There is a way to switch from IDE to AHCI without reinstalling windows, couldn't you apply the same principal to a AHCI -> RAID switch?
This is how SRT works once configured and enabled for the first time:
You access a block of data for the first time from your mechanical Hard Drive, at which point it is also copied into cache.
The next time you access that same block, it will be accessed from cache instead.
As you access additional blocks of data from your mechanical Hard Drive, they will also be cached, and accessed from the cache upon subsequent requests for that block.
This process repeats until your cache is full at which point the oldest blocks in cache are pushed out of cache.
Note that I reference "blocks" because SRT works on the block/hardware level, not the file level or the program level. If you have a 3GB file that you only regularly access a few hundred megabytes of, only the blocks that contain those few hundred megabytes of data are likely be cached (as opposed to the whole file)
Until the cache actually fills up, it will continue to cache every single block that you access regardless of how frequently you access it. Only when the cache becomes full does anything start to be pushed out of cache, and at that point it will start with the oldest data in cache, not based on frequency of usage. It doesn't use any sort of algorithms to try and predict what you might need cached or anything like that. It doesn't differentiate between a file you've accessed thousands of times and a file you've only accessed once. It is based ONLY on which blocks you've most recently accessed.
Of course, anything you use on a regular basis is unlikely to become the oldest data in cache as it becomes the newest data in cache again every time you use access those blocks.
You can only use up to 64GB max as cache. This is not really a limitation because in practice the caching is actually pretty efficient. Take the game World of Warcraft for example. A modern install of WoW is about ~22 Gigabytes but a lot of that comes from old map files from past expansions and old content. A WoW player might only access ~5GB of data or less on a regular basis as they play the game, unless they just love exploring old content for no reason. There are countless other examples of this, such as games that have both a single-player and a multiplayer. If you just play multiplayer, it's likely not going to end up caching all the files for the single-player campaign, etc. In that respect 64GB ends up being a lot of data. Nothing is likely to end up pushed out of cache unless it has been a significant amount of time since it was last accessed.
Great posting, I would like to add to that.
My understandig is that Intel SRT does a bit more than that, at least that's what I have read. Intel SRT should ignore very large files (ISO, TV recordings, big archives, ...) that would not profit very much from being cached.
I kind of see that in my testings since SRT does not improve sequential read speed above the speed of the HDD.
Take a folder containing many tiny files and some large ones, games often look like this. Copy that folder multiple times from the cached HDD to a even faster SSD and you should see that a) the tiny files get copied a lot faster and b) the large parts don't change in speed at all - even if the SSD used for caching is capable of much higher seq. read speeds than the HDD. Windows 8 has a beatiful graph in the new file copy dialog, makes it obvious.
I have never ever seen a read speed higher than the exact maximum of the HDD with a C300 128GIB added as cache.
But since you seem to have quite a bit of knowledge, there is something I could not find an answer for.
How does SRT act if it is set to "maximize", also caching write?
More precise, how much space does SRT take for writing on the cache? I like maximize, but if it caches let's say up to 30gig when writing, it would be quite a bummer losing 30gig of read cache. ~10% as write cache would seem the right balance to me, but again, I could not get any information on that.
Any more insight into "maximized" would be appreciated!