Todd said:
From the HD-Tach report
SATA 3 machine:
Seagate ST500DM0 02-1BD142 KC44
ICH10R SATA 2 internal controller
SIIG SC-SA0L11-S1 PCIe SATA 3 controller
SATA 2 machine:
Seagate ST350051 4NS SN12
PCH controller
Factory settings
You do realize, there are about three different performance levels
for SATA III controllers.
1) SATA III native on the Southbridge. This is likely to reach
the full speed of your new SATA III SSD. The reason full speed
is possible, is the hub bus from the Southbridge to the rest
of the system, runs at 1GB/sec or 2GB/sec. Plenty of bandwidth,
as long as there aren't too many competing subsystems blasting
away at the same moment in time. When people "benchmark", normally
the other subsystems are quiet. With such a "native" port,
you might get to see your >500MB/sec SSD perform properly.
Some of the newer Intel motherboards, have two native SATA III
ports on the Southbridge. And AMD Southbridges, I think some of
those have six SATA III ports. Those are examples of "native" designs.
2) SATA III chip on PCI Express x1 Rev.2 lane. SATA III is 600MB/sec
as a cable rate. PCI Express x1 Rev.2 is 500MB/sec on the diff wires
leading to the slot. It means the SATA III chip on the add-in card
will be robbed of its peak rate.
Note that certain early SATA III chips, aren't even capable of full
SATA III under the best of conditions. Some can only manage 345MB/sec
downhill, with a wind blowing at their back. An enemy of high performance,
is non-pipelined protocols. "Full" rates only come, if the hardware
can be convinced not to wait for Acknowledgements. One of the reasons
the older USB standards can't reach "full" rates, is the kind of
protocol used (polling). It's not clear why the 345MB/sec chip
can't go faster, but some design issue is likely responsible.
I'm not expecting "a new driver" to fix that.
3) SATA III chip on PCI Express x1 Rev.1 lane. In this case, the single
PCI Express lane runs at 250MB/sec, and now we're no longer in SATA III
territory. Now, the new add-in chip has been compromised by the
PCI Express bottleneck, to a significant extent.
Based on your test results, you're seeing case (3). You've placed
a SATA III chip on add-in card, in a PCI Express x1 Rev.1 slot.
(I think your benchmark is measuring "burst" transfer rate,
rather than sustained transfer rate. Bursting to the cache chip
on the hard drive, can run faster, until the cache chip is full
and the platters become the limitation.)
*If* your motherboard has multiple video card slots, and you configure
things right, you may be able to move the SATA III card to a Rev.2 slot,
like an x4 or x16 Rev.2 slot. PCI Express auto-negotiates, and the hardware
can tell how many lanes have wires running to them. That will change
your case (3) to a case (2), and you'll see an improvement while
using your new SATA III card. At least, on burst transfers. If
you're benching with a *good* SSD, then your sustained transfer would
be faster, when you use a video card slot for the x1 SATA III card.
It's pretty hard to find an add-in card chip, which is not compromised.
I'm still looking. Highpoint brand, seems to be experimenting in this
area, and if any company can break the "crap gap", that's where
I'd look. Many others are content to ship the cheapest Marvell chip
they can find, knowing the users aren't smart enough to figure out
what they got.
The same kind of careful analysis applies, when adding USB3 cards to computers.
You could have a Rev.1 or Rev.2 slot, and your USB3 will not run full rate,
if stuffed in a Rev.1 slot. Fortunately, the USB3 enclosures haven't
been doing that well either, so the 200MB/sec a USB3 enclosure might
offer, won't uncover the difference in the PCI Express slot revision.
It's only when you use a Blackmagic USB3 video capture, and their
software tests your transfer rate, it'll notice your lacking Rev.1
slot and refuse to do captures. That's the only case I know of
right now, where it matters.
HTH,
Paul
