Relic, What you wrote is basically true, one PCI-E lane is used for the various Marvel 91XX SATA controllers by all mother board manufactures, so it is immediately limited to one PCI-E lane's bandwidth, of 5Gb/s, short of the SATA 6Gb/s maximum theoretical rate. Not that any single HDD or SSD that exists today can even perform at the more reasonable 600MB/s actual bandwidth quoted for the SATA 6GB/s interface anyway.
The one catch is that Marvell advertises their products for use with one PCI-E 2.0 lane. From Marvell's website:
- 88SE9128 (PCIe ×1 to 2 SATA 6 Gb/s ports RAID controller)
- 88SE9125/88SE9120 (PCIe×1 to 2 SATA 6Gb/s ports I/O controller)
See for yourself:
Whether or not the chips themselves are limited to one PCI-E lane, I do not know, or if they can be implemented with multiple PCI-E lanes, but no one does AFAIK.
But all of this was not my point. The problem with the Marvell SATA 6Gb/s interface is it's poor "user experience", ie 'drives not being recognized by the BIOS or OS when connected, 'drives "disappearing" from the OS randomly, RAID configurations failing or failing to configure correctly, and the data transfer rate falling off drastically over time when the interface is used for long intervals. Those are the most common complaints regarding the Marvell SATA chipsets seen in forums, such as this one or in any mother board manufacture's forum. Those problems seem to be associated with the drivers that Marvell provides, as well as the lack of documentation for their products. Intel provides the only Marvell documentation I have ever seen, that being one five year old document. Nothing is available from Marvell's website that is of any use to the end user. IMO in some cases the problems with the Marvell interfaces is due to user ignorance and error, which is exacerbated by the lack of documentation, but certainly not all of it.
On the other hand, reviews of mother boards using the Marvel SATA interface chips don't complain about them at all as far as I have seen. Prior to the release of the Intel 6 series chipsets, the Marvell SATA 6Gb/s chips were among the best performing interfaces according to several reviews I have seen. Also, Marvell has recently released a new AHCI driver (Beta version) for their chipsets that functions much better according to users and in my personal experience.
Thanks for the info. I have one of the serise 6 chipsets. I have a batch file that measures restart time - the time to shut down, post and come to the desktop. I first tried it with my X25-M boot drive and it was 28 seconds. I disabled the Marvel controller and it went down to 23 seconds. I have so far been able to leave Marvel disabled but I am sure the time will come that I will need it.
Relic, You're welcome. Those are some quick restart times, thanks to the SSD for the most part I'd say. On one PC I have with the Marvel 9120 chipset, during POST I can see the Marvell boot messages (set in UEFI to do so) which takes a few seconds, disabling the Marvell chipset saves that time as you have seen.
It's not my intent to pronounce the Marvell SATA interface as not worthy of use, but to simply (hopefully) describe the realities of it's functioning as they are. I have not used it extensively, but when I have they've worked very well. I should say that is with the new AHCI driver and not in a RAID set. I have it connected to an eSATA port on a PC, using an external HDD case with a standard internal HDD. Works just fine, fast, no glitches. I see people in forums jumping on the hate Marvell band-wagon when someone complains about it, which I do not want to instigate. Experiment with your's sometime, and see if your mother board manufacture has posted the Beta AHCI driver on the download page for your board, I've seen zero complaints when using that driver.