3 ms·

With a decent filesystem and a ram buffer the better microsd cards still beat an old 5400rpm 128GB HDD's sequential speed when doing random writes of ~500kB fil

With a decent filesystem and a ram buffer the better microsd cards still beat an old 5400rpm 128GB HDD's sequential speed when doing random writes of ~500kB files by a fair margin. 150MB/s is nowhere near SSD speed, but it would saturate a SATA I port.

yourusername 4y ago
Considering 100 GB HDD's came out 20+ years ago that's not very suprising. If people are comparing SD cards to HDD's they're probably comparing to HDD's made in the last 10 years.
- Schroedingersat 4y ago
  OP specifically cited an old hdd in the pre-ssd days. New HDDs are still at best on par with a high end microsd used as internal storage (compare an A2 microsd 'rated' at 2000 but actually getting about 800 iops vs a high end hdd benchmarking at 200-300). Mounting without a buffer on a badly chosen filesystem in what could well be the M1's abysmally slow sd reader isn't a representative test. Hare braned formatting tools will also often offset the filesystem by 512 Bytes (necessitating two writes and two reads for one 4k write) or use blocks that aren't 4kiB
  - hot_gril 4y ago
    FWIW my MacBook Pro is the 2015 Intel version, and my card is HFS+ formatted (Android would use FAT32 or Ext4). I got similar performance with an external USB reader IIRC, but maybe it's time to test again.
    - Schroedingersat 4y ago
      Aye, there's your problem. https://web.archive.org/web/20151222043711/https://support.apple.com/en-us/HT204384 https://web.archive.org/web/20151222043711/https://support.a... It's like connecting a QLC SSD through an IDE port and complaining it's slow. It /is/ slow but not that slow.
      - hot_gril 4y ago
        Is the issue the 480mbit/s USB limit? That's still a lot more than what I was getting with random writes.
        Schroedingersat 4y ago
        USB 2 has very poor latency for this use case and the sd connector lacks the pins for newer sd cards. A usb 3 UHS III reader would be able to talk to the card at full speed, then it's a matter of convincing the OS to trim it and buffer as if it's an internal hdd. Kinda expensive and pointless to go to that effort to get 2x hdd performance when there are NVME enclosures for cheaper SSDs that have 100x the IOPs though.
hot_gril 4y ago
I'm not sure about that test cause there are too many variables, but in the typical real-world usage of like cloning a git code repo onto my SD card or opening a zip file, it was always painfully slow. I wasn't even benchmarking, I was just trying to get stuff done and not wanting to wait on it. In fact I moved the files to an HDD to do it one time.

yourusername 4y ago

Considering 100 GB HDD's came out 20+ years ago that's not very suprising. If people are comparing SD cards to HDD's they're probably comparing to HDD's made in the last 10 years.

Schroedingersat 4y ago
OP specifically cited an old hdd in the pre-ssd days. New HDDs are still at best on par with a high end microsd used as internal storage (compare an A2 microsd 'rated' at 2000 but actually getting about 800 iops vs a high end hdd benchmarking at 200-300). Mounting without a buffer on a badly chosen filesystem in what could well be the M1's abysmally slow sd reader isn't a representative test. Hare braned formatting tools will also often offset the filesystem by 512 Bytes (necessitating two writes and two reads for one 4k write) or use blocks that aren't 4kiB
- hot_gril 4y ago
  FWIW my MacBook Pro is the 2015 Intel version, and my card is HFS+ formatted (Android would use FAT32 or Ext4). I got similar performance with an external USB reader IIRC, but maybe it's time to test again.
  - Schroedingersat 4y ago
    Aye, there's your problem. https://web.archive.org/web/20151222043711/https://support.apple.com/en-us/HT204384 https://web.archive.org/web/20151222043711/https://support.a... It's like connecting a QLC SSD through an IDE port and complaining it's slow. It /is/ slow but not that slow.
    - hot_gril 4y ago
      Is the issue the 480mbit/s USB limit? That's still a lot more than what I was getting with random writes.
      - Schroedingersat 4y ago
        USB 2 has very poor latency for this use case and the sd connector lacks the pins for newer sd cards. A usb 3 UHS III reader would be able to talk to the card at full speed, then it's a matter of convincing the OS to trim it and buffer as if it's an internal hdd. Kinda expensive and pointless to go to that effort to get 2x hdd performance when there are NVME enclosures for cheaper SSDs that have 100x the IOPs though.

hot_gril 4y ago

I'm not sure about that test cause there are too many variables, but in the typical real-world usage of like cloning a git code repo onto my SD card or opening a zip file, it was always painfully slow. I wasn't even benchmarking, I was just trying to get stuff done and not wanting to wait on it. In fact I moved the files to an HDD to do it one time.