Benchmark Results: Iometer
We used an 8 GiB LBA range for each of our Iometer benchmarks, running each test for 90 seconds. We also aligned the read and write access patterns to a 4 KiB sector boundary.
This measurement is useful for looking at the read and write performance of common transfer sizes at queue depths ranging from one to 32. Queue depths in a typical client environment are generally quite low, so the results generated between one and four are of particular interest. Read operations are typically a lot more prevalent than writes operations, so read performance is also most relevant.
Plextor's M5 Pro dominates read performance across all transfers sizes and queue depths, while the Vertex 4 comes out on top in write performance, peaking high and early.
In order to compare our findings to what each manufacturer says its drive can do, we have to convert the MiB/s results to IOPS. Notably, the "up to" results that vendors like to use are typically based on a queue depth of 32, and it's possible to see significant variation in Iometer results depending on the span of the test file and prior write history.
As expected, there is some variation between our benchmark results and those specified by the manufacturers. This comes down to differences in how each company determines its performance specifications, the state of the drives, and the test system used. We're happy to see that Plextor's M5 Pro, though rated for the highest I/O performance, also finishes first in what we measure.
| Header Cell - Column 0 | Vendor-Rated "Up to" 4 KiB Read IOPS | Observed 4 KiB Read QD 32 IOPS | Vendor-Rated "Up to" 4 KiB Write IOPS | Observed 4 KiB Write QD 32 IOPS |
|---|---|---|---|---|
| Samsung 830 | 80 000 | 76 309 | 30 000 | 39 657 |
| Crucial M4 | 50 000 | 46 824 | 50 000 | 59 175 |
| Vertex 4 | 90 000 | 85 159 | 85 000 | 82 335 |
| Plextor M5S | 73 000 | 73 888 | 70 000 | 69 346 |
| Plextor M5 Pro | 94 000 | 90 883 | 86 000 | 84 081 |
But those numbers are at a queue depth of 32. Let's instead have a look at 4 KiB read and write performance at a queue depth one, which is where you're going to see more activity. We again converted the MiB/s results to IOPS in our chart below.
Plextor's M5 Pro comes out on top again for read performance, but trails the Vertex 4 on write performance (though not by much).
| Header Cell - Column 0 | Iometer, 4 KiB Read QD 1 IOPS | Iometer, 4 KiB Write QD 1 IOPS |
|---|---|---|
| Samsung 830 | 5 721 | 15 849 |
| Crucial m4 | 5 849 | 15 900 |
| Vertex 4 | 7 036 | 16995 |
| Plextor M5S | 7 076 | 15 823 |
| Plextor M5 Pro | 7 729 | 16 688 |
