I have a ClockworkPi uConsole and although I’m having great fun with it, it comes with just 32GB of SD card storage. Even if it were bigger, SD cards are dog slow compared to almost anything else. I bought a HackerGadgets NVMe adapter and Crucial P310 1TB NVMe SSD to speed things up.

Sequential operations

In these tests, /dev/mmcblk0 is the SD card, and /dev/nvme0n1 is the NVMe adapter.

Reads

root@uconsole /h/me# time dd if=/dev/mmcblk0 of=/dev/null bs=1M count=8192
8192+0 records in
8192+0 records out
8589934592 bytes (8.6 GB, 8.0 GiB) copied, 198.193 s, 43.3 MB/s

________________________________________________________
Executed in  198.20 secs    fish           external
   usr time    0.06 secs    0.03 millis    0.06 secs
   sys time   23.06 secs    2.00 millis   23.06 secs

root@uconsole /h/me# time dd if=/dev/nvme0n1 of=/dev/null bs=1M count=8192
8192+0 records in
8192+0 records out
8589934592 bytes (8.6 GB, 8.0 GiB) copied, 21.366 s, 402 MB/s

________________________________________________________
Executed in   21.37 secs    fish           external
   usr time    0.03 secs    1.59 millis    0.03 secs
   sys time   10.38 secs    0.86 millis   10.38 secs

Summary: The NVMe drive was 9.3x faster than the SD card and took 55% less CPU.

Writes

I couldn’t write to the raw SD card because all of it is allocated in the partition map and it’s what I boot off of as of this writing. Instead, I formatted the NVMe drive with the same options as the root filesystem and mounted it at /tmp/nvme so that at least it’s an apples-to-apples comparison

root@uconsole /h/me# time dd if=/dev/zero of=/sdfile bs=1M count=8192
8192+0 records in
8192+0 records out
8589934592 bytes (8.6 GB, 8.0 GiB) copied, 331.464 s, 25.9 MB/s

________________________________________________________
Executed in  331.47 secs    fish           external
   usr time    0.05 secs    0.25 millis    0.05 secs
   sys time   25.98 secs    2.16 millis   25.98 secs

root@uconsole /h/me# time dd if=/dev/zero of=/tmp/nvme/sdfile bs=1M count=8192
8192+0 records in
8192+0 records out
8589934592 bytes (8.6 GB, 8.0 GiB) copied, 21.0683 s, 408 MB/s

________________________________________________________
Executed in   21.10 secs    fish           external
   usr time    0.03 secs    2.16 millis    0.03 secs
   sys time   18.92 secs    1.13 millis   18.92 secs

Summary: The NVMe drive was 15.7x faster than the SD card, and took 27% less CPU.

Random operations

These use the fio benchmark. First, the SD card:

root@uconsole /v/tmp# fio --profile=tiobench
seqwrite: (groupid=0, jobs=1): err= 0: pid=3128: Mon Jul  6 19:40:57 2026
  write: IOPS=1045, BW=4181KiB/s (4281kB/s)(30.5MiB/7482msec); 0 zone resets
randwrite: (groupid=1, jobs=1): err= 0: pid=3161: Mon Jul  6 19:40:57 2026
  write: IOPS=692, BW=2771KiB/s (2838kB/s)(30.5MiB/11287msec); 0 zone resets
seqread: (groupid=2, jobs=1): err= 0: pid=3206: Mon Jul  6 19:40:57 2026
  read: IOPS=2917, BW=11.4MiB/s (12.0MB/s)(30.5MiB/2680msec)
randread: (groupid=3, jobs=1): err= 0: pid=3255: Mon Jul  6 19:40:57 2026
  read: IOPS=3075, BW=12.0MiB/s (12.6MB/s)(30.5MiB/2543msec)

Run status group 0 (all jobs):
  WRITE: bw=4181KiB/s (4281kB/s), 4181KiB/s-4181KiB/s (4281kB/s-4281kB/s), io=30.5MiB (32.0MB), run=7482-7482msec

Run status group 1 (all jobs):
  WRITE: bw=2771KiB/s (2838kB/s), 2771KiB/s-2771KiB/s (2838kB/s-2838kB/s), io=30.5MiB (32.0MB), run=11287-11287msec

Run status group 2 (all jobs):
   READ: bw=11.4MiB/s (12.0MB/s), 11.4MiB/s-11.4MiB/s (12.0MB/s-12.0MB/s), io=30.5MiB (32.0MB), run=2680-2680msec

Run status group 3 (all jobs):
   READ: bw=12.0MiB/s (12.6MB/s), 12.0MiB/s-12.0MiB/s (12.6MB/s-12.6MB/s), io=30.5MiB (32.0MB), run=2543-2543msec

Disk stats (read/write):
  mmcblk0: ios=15548/15656, sectors=124384/125480, merge=0/27, ticks=4870/18365, in_queue=23235, util=93.05%

And then the NVMe drive:

root@wizzle /t/nvme# fio --profile=tiobench
seqwrite: (groupid=0, jobs=1): err= 0: pid=4865: Mon Jul  6 19:47:10 2026
  write: IOPS=21.6k, BW=84.4MiB/s (88.5MB/s)(30.5MiB/362msec); 0 zone resets
randwrite: (groupid=1, jobs=1): err= 0: pid=4870: Mon Jul  6 19:47:10 2026
  write: IOPS=21.2k, BW=82.8MiB/s (86.8MB/s)(30.5MiB/369msec); 0 zone resets
seqread: (groupid=2, jobs=1): err= 0: pid=4871: Mon Jul  6 19:47:10 2026
  read: IOPS=13.3k, BW=51.9MiB/s (54.4MB/s)(30.5MiB/589msec)
randread: (groupid=3, jobs=1): err= 0: pid=4876: Mon Jul  6 19:47:10 2026
  read: IOPS=13.1k, BW=51.3MiB/s (53.7MB/s)(30.5MiB/596msec)

Run status group 0 (all jobs):
  WRITE: bw=84.4MiB/s (88.5MB/s), 84.4MiB/s-84.4MiB/s (88.5MB/s-88.5MB/s), io=30.5MiB (32.0MB), run=362-362msec

Run status group 1 (all jobs):
  WRITE: bw=82.8MiB/s (86.8MB/s), 82.8MiB/s-82.8MiB/s (86.8MB/s-86.8MB/s), io=30.5MiB (32.0MB), run=369-369msec

Run status group 2 (all jobs):
   READ: bw=51.9MiB/s (54.4MB/s), 51.9MiB/s-51.9MiB/s (54.4MB/s-54.4MB/s), io=30.5MiB (32.0MB), run=589-589msec

Run status group 3 (all jobs):
   READ: bw=51.3MiB/s (53.7MB/s), 51.3MiB/s-51.3MiB/s (53.7MB/s-53.7MB/s), io=30.5MiB (32.0MB), run=596-596msec

Disk stats (read/write):
  nvme0n1: ios=13873/15650, sectors=110984/125184, merge=0/4, ticks=866/486, in_queue=1354, util=49.13%

Summary

BW measurements are in MB/s.

Conclusion

In the words of the great Ferris Bueller, “It is so choice. If you have the means, I highly recommend picking one up.”