Benchmark Results: 4 KB And 512 KB Random Reads
At low queue depths and small random reads, we see the Vertex 3 Pro suffer in IOPS.
The header of these two charts is the same, except for one important detail. The first employs a queue depth of one, and the second uses a queue depth of 32. Why are these figures important? Well, look what they do to performance.
Native Command Queuing, part of the SATA specification, was originally designed to improve the performance of mechanical disks by allowing the hard drive itself to optimize the order in which read and write commands get executed. Of course, the physics of a hard drive don’t apply to SSDs. However, the multi-channel architecture of a solid state drive enables it to similarly field multiple concurrent requests—though Intel claims the bottleneck isn’t the drive, but rather the host system itself. Today’s SSDs are consequently dependent on high queue depths in order to realize the specifications you see manufacturers quoting.
With a QD of one, the new Vertex improves on what we've already seen from SandForce, augmenting performance in situations unfavorable to the way modern SSDs run best: with lots of concurrent requests. Cranking queue depth up to the other (very unrealistic on the desktop, but much more plausible in the enterprise space) extreme of 32, the Vertex 3 Pro is suddenly pushing ~217 MB/s.
Once we move to larger random reads, we see OCZ's newest drive take the lead again.