I wrote a bunch about SSD performance in this thread:
http://forums.anandtech.com/showthread.php?t=2425370. Short to say, every SSD uses RAID0 already to increase speeds.
With PCIe, this parallelization can continue to increase speeds indefinitely. But that also means a redesign of the controllerchip is necessary. Some PCIe SSDs are just PCIe -> SATA controllers with a regular SATA->NAND controller. A
native PCI-express design of the controllerchip is really the best. NVMe products are the new generation, which also addresses the SATA overhead, which is becoming a bottleneck for really fast SSDs.
But will all this extra speed matter? That is the real question. We can talk long about it. But you can
test it yourself! Yes you can. On your very system. You only need a stopwatch and some time.
1) Reboot the system, to purge all filecache from the system. After booting let the system settle to be really idle.
2) Start your favourite application, like a modern game. Measure the time it takes before the game finishes loading, or other in-game loading measurements, like changing zones in World of Warcraft.
3) Now close the application. Do not reboot. Wait a couple seconds. Launch the same application again and repeat the same measurements. They will be lower than the first time.
What you have measured is the difference between reading from disk (SSD) and reading from memory (filecache). This measurement is only valid if you have enough RAM to store all caches used during the experiment. You can see this with Windows task manager on the Performance-tab where
Cached is a high number after doing step 2) but
Free should still be above 1GB.
How to interpret the results?
uncached: 15.6 seconds
cached: 12.4 seconds
(Try to take an average or median of three or more attempts.)
Just suppose you got results like this. This means the difference is 3.2 seconds. This is the
maximum real-world difference you would ever see on your current system if you had the
fastest SSD that could possibly exist. Actually, by definition, it should be less fast. But you get the point: this is the
absolute limit of performance improvement of a faster SSD.
And you probably will be disappointed by the difference - it might be much less than in my example. This is because storage is no longer the major bottleneck as it used to be with harddrives. Today, the CPU is very much a bottleneck in everyday computing. Except networking perhaps, but that is another story. The issue is compounded by the fact that often only one CPU core can be utilised. Thus, you would need a much faster CPU (10GHz) to unlock the power of fast SSDs. Bottlenecks are like a pie-chart: if you cut away one part, then the other parts take up more % of the whole. The result is that
by having faster storage, you move the bottleneck back to the CPU.
