1 hour ago, TerryE said:

 

I've just been doing a bit of research and it turns out that this isn't the case.  Reputable manufacturers do have an on-card flash controller in front of the flash chip(s) ( and this is often an ARM chip! ) and this controller does implement wear leveling.  See Bunnie:studios - 2Gbyte microSD card and Bunnie:studios - On MicroSD Problems which describes an epidemic issue of knock-off clones where the counterfeiters save cost by leaving out this controller and use a very simple interface which does no wear leveling.    If you sudo cat /sys/block/mmcblk0/device/cid this will output a 32 hex digit output.  Use https://goughlui.com/static/multicid.htm or layout below to decode.  It turns out one of my SD cards is a bit suspicious, and has a questionable life, but luckily I only use them as /boot and effectively never write to them, so this isn't an issue for me.

 

 

 

Thanks Terry very interesting. This may explain why I have had a lot better results with SD cards since I moved to using SANDISK only, and only ever bought from reputable sources.

 

The microSD standard does not support TRIM; only SSD -- but the Linux kernel only supports TRIM over eSATA and USB3 interfaced controllers.  (eMMC does which is why the new RPi clones which support eMMC and USB3 have a big I/O performance hike over RPi.)

 

I believe that we have the same potential issue with cheap SSD clones, which is why provenance of SSD and microSD devices is of material importance.

I've been half considering getting a RPi CM3 (The only version which does include eMMC) but I'm not sure my eyes are up to the level of soldering required.