with Thunderbolt 3 support was released the following year. Yet it wasn’t until October 2020 that we were able to buy a Thunderbolt hub.
I used my two fastest Thunderbolt 3 SSDs: to protect the innocent, I’ll here refer to them as A and B. When connected directly to a Thunderbolt port on the back of my Mac Studio M1 Max, and benchmarking using my normal method in Stibium, disk A delivers read/write speeds of 3.0/2.1 GB/s, and Disk B 2.8/2.8 GB/s.
I then repeated write and read tests in exactly the same way, for the following hardware combinations:
- both SSDs connected via the hub to the same Thunderbolt port on the Mac Studio;
- both SSDs connected via the hub to a Thunderbolt port on my MacBook Pro 16-inch M1 Pro;
- ditto, but with my Studio Display also connected to the hub, and running in standard resolution.
Measuring transfer speeds to both SSDs simultaneously was more of a challenge. Because of the marked difference between their write speeds, I found it impossible to run concurrent write tests, as the slower SSD always had the benefit of finishing later, when the other SSD was already done. However, their read speeds are more similar, and tests could be started and completed at almost the same time. I therefore measured read speeds for each SSD individually, and both of them together.
Few if any other benchmarking tools can be run in tandem in this way, because of the protocols they use for tests. Fortunately, Stibium is designed to run straight through its set of read tests as quickly as possible, and should therefore be reading continuously from both disks throughout the test.
Direct or hub?
Simply adding the hub to the connection, as in the first two combinations, had little effect on solo read speeds measured on either SSD. The largest fall in read speed was seen in Disk B, which fell from 2.8 to 2.6 GB/s on the Studio, but still achieved 2.8 GB/s on the MacBook Pro.
It was Disk A that showed the greatest fall in solo write performance, from 2.1 to 1.5 GB/s on the MacBook Pro. Write speed through the hub then appeared to perform as if the SSD was only using two of the four lanes of PCI Express, as some older and cheaper Thunderbolt SSDs do in normal use. Disk A’s solo write speed was almost identical on the MacBook Pro and the Studio at 1.5 GB/s. This was entirely unexpected, but consistent across all write tests using Disk A.
Simultaneous read speeds, when corrected for the slight asynchrony that occurred in testing, were around 1.5 GB/s for both disks, on both Macs. Taken as a total, they reach the practical limit of Thunderbolt 3 disks at 3 GB/s, as would be expected.
With a Thunderbolt Display
All measurements of disk transfer speed were lower when a Studio display was also connected to the hub, although the reduction was in most cases less than 10%. The largest fall was in Disk B’s write speed, which dropped from 2.8 on the hub without the display, to 2.1 GB/s with the display. However, solo read speeds for each disk remained close to the magic 3 GB/s, at 2.8 for Disk A, and 2.6 GB/s for Disk B, and simultaneous read speeds still totalled close to 3 GB/s.
- Connecting Thunderbolt 3 SSDs via a hub can significantly reduce write speeds on some disks, but not others.
- Connecting Thunderbolt 3 SSDs via a hub should have little effect on their read speeds.
- Total simultaneous read speeds for two SSDs connected via a hub should be close to the practical maximum of 3 GB/s.
- Simultaneous connection of a display is likely to reduce SSD performance, and in some cases that may be substantial. Read speeds are likely to be less affected than write speeds.
- Thunderbolt 4 hubs are a valuable way of adding ports to a Mac, but their total transfer speed to the host can’t exceed that normally achieved over a single Thunderbolt port. If you don’t need the additional ports, it’s better to connect Thunderbolt devices directly when you can.
As usual, if you want to read the full review, it will be published in a future edition of Mac|Life and MacFormat magazines.