On July 27, 2023, I ordered my first batch of microSD cards from AliExpress: three Kioxia Exceria G2 32GB’s, a Kodak Ultra Performance 32GB (the gold version), three Hiksemi NEO 8GB’s, two Hiksemi NEO 32GB’s, three Microdrive 16GB’s (yes, the ones with the Bart Simpson design on them), three SanDisk Ultra 32GB’s, a “Sansumg” PRO Plus 2TB, a “Sansumg” EVO Plus 2TB, three Cloudisk 32GB’s, and a Kodak Ultra Performance 64GB (the black/gold version). Altogether, I spent $86.78.
At about the same time, I took an old laptop, installed Ubuntu on it, and started writing a new card testing program: something that would test for fake flash and determine the card’s true size, run benchmarks on the card’s read/write performance, and then put the card through endurance testing until it failed completely.
A few weeks later, the orders started to show up.
On August 16th, I plugged a couple of card readers into that laptop. I picked out a couple of cards from my initial batch — one of the Microdrive 16GB’s and the “Sansumg” PRO Plus — and set my program to work, writing to them and reading back from them non-stop. Over the following months, all of them made their way into my card readers, and I started them on the same path of endurance testing.
Out of that humble setup has emerged something even more ambitious: 11 machines, 83 card readers, and 121 cards, running around the clock, stress testing these cards with one goal in mind: see which ones can last the longest.
The “Sansumg” 2TB PRO Plus (which was actually closer to 8GB in actual capacity) was the first to fail. It would be followed — nearly two months later — by one of the Bekit 8GB’s, then by one of the Hiksemi NEO 8GB’s just a few days later. In total, I’ve sent 115 cards to their graves — after having written over 14 petabytes of information to them. But out of that initial batch of 21 cards, there’s still a few survivors: all of the Kioxia Exceria G2 64GB’s, one of the Hiksemi NEO 8GB’s, and one of the Microdrive 16GB’s. Somehow — defying all logic — the one Hiksemi NEO 8GB still hasn’t experienced a single error, even after completing over 100,000 read/write cycles. (One of the Kioxia Exceria G2 64GB’s hasn’t experienced any errors either — but it has only completed 1/8 as many read/write cycles as the Hiksemi NEO 8GB has.)
There’s a story to be told from the ones that have failed. However, there’s a story to be told from the ones that haven’t failed as well. It’s not a complete story, but I believe that some story is better than no story at all. So on the two-year anniversary of the start of this project — when I started testing those first two cards — let’s tell that story.
Continue reading “The Great MicroSD Card Survey: Two Years Later”