As an example, my Adata XPG S8200 Pro can reach read and write speeds of up to 3500/3000 mb/s and even in testing (using CrystalDiskMark) the supposed speeds are around that range. But when I actually transfer files, even within the SSD, the actual speeds is so much slower?
In: Technology
First of all, are you *moving* the data or *copying* the data?
**Moving** the data on the same drive does not actually *move* the information on the physical drive. The OS merely changes the pointer to that data to make it appear in a different location within your filesystem.
In other words, the data will still be stored at index XYZ on the drive, but it’s folder (the pointer) is now “D:/Folder 2” instead of “D:/Folder 1”
**Copying** data is a much more intensive process. When you do this between 2 different drives, each drive only needs to perform a single operation (i.e. a read operation on the origin drive, and a write operation on the destination drive.) This means that each drive can devote all its available resources to that single operation.
However, when you copy data on *the same drive*, that drive now needs to perform *two* operations – reading *and* writing. On a single SSD, the data must be read, written to a buffer, erased and rewritten in steps, all using the same controller and resources on that SSD. This can be *much* slower than if copying between two different drives.
I can’t tell by looking at CrystalDiskMark, but it may be generating test data at runtime to write to the destination, instead of reading test data from one disk and writing to another. This actually seems very likely the case, because it wants to benchmark one device, your SSD, as isolated as possible. If it involved the read speed of some source drive, then the test can only run as fast as the slowest device.
So consider that, your one drive may be slower than your other drive. It may read less at a time, it may be on a smaller or slower bus – so less data can be transferred across the system, it may have to seek to find the data – especially on a slow, physically rotating magnetic platter with a read/write head mounted on an armature, the data on the source may be fragmented, and worse overall, both drives may share the same bus! If they’re both on the same wire, then when one drive is active, the other is idle – data has to be read to a buffer in system memory, and then written back down the same bus to the other device.
There are many factors that can effect performance. Benchmarks in isolation are poor, and consumer benchmarking software typically sucks across the board. I know benchmarking video cards and gaming rigs are a big driver in selling hardware, but often it doesn’t translate well to performance during actual use. I’m a former game developer, performance profiling is my job, and it’s damn hard to get right and get meaningful data and analysis.
Also, CrystalDiskMark measures tranfer rates as MB/s, base-10, whereas Windows Explorer uses MiB/s, base-2, but calls it MB/s for historical reasons. The same transfer will have different values because they’re different units in different bases. Base-10 makes for bigger numbers and looks more impressive, and you see it a lot with drive manufacturers because the bigger numbers drive sales. And only they use base-10, for obvious reasons!
Transferring data on the same SSD is not a reliable measure, for a couple of reasons. The first is that depending on how the SSD is connected (for instance if it’s SATA) you can’t do reading and writing at the same time, because at times it reads and at times it writes.
The second is that unless you have partitioned your SSD into multiple partitions, moving data is not really moving the data it is altering the meta data only which can involve lots of random seeks and writes to various places and would be get you closer to your drives random IO performance than sequential IO performance.
one of the reasons may be that in order to transfer a file, it’s necessary to locate all of its pieces on the drive, then find a new place for them on the drive – probably, not one monolithic area, but much smaller multiple areas.
however that was much more important for old HDDs, SSDs are better in that regard.
Tested speeds are from one medium to another, i.e. from one SSD to another. Just like HDDs, SSDs suffer from sequential write. Consider a table. You have a pencil. You want to draw (write) a picture (file) on the left side of the table and then copy that to the right side. You move your hand constantly from the left side of the table to the right, slowing down the copying process considerably if you already have the picture on a piece of paper in front of you and are copying it on only the left side of the table (copying from one medium to another). See how a HDD functions inside and you will understand why the speeds are slower when using the same medium (copying from one area/folder to the HDD/SSD to another).
Hope this helps
Latest Answers