I think almost everyone who has got an M1 Mac has tried running some sort of benchmark. This article looks at one specific feature – SMB Target Disk mode – assesses how fast it is, finds a nasty problem with it, and concludes by considering M1 performance more generally.
Setting up SMB Target Disk mode
Apple’s instructions assume that both Macs involved, which I’ll refer to as the Target and Host, are running at the time. I’ll begin with a more likely scenario, with both Macs shut down, and in my case the Target is an M1 Mac mini, and the Host is an M1 MacBook Pro.
Start the Target up with the Power button held in until the display shows Loading Startup Options, then release it. This takes you to the Startup Options screen. Select the Options icon, then click Continue underneath it. Once the main Recovery mode window has loaded (offering its four options), open the Utilities menu and select the Share Disk command. Work through the sequence to select the disk until it offers you a button to turn sharing on. Leave the Target then, and turn your attention to the Host.
Ensure the Host starts up fully before reaching for your Thunderbolt cable. If you connect the Host and Target using a Thunderbolt cable (at least) before you’ve started the Host up, it will enter an endless restart loop, in which that Mac starts, takes you through login, then shortly afterwards restarts again. Provided that you don’t connect the cable until the Host has started up fully, both Macs appear stable.
Once you’ve connected the two Macs using a Thunderbolt 3 cable, on the Target click to start sharing. It should then appear on the Host as a Network disk, to which you connect in the normal way, following authentication.
To disconnect, eject the Target on the Host and, on the Target, click on the button to stop sharing. You can then restart the Target back into normal mode.
File transfer performance
This is fast, really fast: don’t waste your time testing with files less than 10 GB, as your timing errors will be too great. I used two test files, one of 12.2 GB, the other 13.61 GB. Going from Target to Host, these transferred at 0.8 and 1.5 GB/s, and from Host to Target at 1.0 and 0.9 GB/s. On large files you should therefore expect to see rates ranging from 0.8-1.5 GB/s in each direction. For comparison, benchmark results for Apple’s high-speed internal SSDs are typically around 2.8 GB/s (see below). Over Thunderbolt, this is comparable to what you’d expect from 10 Gigabit Ethernet.
M1 Mac performance
As I wrote above, I have two M1 Macs, a mini and a MacBook Pro of similar specification. They have the better processor option with 8-core CPU and 8-core GPU, with 16 GB of memory and 500 GB internal SSDs.
Unsurprisingly, their benchmark performance is almost identical. Geekbench 5 scores are around 1730 (single-core) and 7600 (multi-core), and their OpenCL scores are just under 20,000. They both comfortably outperform my MacBook Pro 16-inch late 2019 with an 8-core i9 (1160 single, 6740 multi), although that does better with OpenCL at 30,500. More surprising, perhaps, is comparison with my mainstay Mac, a base specification iMac Pro with 8-core Xeon-W. That’s similar to the i9 on single-core, rather better with around 8,200 on multi-core, and far better with OpenCL at 52,400.
All my active Macs turn in similar performance on the Blackmagic Disk Speed Test, with both read and write speeds of around 2.8 GB/s to their internal SSDs.
Real versus virtual memory
Debate has raged over whether these initial low-end models, with a maximum of 16 GB of physical memory, could possibly substitute for an Intel Mac which currently works well with 32 GB or more of physical memory. It’s never possible to predict real-world performance from benchmarks and specifications, and the tight integration of the M1 design makes prediction even more prone to error.
For heavyweight tasks, an Intel Mac may require more memory than an M1, because its memory isn’t Universal, accessible to CPUs and GPUs. The M1 may well have to rely more heavily on virtual memory, which many of us still associate with periods of disk churning. Here it’s worth comparing transfer speeds for different storage types.
Memory built into the M1 chip carrier is 4266 MT/s LPDDR4X SDRAM, which should achieve transfer speeds of around 30 GB/s, compared with 21 GB/s that you’ll get from the 2666 MHz DDR4 fitted in the latest iMac, or my iMac Pro. With an internal SSD now achieving 2.8 GB/s through a T2 controller (for current Intel Macs) or the M1 chip, its performance should be similar to that of PC-2700 DDR SDRAM, last fitted to iMacs in 2003-04. Although one order of magnitude slower than the M1’s memory, good virtual memory management should maintain good performance for many tasks within the M1 chip’s limit of 16 GB.
Currently, the high-performance memory built into the M1 chip carrier is only available in 8 and 16 GB units, and appears to have been produced for low-volume markets. I expect that next year, as Apple starts to roll out new models with higher specifications, we’ll see them offered with more memory. But if Apple were to drop back to the type of DDR4 memory modules which could be upgraded by users, performance and energy consumption would both suffer badly.