Tech Hobbyist Destroys 51 MicroSD Cards To Build Ultimate Performance Database (tomshardware.com) 17
Tech enthusiast Matt Cole has created a comprehensive MicroSD card testing database, writing over 18 petabytes of data across nearly 200 cards since July 2023. Cole's "Great MicroSD Card Survey" uses eight machines running 70 card readers around the clock, writing 101 terabytes daily to test authenticity, performance, and endurance.
The 15,000-word report covering over 200 different cards reveals significant quality disparities. Name-brand cards purchased from Amazon performed markedly better than identical models from AliExpress, while cards with "fake flash" -- inflated capacity ratings -- performed significantly worse than authentic storage. Sandisk and Kingston cards averaged 4,634 and 3,555 read/write cycles before first error, respectively, while Lenovo cards averaged just 291 cycles. Some off-brand cards failed after only 27 cycles. Cole tested 51 cards to complete destruction during the endurance testing phase.
The 15,000-word report covering over 200 different cards reveals significant quality disparities. Name-brand cards purchased from Amazon performed markedly better than identical models from AliExpress, while cards with "fake flash" -- inflated capacity ratings -- performed significantly worse than authentic storage. Sandisk and Kingston cards averaged 4,634 and 3,555 read/write cycles before first error, respectively, while Lenovo cards averaged just 291 cycles. Some off-brand cards failed after only 27 cycles. Cole tested 51 cards to complete destruction during the endurance testing phase.
yort (Score:3, Insightful)
The kind of destruction to get behind (Score:5, Insightful)
Re:The kind of destruction to get behind (Score:5, Interesting)
This is the kind of destruction we love to see. Destruction to determine product quality and help decide what to purchase. Not dumb destruction of brand new quality products in order to generate dumb clicks and dumb comments.
Looking forward to when the high endurance cards finally reach the 1% failure state so that we can find out whether these things really are better than the standard cards. So far, the first failure of SanDisk High Endurance was *way* earlier than SanDisk Pro, on average, so I won't be surprised if it turns out that the whole high endurance thing is a scam.
Re: (Score:3)
Re: (Score:3)
I wouldn't say scam, but, its a marketing trick for "wear leveling" that standard microSD doesn't do. But just because you wear level doesn't make it more durable. Durability needs more NAND layers and more NAND chips and striping to truly shine, and microSD simply can't fit all that.
My point is that the microSD Pro cards exhibited their first error after half again more cycles, on average, than the "high endurance" cards, suggesting that the high endurance cards might actually have lower endurance than Pro cards, which is... unexpected. The Pro cards, IIRC, do more wear leveling than the standard cards. I have no idea whether high endurance cards do more wear leveling, have a lower number of levels per cell, have more spare cells, or something else entirely.
Re:The kind of destruction to get behind (Score:4, Interesting)
But just because you wear level doesn't make it more durable.
Actually for common workloads that is *exactly* what it means since durability is not considered on a block level but on a device throwing an error level. The vast majority of workloads don't fill the cards to capacity meaning wear levelling objectively increases the life and write durability to the device. The question is how much. If every load you have fills the entire memory card to capacity every time then wear levelling does nothing. If on the other hand you have a real workload, like say a camera then half filling the card every time will give you double the expected life (if the wear levelling is perfect, it's not but pretend it is for easy math).
But in any case you're ignoring the fact that these cards also do in fact often have significant differences in the NAND chips and their construction. E.g. Sandisk Endurance SD cards use 3D NAND, whereas their normal ones use standard TLC NAND. Also the Endurance ones use chips with a higher temperature rating, different plastic casing (higher impact resistance) and are actually rated as water proof for complete submersion vs simply trying your luck.
Useful for RPi owners, read wear tests need too (Score:3)
A dirty little secret of SD cards is they can fail from continuous reads too, such as when used in a looping media player. Reading cells cause some leakage to adjacent cells and the cards normally rewrite cells after many reads. However that happens in the background when the card is idle, so if you read without ever letting the card idle long enough they will also fail, even with no writes.
Ok, but... (Score:2)
Ok, great, someone has done the research to know which reliable cards to buy. However, how does one avoid counterfeit cards masquerading as these cards? For example, the Kingston Canvas Go! Plus 64GB card scored the best. When I go to Kingston's site to purchase one ( https://www.kingston.com/en/me... [kingston.com] ) I can't. I'm only given a link where to purchase from. I would feel much better about my odds if I could purchase directly from Kingston. I refuse to order from Amazon because I have been burnt by counterfei
Re: (Score:2)
Ok, great, someone has done the research to know which reliable cards to buy. However, how does one avoid counterfeit cards masquerading as these cards? For example, the Kingston Canvas Go! Plus 64GB card scored the best. When I go to Kingston's site to purchase one ( https://www.kingston.com/en/me... [kingston.com] ) I can't. I'm only given a link where to purchase from. I would feel much better about my odds if I could purchase directly from Kingston. I refuse to order from Amazon because I have been burnt by counterfeit goods (too many times!) too often. If Amazon has this problem I'm sure other outlets have the issue too.
This is a real concern. If brand name and model and external packaging can't be trusted, then the only way to truly determine endurance is to write enough to destroy the card, which is, of course, not practical because the testing has to be done for each physical card.
Re: (Score:3)
Or remove the partition and reformat. In 2019, just before the pandemic, I was scheduled to give a few beginner adult classes at the local community college in the use of the Raspberry Pi. The students were required to have their own Pi but to make things easier I decided to purchase 60 used 32GB MicroSD cards on eBay that were pulled from smart phones. I figured I would prepare them with the software needed for the class. Of the 60 cards 13 of them were counterfeit. When I would write to them the data woul
Re: (Score:2)
Re: (Score:3)
I refuse to order from Amazon because I have been burnt by counterfeit goods (too many times!) too often. If Amazon has this problem I'm sure other outlets have the issue too.
No the problem is Amazon. Amazon is not a reseller, Amazon is a flea market. Amazon does not care if someone creates a fake Kingston shop within their marketplace and sells them next to the official shop without regular people being able to tell them apart.
You need to buy electronic components from official resellers. I would trust those in the "industrial" category. If I select "USA" https://www.kingston.com/en/wh... [kingston.com] then in between others, Mouser / Digikey appear and these one are of trust. The only probl
Re: (Score:2)
I think Best Buy is still around? They probably even do delivery now if that's your obsession.
As far as online retailers, I used to be partial to NewEgg but somewhere around 10 years ago they went way downhill.
Re: (Score:2)
I never purchased SD cards on Amazon. The last counterfeit Item I purchased there (and the last straw) was a set of Mitutoyo calipers that were rather expensive but the company has an awesome reputation for quality measuring tools. I was having problems getting the batteries to make proper contact in the battery holder. When I reached out to Mitutoyo and explained the issue they asked which brand and size of coin cell batteries I purchased. When I told them I used the included batteries they stated they don
And data retention aka bit-rot? (Score:2)
That's another problem with flash - all my picture frames show corrupted data after few months.
Now that's super useful (Score:2)
The tests I care about other than reliability and straight streaming read or write speed are random read/write which some cards do much better than others — I expected this for writes, but not for reads! — and how long a device can sit around unused and retain data.
TFA notes that there are application performance class ratings, but they are very particular and what I really want to know is what happens with lots of small files.