Marvell 88SE91xx / 88SE92xx based PCIe SATA-III controllers

[Hervé]

These have been mentioned on other forums before but I got an opportunity to buy one of those PCIe SATA controllers for a few Euros. The card was advertised as based on Marvell 88SE9128 chipset but the model I received turned out as a 88SE9123. It certainly does not seem to make a difference and I reckon that most if not all Marvell 88SE9xxx-based SATA controllers would work exactly as this one does.
 
The card comes as a basic and very compact PCIe x1 card with 2 x SATA-III 6Gb/s ports. Variants with 4 x ports or more exist too (there are various manufacturers offering SATA controllers based on those Marvell chipsets on PCIe x1/x2/x4 cards: Syba, Startech, IOCrest, etc.). It works OOB in Snow Leopard, Lion, Mountain Lion, Mavericks and Yosemite. By default, it's listed as "Unknown AHCI controller" under SL and as "Generic AHCI controller" in more recent OS X versions such as Mavericks or Yosemite, which can easily be fixed through a little patch of AppleAHCIPort kext. Just insert the card in a PCIe slot (x1, x4, x8 or even x16) and attach the disk(s). Perfect for older desktop PCs that do not have a SATA-II/SATA-III/AHCI capable controller and wish to enjoy SATA HDDs or SSDs at PCIe x1 speed (250MB/s for v1.0, 500MB/s for v2.0) and, for SSDs, enable Trim.
 

 
 

WS670_SL:~ admin$ lspci -nn
pcilib: 0000:05:00.0 64-bit device address ignored.
[...]
04:00.0 SATA controller [0106]: Marvell Technology Group Ltd. 88SE9123 PCIe SATA 6.0 Gb/s controller [1b4b:9123] (rev 11)
04:00.1 IDE interface [0101]: Marvell Technology Group Ltd. 88SE91A4 SATA 6Gb/s Controller [1b4b:91a4] (rev 11)
{...]
WS670_SL:~ admin$ 

`
The PCIe bus obviously drives the overall controller performance. As such, the true speed obtained on drives (and especially SSD) depends entirely on the PCIe version. As a reminder, theoretical unidirectional speeds per PCIe lane (in MB/s) are as follows:

version\slot    x1        x2        x4        x8        x16
PCIe 1.0   |   250       500       1000      2000      4000
PCIe 2.0   |   500      1000       2000      4000      8000
PCIe 3.0   | ~1000     ~2000      ~4000     ~8000    ~16000   (round-up values, real rates are about 1.5% lower)

`
Performancewise, I obtained the following Blackmagic results on my old Precision 670 (PCIe v1.0, i.e. 250MB/s lane) with my Toshiba SATA-III SSD (Trim enabled) :

• Read: 215MB/s
• Write: 145MB/s

'not a huge improvement compared to results obtained with the SSD off the integrated SATA-1 (150MB/s) controller, but at least Trim is available.

[Hervé]

StarTech PEXSAT34 works OOB under OS X version from SL 10.6 to Yos 10.10. Not tested beyond Yosemite (deskop used for testing is far too old) but no reason why it would not. It's a PCIe 2.0 in x4 format offering 4 x SATA-III ports under Marvell 88SE9123 controller. Port #1 can be shared with eSATA port at the rear.

NB: in my Precison 670 where the card was fitted to a PCIe x4 v1.0 slot, a single port only operates @250MB/s, i.e. single PCIe lane, which contradicts what StarTech support stated to me: that a single SATA port would run through all 4 lanes. In fact, it seems to me that the card operates at 1 x PCIe lane per SATA port and that there is no concatenation of lanes at all; at least on my old dinosaur.  I also obtained slightly poorer benchmark results than with above PCIe x1 model.