You missed the part where I said I used dd, binary compare and md5sum to test. IE each and every block of the SDcard has had some random contents copied onto it - then is verified in read. This is all the windows test software does.1: Where were the problem SD Cards purchased? (Asking simply due to the high instances of false-capacity counterfeits on the likes of eBay, dhgate, and so on, over the last decade or so. Further info is available here, for anyone interested.)
2: Have the SD Cards been tested as genuine using a dedicated tool for this specific purpose? Specifically, F3 on Linux/Mac OS X, or H2TestW on Microsoft Windows.
From a command line of a linux PC with a USB card reader (device /dev/sdb on my PC) it looks something like this:
Produce 4GB (decimal) of random data
# dd if=/dev/random of=randomdata bs=4000M count=1
Copy data onto card
# dd if=randomdata of=/dev/sdb
<wait for completion plus a long pause for cache to clean, then unplug and re-plug usb device)
# md5sum /dev/sdb
# md5sum randomdata
If the two results match then you can be pretty sure that 1) every sector copied ok 2) the card is not a fake (remapped) as this would drastically change the file hash. This is one of several tests I did but is the most informative.
PS this isnt exactly what I typed but close enough to show the idea, I wrote and read the actual number of blocks the device reported instead ....