With SSDs, every access has to go through an flash translation layer, or FTL. So whenever the computer wants to read data from the drive, it must first lookup its physical address in the FTL, and then it will get the data from the actual NAND chips internally.
When it comes to writes, this process also occurs, where the SSD has to find a the appropriate place to put the data. However, when write caching is on, as soon as the SSD decides where to put the data, it will tell the computer "hey I'm done writing." The data technically hasn't made it to the NAND chips at this point, but the computer thinks the operation is complete.
Now the stated write IOPS number is "best case." This implies write caching is ON, and the drive is in a clean state, where absolutely no writes have been done. You can even see it in the datasheet (look at the disclaimer #2 on the first page: https://www-ssl.intel.com/content/dam/www/public/us/en/documents/product-specifications/ssd-530-sata-specification.pdf) Also, on the latter pages of the datasheet, the incompressible 4k random write IOPS is between 17K - 23K. This implies the SSD controller is using compression (SandForce).
Lastly, you must also consider the capacity of the SSD. The larger SSDs tend to be faster because there are more NAND channels operating in _parallel_.
I invite you to study what happens to SSDs as they begin to get aged. You'll notice the write IOPS can go down substantially due to garbage collection. Here is a link to a benchmark: Samsung SSD 840 Pro Review | StorageReview.com - Storage Reviews The graph of interest is the "preconditioning curve."