Among issues reported in Apple’s original M1 Macs were poor write speeds to external SSDs in certain configurations. Examples quoted include speeds as low as 1.0-1.1 GB/s for OWC Envoy Pro models. OWC engineers discovered that these could be improved to around 1.8 GB/s by connecting the Mac to a Thunderbolt display, and kindly wrote this up on the OWC blog.
If you use an original M1 Mac with an external SSD and have the option of connecting its display by HDMI or Thunderbolt, it’s now clear that it’s worth checking to see whether using Thunderbolt for the display might also improve write performance to the SSD. This doesn’t appear to affect read performance, though, only write.
Since that report a year ago, there have been various rumours about similar issues affecting later M1 models including those using M1 Pro and even M1 Max chips. If that were to be correct, it would be of even greater consequence, as those models support multiple external displays including Apple’s new Studio Display. Moreover, it wouldn’t be consistent with theory proposed by OWC to account for the problems in original M1 Macs, which is dependent on them only supporting a single display. This article assesses whether similar problems afflict the Thunderbolt 4 ports on later models of MacBook Pro with M1 Pro or Max chips.
Testing was performed using a Samsung X5 SSD, which is known to throttle after relatively small amounts of data have been written to it. To ensure that throttling didn’t occur during testing, the amount of data written was limited to 10 GB in each test. Because of the robust statistical methods used to analyse individual test results, this still yielded reproducible data, and is a good illustration of the importance of sound and robust analytical methods when measuring disk performance.
The host used for these tests is a MacBook Pro 16-inch 2021 with an M1 Pro chip (MacBookPro18,1), 32 GB of memory and 2 TB internal SSD running macOS 12.3.1 (Build 21E258) in Full Security mode. That was tested with an Apple Studio Display connected by its supplied Thunderbolt cable, and a Samsung X5 500 GB SSD connected using a passive 0.8 m Thunderbolt 4 cable. Both the display and the external SSD were running current versions of their firmware.
Write speed measurements were made using Stibium version 1.0 build 55, using the Write Series feature to write two complete sets of test files, making a total of 32 files ranging from 2 MB to 2 GB. The SSD was freshly formatted in APFS (unencrypted) at the start of testing. Following each test, all test file were removed and the SSD given time to ensure its SLC Write cache could refresh, to ensure that write speeds weren’t throttled during any test.
There are three Thunderbolt ports on this model. For these test purposes, I numbered each to ensure clarity. Port 1 is that at the back left, next to the MagSafe port; port 2 is the front left port, between port 1 and the audio socket; port 3 is on the right side.
Testing consisted of connecting the display and external SSD to every possible combination of Thunderbolt ports in succession. Thus, when the display was connected to port 1, the external SSD was tested when connected to ports 2 and 3, and so on. Initial testing was performed without the display connected, and the external SSD connected to ports 1, 2 and 3 in succession.
In every case, overall write rate as estimated using Theil-Sen regression was between 2.2 and 2.3 GB/s. There was no evidence of any significant difference between combinations of ports in calculated averages, linear regressions, or 20% trimmed means. The graph below shows the standard pattern of write rates by file size, which again was consistent across every combination of ports.
As is often the case, write speeds at the smallest file sizes were lower, rising from around 1.7 GB/s for 2 MB test files. Speeds then rose steadily to reach the sustained maximum at 60 MB and above. With the small number of tests performed (I normally run sets of 10 x 16 files rather than 2 x 16), there is a single outlier at 10 MB with a very low write rate of 0.8 GB/s. There’s also no evidence of any fall in speed with larger file sizes.
There is no evidence of any favoured combination of Thunderbolt ports which results in faster write rates to external SSDs in this MacBook Pro model.
Anecdotal evidence reported by others strongly suggests that this may not hold for Macs with the original M1 chip, however if you have a MacBook Pro 16-inch 2021 no similar effect occurs and you can connect your display and external SSD however you want without affecting write performance.
Although I haven’t tested my Mac Studio Pro in the same way, I see no reason to suspect that model is any different, although using the front Thunderbolt ports on the Studio Ultra could be more interesting.