Benchmark Results: 4 KB And 512 KB Random Reads
Now we have a better idea of why the SSD 510 performed so poorly in our specific Iometer workloads. Random 4 KB reads are low—though right at the 20 0000 IOPS mark we were told to expect. In comparison, the recently-handicapped Vertex 2, with its 25 nm NAND and ECC overhead, turns in more impressive numbers.
At lower queue depths the Intel drive isn’t comparatively in as bad of shape. However, it still finishes last, even with a queue depth of one.
Our results in Iometer map to CrystalDiskMark, where, with a queue depth of one, the SSD 510 falls in behind the X25-M, yielding a last-place finish.
Bumping queue depth all the way to 32 helps push the SSD 510 up above the Vertex 2 (severely impacted by processing overhead when it has to work with small files). However, everything else continues to outperform.
Larger files give Intel’s drive the room to benefit from its 6 Gb/s interface, and we see the SSD 510 almost exactly match the performance of Crucial’s RealSSD C300, based on a previous-generation Marvell controller.
The Vertex 3s are still much faster though, while the 3 Gb/s drives lag behind.