(PeteCresswell) said:
In trying to track down a problem, I stumbled on to the fact that
one of my drives seems to support a vastly faster transfer mode
(6 gigs/second) than the others (300 megs/second).
e.g.
Slow:
http://tinyurl.com/bn8x2do
Fast:
http://tinyurl.com/cd7dlfw
I'm assuming that the limiting factor for the faster drive is the
onboard SATA controller it's plugged in to.
The Questions:
- Have I got it right so far?
- If so, can anybody recommend a SATA card that would take
advantage of the faster max speed?
Newer hard drives, support SATA III, but it's really a waste
on a hard drive. The best platter limited transfer now is around
180MB/sec. Your average garden variety HDD is 135MB/sec near
the beginning of the disk. Being able to burst at 600MB/sec
doesn't really help with the platter based limitation.
You can get SATA controllers with SATA III interfaces, but they're
a bit limited on the PCI Express side. If the chip has a
PCI Express x1 Rev.2 interface, that gives 500MB/sec. Whereas
the SATA III is 600MB/sec. (In both cases, those numbers ignore
packet overhead, which is still a factor.) So the SATA controller
chips, added via a card, may be limited a bit by the motherboard
connection. In theory, a modern motherboard with SATA III in the
Southbridge, might do slightly more.
So this is mainly a concern for SSD owners, when they want to
benchmark and get the very last MB/sec of their new purchase.
There are SSDs that can function as "test" devices for the transfer
rate. A hard drive, on the same port, really isn't coming near the
limits in a practical way. A SATA II interface can sustain a long hard drive
transfer, without limiting the disk.
So, say you have an older motherboard, with PCI Express x1 Rev.1
slot. That is limited to 250MB/sec per lane, so even if you plug
a SATA III chip into that, it can't really run at a SATA III rate.
If the motherboard has a PCI Express x1 Rev.2 slot, that will do
500MB/sec per lane (one lane). That's a slightly better match, but
may still hold back the SATA interface slightly. There are also
cards that have a PCI Express x4 interface on them, but as long
as the SATA chip has the x1 lane interface, there will still be
that limit. And then, the SATA on an Intel or AMD Southbridge,
is the next best option for "flat-out testing" or benchmarking.
As for whether it does AHCI in the driver, you'll have to look
at the driver files to learn more there. Or the description
of the chip or something.
You can do some architectural analysis of some Highpoint cards,
to see whether there are bandwidth bottlenecks or not, and whether
they could come close. Visually, this one looks like a candidate,
as it has a x4 edge connector, so perhaps it doesn't limit bandwidth.
The controller chip may have x4 PCI Express on it. So visually,
I'd look more closely at this one.
http://www.newegg.com/Product/Product.aspx?Item=N82E16816115114
By comparison, this card has a PCI Express switch chip and two
SATA controllers. Which means it goes from PCI Express x4 Rev.1
to a couple PCI Express x1 Rev.2 ports internally, so has the
typical potential for slight limitation. This is basically like
buying two x1 cards, and combining all the ports onto the one
card. Occasionally, putting two chips behind a switch, causes
a few problems (check reviewer comments).
http://www.newegg.com/Product/Product.aspx?Item=N82E16816115077
And to be sure about some of those things, you need to see benches
of them running real good SSDs. An early Marvell chip for example,
a SATA controller, has a SATA III interface, but the transfer
rate seems to be limited by the controller inside the chip
itself. People get perhaps 345MB/sec from it, so only a bit
above SATA II. So you'd start with some arch reviews, then
take the card model number and see if anyone has benched it with
a 550MB/sec SSD.
And the purpose of all this "vetting", is for your next SSD purchase,
rather than getting 135MB/sec sustained from a $50 SATA hard drive.
Virtually any SATA II port will do that for you. Hard drives
do have a cache chip, and the cache chip, for a short time,
can sustain a burst transfer, but we can't even be sure
on a hard drive, that there isn't a limitation there as well.
Even the benchmark utilities, can have trouble measuring the
burst transfer rate properly.
Paul
