For those who have asked: I do not presently have any plans to add more tests of other SSDs into this article. The primary purpose was to illustrate the differences seen between AHCI and IDE, and not necessarily demonstrating the differences for every SSD controller. I didn't include Fusion-IO, RunCore, Micron, Toshiba, Marvell, Intel, Mtron, MemoRight, or PhotoFast... to name a few.

This fixed the issue with my large SATA disk disappearing when resuming from sleep.

#support.microsoft.com/kb/977178

I came here trying to find whether I should use IDE or AHCI with a new Sandforce SSD and this article answered my question perfectly! Thanks for the great article :)

NCQ is not a feature you want on SSD!! As SSD has no spinning storage, but instant access, so NCQ is just an overkill as your access-times show.

I see alont of SSD reviews, but they all reference Intel Platforms and Intel Drivers etc...
It would be nice to see some Tests/Reviews of Benchmarks that provided results for both camps ( Intel & AMD )
I would also be appreciative to see Microsofts AHCI Driver tested/included as well as AMD's ATI AHCI Driver tested/included. As well as the Marvell AHCI driver that is more commonly seen in AMD platforms verses what shows up or is seen in the Intel Platforms.....

Thanks for this review, but it is hard to apply what I have read to my AMD Platform which uses ATI AHCI Drivers &/or Microsoft AHCI Drivers if I use the MS Windows 7 generic AHCI driver......

Both bas and olin have missed the mark here by trying to oversimplify the issues. There is no direct one-to-one relationship between or among IOPS, access time, and bandwidth. Too many factors intrude for such simple relationships. For truly independent IO's, access time can put an upper limit on IOPS, but more than a small percentage of IO's are logically and physically dependent. Depending on the intelligence of the drive and OS software in dealing with various forms of IO, huge differences in performance can arise. Depending on the page and block saturation in an SSD, writing the same file to that device can produce vastly different results, even in the absence pf interfering or block overlapping IO. Different fill methods for blocks and pages can cause what appear to be random changes in performance. Even in the simplest of cases with known algorithms, understanding these performance fluctuations can be complex.

If you would like to push a drive to the maximum performance in a real world situation, you should put a single large paging file on the drive and use programs that drive the paging rate as high as possible. When that settles down to a constant rate over time, then a measurement of paging rate will tell you what the maximum performance is in a real world situation. I am not sure that TRIM will actually improve performance in a busy drive, and it certainly loses the ability to recover any deleted file. A smart TRIM handling on a drive would queue the TRIM requests up and satisfy them when the drive is not otherwise busy, rather than giving them the priority they have under existing classification. TRIM could certainly help in a situation where the read to write ratio is high and the drive is not particularly busy, but this also would lessen the advantage the SSD has over a drive with a large built-in cache.

...continued...

Where am I wrong? AHCI adds commands to the protocol that are rubbish, like NCQ.
There is no point in doing NCQ as there is no rotation of the drive where command ordering is needed or even wanted.
NCQ has been proven to hinder the performance as it can stall the controller by making the computer WAIT before giving new commands.
SSD is a matrix and ordering is silly as all cells deliver data at the same speed, there is no ordering wanted to optimise the rotations in where data can be read.
But AHCI is still doing it, as such it lowers performance, and this can be noticed when using better benchmarks that can do proper I/O reading and writing combined with multiple commands at the same time. That benchmark has been here for ages, called ATTO.
Give ATTO a high command-que depth and see what happens, it will show the difference. As ATTO gives read and write speeds, for me it's the best harddisk/ssd benchmark on the planet when used properly.
Meaning VERY large Block-size and high Que-depth, then it gives pretty accurate numbers.

TRIM works in both modes. Thank you for pointing that out.

I can't have my HDD run in AHCI together with my SSD. It won't load the OS and keeps on crashing.