Consistency
Most benchmarks hammer the SSD when it's in a fresh or near-fresh state. What we really want to know is how the SSDs respond when they've been degraded, as happens over time, before technologies such as TRIM ameliorate the damage.
What's needed, ideally, are tests that keep the drives occupied for a while, hamstringing their performance, and then benchmark them in this heavily-used state.
Here's where PCMark 8's expanded storage tests come in. The consistency test measures the performance consistency and degradation tendency of a storage system. In the first part of the test, the cycle continues until a steady degraded level of performance has been reached.
Starting by writing 128KB writes across the whole drive and then degrading it further by random writes varying in size between 4KB and 1MB, the 10 trace benchmarks are run in between the degradation phases. Once in steady phase, achieved by 45 minutes of random writes, the tests are then run again between each phase.
In the second part, the recovery of the system is tested by allowing the system to idle and measuring the performance with long intervals. The final recovery phase allows the drive five minutes of idle time, ostensibly for TRIM commands, and then benchmarks them against the 10 different activities.
So, you have performance when the drive is degraded, when it's in a steady state, and when it has a chance of recovery. In all, there are 18 phases to the test - eight degradation, five steady and five recovery - and the best drives should be able to keep up a high level of performance in each area. The benchmark returns a multitude of data, from latency, bandwidth, access time, playback time to busy time.
We've chosen to focus on the Photoshop Heavy and Battlefield 3 trace tests, out of a possible 10, thus mirroring heavy 2D and 3D loads on the storage system.
What we really want to see is a flat line high up on the graph that denotes consistently high performance in all states; G.Skill's Phoenix Blade does just that.
Yet Plextor's performance is significantly lower in both regular and Turbo states, though the Turbo-enabled iteration's speed jumps up to over 400MB/s in a recovered state.
This benchmark tells us that Plextor's Photoshop performance in a heavily used state is about as good as a SATA drive's.
The Battlefield 3 gaming trace imposes a different type of load on the drives. The M6e's peak bandwidth is practically identical in both states. More importantly for the enthusiast, the numbers are lower than a decent SATA drive's until the recovery phases kick in.
