from Hacker News

How reliable are MicroSD cards?

by edent on 6/2/25, 6:00 AM with 42 comments

• by duskwuff on 6/2/25, 8:10 AM

  Re. brands: A bunch of the names on this list are literally just brand names, not manufacturers. Figuring out the actual manufacturers could be interesting - this can be accomplished either by querying the cards, or by physically inspecting the card. (A couple of physical features, like the array of test points often found on the back of the card, are manufacturer-specific.)

• by jl6 on 6/2/25, 6:44 AM

  Great to see someone putting in the effort to do this. It’s really difficult for consumers to evaluate the quality of “cheap” electronics that don’t get in-depth attention from major review publications, which themselves are dwindling. RIP Anandtech. Usually the best we get is advertorial listicles regurgitating the specs, with some affiliate links.

• by kevincox on 6/2/25, 8:11 AM

  I'd love to hear more about their 0.1% error rate threshold. There is a huge difference between 1. Read returns an error but succeeds on the next attempt 2. Read returns an error and continues to do so 3. Read returns the wrong data "successfully".

• by jauntywundrkind on 6/2/25, 3:20 PM

  These cards are doing kind of awesome? For endurance:size of consumer drives, 600 drive cycles (Terabytes written (tbw)/terabytes capacity) is pretty much the best advertised endurance you're gonna get, is very standard. Definitely a gamble but seeing some of these SD cards spit out 4000 cycles is incredibly impressive in that light.

  In practice consumer NVMe/sata drives do tend to last longer. The ever excellent TechReport did a test to failure 10 years ago, and all drives made it to 3000 cycles. The Samsung 840 Pro almost made it to 10k cycles! https://techreport.com/review/the-ssd-endurance-experiment-t...

  I'd love to see some modern drives tested. It would be really interesting to know who makes flash that in practice is the highest endurance: kioxia, micron, whose flash lasts longer? TechReport has an excellent SSD database that shows all the parts & specs on those parts, so one can go look at a SanDisk/WD SN7100 and see oh it uses Kioxia 218-Layer BiCS8 3D TLC; there's not really that many people making flash itself. It's be lovely to have some exhaustion/test to failure for these consumer stacks. And what stacks they are! 218 layers! I can't imagine endurance has gone up over the decade, especially given how strongly drives 10 years ago overperformed!! https://www.techpowerup.com/ssd-specs/western-digital-sn7100...

• by jorvi on 6/2/25, 9:42 AM

  List of results with some slight inferences where needed:

  On median, SD cards will experience:

  - first error at 1450 r/w cycles

  - failure at 3100 r/w cycles (total failure or 0.1% of sectors)

  SD card endurance:

    - 5*    Big name brand 'Endurance' / 'Industrial' series
    - 2*    Adata
    - 4*    Amazon Basics
    - 4*    Delkin
    - 1*    Gigastone
    - 5*    Kingston
    - 4*    Kioxia (Plus and G2 only)
    - 4*    Lexar
    - 1*    onn.
    - 4*    OV (terrible read/write performance)
    - 4*    PNY
    - 1*    Sandisk (after WD acquisition)
    - 1*    Silicon Power
    - 4*    Transcend

• by PaulKeeble on 6/2/25, 7:13 AM

  This is the sort of testing you would hope that someone like Storagereviews would ultimately do but hasn't. They have all these SD cards and have tested them in various performance scenarios, would be nice for them to also test the endurance of the cards as well.

  At ~3000 total drive writes that is quite a lot of endurance for a raspberry pi, especially for a 128GB drive where the OS and apps only take up ~20-30GB. That is a lot of OS updates and database writes.

  Still glad someone is doing this. I just went with the endurance cards with A2 ratings (Samsung in this case) but none of my Sandisks failed in my Raspberry pi's over the years they all still work but I haven't hammered them. More recently I moved to an Orange PI 5 plus which has an SSD now and that ideally is what they move to for performance and durability reasons.

• by nirui on 6/2/25, 1:22 PM

  > SanDisk/WD... noticed that they were sensitive to brownouts

  Damn, I had a hit on this but this test made it clear.

  I have few SBCs with SanDisk TF cards as their data storage. Those were powered with a cheap USB power adapter, and as a result, these SBCs completely locks themselves up after just few days and become inoperable. I even contacted the manufacturer of the SBCs for repair.

  But before actually sending the boards out, I brought a new power adapter, replaced the old one with it, and the problem is completely gone. The SBCs are all good, the problem is the power supply.

  Though, I still hope that those SBC manufacturers could switch to other storage device. SSDs, even the basic ones are much more reliable that whatever TF card listed on the test.

• by imtringued on 6/2/25, 7:53 AM

  SD and microSD cards are designed to be used in battery powered devices. SBCs are not battery powered. If anything, they are microSD card torture devices.

  Basically 99% of the time, card corruption has nothing to do with the microSD card in question and more to do with the quality of your power supply.

• by bArray on 6/2/25, 1:42 PM

  I generally treat SD cards as memory that will almost certainly fail at some point. Once you accept that, they are more easy to deal with. If you want reliable memory, get somewhere external to store it.

  I had a HGST spinning disk in my laptop from prior to the WD acquisition, and it failed perfectly. It dealt with errors all of the time and never lost data, but most importantly was that when it finally kicked the bucket, it became read only and I was able to recover all data without issue.

• by retrodaredevil on 6/2/25, 1:15 PM

  I've started to exclusively install DietPi on all my Raspberry Pi devices. It has great defaults including RAM log. Additionally, it's not limited to Raspberry Pis, it can be installed on a range of SBCs.

• by geor9e on 6/3/25, 12:19 AM

  I will never use SDCards to run any 24/7 operating system again, after my Home Assistant for Rasperry Pi became unusable twice. The hours of troubleshooting and getting it back online was not worth the $200 an NUC/mini with actual SSD cost. SD cards are incredibly slow too.

• by addandsubtract on 6/2/25, 10:05 AM

  I switched away from RaspberryPi's because dealing with their SD cards and the pain of setting up an alternative boot drive was too much for me. I'm happy to use an ESP32 for smaller projects or an N100 mini PC for equal or more demanding projects.

• by fanatic2pope on 6/2/25, 12:42 PM

  When I used to use rpi's I had a series of steps I would take to reduce/eliminate most writes, while not turning the system full read-only. Simple things like disabling all unused services, moving tmp and syslog to a ram disk, stuff like that. Combined with a good quality sd card and power supply, none of my raspberry pi's had an sdcard failure in years. I've since moved on to more properly "embedded" type devices (ESP32 and Pico) but not because of sdcard issues.

• by amelius on 6/2/25, 6:47 AM

  My experience: if you forget to unmount a microSD card after writing to it, then at least you can still use it. In many cases this cannot be said for USB thumbdrives.

• by LargoLasskhyfv on 6/2/25, 12:42 PM

  Never had trouble with normal or even 'bargain basement' SanDisks when on special sales.

  Just don't swap on them, don't write too much on them either, learn how to log to RAM, compress these, and write only in long invervals(like 1x day), configure ZRAM, and so on. Some still working after more than 10 years now.

  Maybe depends on the boards too, mine are various Allwinner with GAN-powerplugs, sometimes as expensive as the boards themselves. LOL.

• by FirmwareBurner on 6/2/25, 6:44 AM

  I don't see them covering the "high endurance" or "industrial" models of SD cards, those are usually for 24/7 operation like security cams or industrial usage and tend to be way more reliable, especially the dedicated industrial ones which aren't readily available on Amazon because they cost way way more than your typical consumer is willing to spend for something they've been conditioned must be dirt cheap.

  Unfortunately, such posts tend not to be very relevant long term, as non-reputable brands can always switch the flash/controller dies underneath the package to save costs, while maintaining the same packaging design and SKU, so as a consumer you can always be duped into buying something with good reviews which is now a totally different product that performs worse. And that's not counting the fakes.