williamwright wrote:
Nextbase want £63. Other makes are much cheaper. The most expensive I've
found is Sandisk Extreme Pro 128GB microSDXC Memory Card at £27. Which
shall I buy? Anyone with concrete evidence that the cheaper one works OK?

Bill

The Sandisk Extreme stuff is usually OK. Their stuff
with Ultra branding is "meh".

The device is probably TLC, and since your application
is continuous write, I wouldn't expect to be "burning a hole"
in it, with a dash cam. Even if it doesn't have wear leveling,
the write pattern takes care of that.

If you want to benchmark it on arrival, on a computer,
you'd need a USB3 to SDXC adapter, and a decent USB3 port
on the computer. And "don't do stupid stuff" to test either.
Copying a folder with 10,000 4KB files to the USB3 adapter
with the SDXC plugged into it, is a poor test. The intended
application (dashcam), does not have that characteristic,
and a higher consistent rate should happen with the dashcam.
The time to transfer a single large file would be more
representative of the application.

I'd point you at a benchmark result, if I could find a
competent one. I could only find one comment from someone,
that the product did not measure up, but there was no evidence
of what he did wrong. Sandisk has one other product, a UHS-II
version, that manages around 100MB/sec on write, but that
doesn't really tell us anything about the UHS-I version.
Anandtech has a graph for the UHS-II version.

You would think benchmarking is easy, but as it happens,
it's not. The first mistake people make with TLC, is
immediately doing a read benchmark. TLC is "mushy" when
you take it out of the package, and every sector needs
error correction. This slows down the stick and makes it
look bad. To bench on read, you need to *write* the device
first. Then the sectors are fresh as a daisy, and ready for
a read bench. Consequently, your first task then is the
write bench, and the read bench comes second. I gave
a Corsair Neutron poor marks, out of the box, until it
occurred to me it was the "mushy TLC" problem. And
writing a section of the device, and comparing uninitialized
to initialized sections, made a world of difference. We can
thank the geniuses who decided TLC was ready for consumers,
for this nonsense. MLC flash based devices, no special
precautions are needed. But nobody uses MLC for stuff
like this. We're stuck with TLC (3 bits per cell) or
QLC (4 bits per cell).

Paul