It would seem that the choice of a card reader obviously affects the speed of memory cards: models with a USB 3.x interface are always faster than their ancestors with USB 2.0, but are all USB 3.x card readers equally fast? Does USB 3.2 Gen 2 (10 Gbps) make sense for them, or is it overkill and marketing noise?
I wanted to check this on the example of the microSDXC Transcend 340S 256 GB (TS256GUSD340S) with three different card readers. For the purity of the experiment, I chose the readers of the same company (fortunately, they are often purchased). Below are the test results, but first, I will describe the main conditions for their conduct.
Motherboard: Asus Maximus VIII Hero (old, but still good);
Stone: Core i7-7700K at 4500 MHz (45x100x4 + HT);
RAM: 2 x 8 GB Kingston HyperX DDR4-3466 in dual-channel mode;
Two solid-state devices of half a terabyte: WD Black SN750 (for the system) + TS512GSSD452K (storage);
B / p: SSR-750TR (aka Seasonic Prime TX-750).
Card Readers: Transcend HUB5C, RDF9K2 and RDF5
Operating System: 64-bit Win 7 with the latest updates (ESU).
The card itself is interesting in that, according to the manufacturer, it is capable of “providing speeds that go beyond the UHS-I standard”, i.e. above 104 MB / s. How exactly this is achieved – the devil only knows, but I will assume that this is data compression on the fly, performed by a card reader. The following statement on the official website led me to this idea: “Best performance is achieved with Transcend HUB5C card reader and current firmware.”…
First, let’s test the recommended HUB5C by plugging it into a USB 3.1 port. Gen.2 Type C (10 Gbps) and rolling the latest firmware v.2958.
I have to admit: this test utility really shows sequential read and write speeds higher than those provided by the UHS-I standard (104 MB / s), but the performance leaves much to be desired for random operations.
With a different profile, read / write is also very fast, but small-block operations at random addresses do not correspond to the declared class A2 (from 4000 IOPS for reading and from 2000 IOPS for writing in random mode).
I also tried the old AS SSD Benchmark utility, but its results were different and very strange each time, so I decided not to give them.
Then I ran the same tests with the RDF9K2 card reader (firmware 1146, no newer).
I think the result speaks for itself. The card reader clearly ran into the limit of the UHS-I interface, but its firmware turned out to be better optimized for read / write operations at random addresses.
As you can see, on a relatively “slow” interface (5 vs 10 Gb / s), the memory card demonstrated an even higher speed of operations with random block addresses.
The Victoria utility also confirms that the linear read speed has decreased: the minimum is 53 MB / s, the average is 73 MB / s and the maximum is 94 MB / s.
First USB 3.x generation
It remains to check my first USB 3.0 card reader – Transcend RDF5.
Despite its venerable age (the model appeared about ten years ago), it still reads fast modern UHS-I cards. However, at the same time, RDF5 does not have any optimizations at the firmware level – neither for small-block nor for linear operations.
The decrease in read / write speed at random addresses is especially noticeable – almost twice, but in the mode without preliminary creation of a command queue (Q1) and in one thread (T1) they are about the same as in the new HUB5C.
Taking into account the five times lower price and diminutive dimensions, RDF5 still looks like a reasonable option, although it is not necessary to count on records with it.
The tested memory card really overcomes the UHS-I limitations when used with a modern card reader. However, even its peak speeds are well within the bandwidth limit of USB 3.x gen 1 (5 Gbps), not to mention gen 2 (10 Gbps). Therefore, a faster card reader port by itself will not provide advantages, as it was during the transition from USB 2.0 to 3.0.
It should be borne in mind that in practice a faster external interface may mean the use of a modern controller, with which new cards will work faster.
Also, the speed of cards is affected by the firmware of the card reader: some versions are optimized to achieve the maximum speed of linear read / write, while others are optimized for small-block operations at random addresses.
The microSDXC 340S memory card itself caused mixed emotions. If in linear modes it is always nimble (close to the V90 class, although it is certified as VЗ0) and even goes beyond UHS-I, then I do not observe the declared performance class of A2 applications with any card reader.
PS: In general, the whole situation with the declared characteristics of memory cards reminds me of an old anecdote about an elephant in a zoo. The visitor reads an ad next to the aviary: “An elephant can eat two buckets of roots, three buckets of leaves, four buckets of grass in a day …” and asks the caretaker in surprise:
– He really will eat so much ?!
– Maybe he will, but who will give him?