SD speed is not that bad at the end, or it is? And the speed is in the card, not the card reader, right? So, if someone buy a UFS card, the read speed will be of that card, right?
Ooh, gonna ramble a bit.
(but to answer your question, this sort of thing is usually both sides? Reader needs to be able to handle X version of protocol, the card itself needs to be so and so capable)
See the charts of benchmarks for both the internal NVMe SSD and the SD Card at 1:29 or 1:30? Here's my interpretation of what's going on...
SEQ = Sequential; you're working with one file and/or everything you want to work with is lined up for the best possible conditions.
Not surprisingly, the NVMe drive is crushing here.
Q should be Queue Depth, or the number of requests working simultaneously/in parallel. The NVMe drive shows some difference between Q1 and Q8, but doesn't seem to be hitting its theoretical max. Could be a CPU limit (I'm guessing power limit, as a desktop Zen 2 shouldn't have the same issue). The SD card on the other hand shows practically no difference between Q1 and Q8. It's only working one thing at a time?
RND = Random; I work with one file, finish, then jump to another file located elsewhere, and so forth.
Notice that at Queue Depth 1, the NVMe drive is only a few times faster than the SD card. Here's a question: how much of that difference is coming from the 'working with a given file' part and how much is coming from the 'I then jump around to the next file' part?
I'm guessing more from the former than the latter, since at the end of the day, they're both using NAND flash. The medium itself can only be so responsive.
At Queue Depth 32, the NVMe drive pulls away again. So, jumping around randomly from a file to another
one at a time isn't so hot. So the workaround is... why only work one at a time, when you can do a lot at once? As oldpuck mentioned before, cover up latency by employing parallelism. The SD card again seems to be stuck to just one thing at a time?
So to recap:
When does an NVMe drive deliver on the hype (relative to something like an SD card)?
1. You are working with
large, individual files. Here's something that the companies marketing these drives don't want to admit:
this currently does not apply to the average consumer*. Granted, 'currently' does a heavy lifting here, because DirectStorage
ought to be changing this in the future.
2. You can make lots of requests simultaneously/in parallel. Good idea! Also shouldn't be unique! SATA SSDs should also be to do this, as we know that in real life, there isn't much practical difference between NVMe and SATA for the average consumer. I assume, but don't know for sure, that UFS is also capable of this.
*corollary to this: the marketing push for PCIe gen 5 NVMe drives started this year. Guess what? Realistically, there's practically no meaningful improvement for the average consumer over existing gen 3 or 4 drives. Don't fall for the hype of ever increasing sequential speeds unless you know it'll do something for you.