Real-World Write Testing
We try to hammer this home any time we talk about comparing SSDs: upgrading to any modern solid-state drive is going to give you a better experience than the best conventional disk, regardless of the SSD's position relative to its competition. From there, sure, the benchmarks tell us that certain drives are faster than others. But it gets more difficult to tell them apart.
Nevertheless, we wanted to add real-world write testing to help demonstrate the performance delta between SSDs, even at the same capacity points and using similar controller technology.
In our first test, we clone our system drive (a 240 GB Vertex 3) over to each target drive using Easus Todo Backup Workstation 5.3. The source data comes from my personal laptop, so hopefully it's representative of what many of you have on your machines, too. Aside from the OS, I have hundreds of song files, video content, a handful of games, Photoshop, Office 2010, and a bunch of personal documents. Overall, there's a fairly balanced mix of compressible and incompressible data.
Corsair's Neutron GTX and Neutron take the lead in our first test, unaffected by all of the incompressible data that slows down the SandForce-based drives. But the Neutrons aren't in front by much. Intel's SSD 330 is the "worst" performer, finishing the workload in 11 minutes and 35 seconds. The Neutrons wrap up in 10 minutes and 40 seconds. We ran this test multiple times, and we see variations of up to 10 seconds. So, the differentiation from one drive to another isn't as significant as synthetic metrics might indicate.
This is even more true in our second test, where we copy the 2.86 GB H.264-encoded file that we use for tablet/notebook battery life testing. Even though the source data is not compressible at all, really, the SandForce-based drives end up on top. "What?" you ask. "That can't be right." But we're only moving one file, resulting in a very low queue depth. Throughput is lower in a test like this one, and that applies across the board. Regardless, though, the margin of error for this test is 0.3 seconds, so what we see is a practical tie between 13 different drives.