John said:
If you cannot tell by the results I posted, what test should I run
to prove that my SATA 2 port is fully utilized?
As far as I can tell, my Intel Q9550 requires socket LGA 775 and
apparently there is no such motherboard that supports SATA 3. My
CPU rocks, and I'm not likely to replace it just so I can get an
SATA 3 capable motherboard.
Also, I seem to recall a local guru suggesting that SATA 2 should
run very fast.
Thanks.
There are some 520 series results here. The write speed suffers,
on the smaller ones like the 60GB one (since fewer channels are populated).
http://www.pugetsystems.com/labs/articles/Product-Review-Intel-520-SSDs-Cherryville-134
*******
To test a potential 420MB/sec sequential read, you'd need SATA III.
On writes, the interface isn't really a concern.
You'd get a SATA III controller card. But not just any card, because
you have another limitation.
Older motherboards have SATA II ports. You know that.
Older motherboards have PCI Express Rev.2 lanes on the video card
slot, but PCI Express Rev.1 lanes on the "lesser" PCI Express slots.
If you have an add-on card, which ends up connected to a Rev.1 lane,
that can form the "bottleneck", making the card no better than
an existing SATA II port. So the "plumbing" must be exemplary,
for an add-in card to be worth the trouble. And not too many card
companies, really give a damn.
If you buy the first SATA III card you see, you could be disappointed
with the benchmark results. And it's because of the PCI Express interface
details.
In terms of add-in card designs, the first "good one" for older
systems, was the Asus U3S6. It required the user to have at least
one available x4 slot. The lanes on the slot could run at Rev.1 rates,
and the card would still perform. That's because a switch chip on
the add-in card, converts two Rev.1 lanes worth of bandwidth, into
one Rev.2 lane of bandwidth. So if the actual SATA chip on the card
needs a single Rev.2 lane, the switch chip provides it. Asus released
that card, at a MSRP of around $25. But when you look at the "regular"
add-in card companies, if you want a similar design, they try to get $100 for it.
That scares away consumers, and dooms such cards to being discontinued
soon after they come out. So it's pretty hard to retrofit an older
motherboard, and "make a hero" out of it.
(Basic idea around the Asus U3S6 card... Switch chip does "gear change"...)
PCI-Express x4 Rev.1 --- switch chip --- PCI Express x1 Rev.2 --- USB3 chip
--- PCI Express x1 Rev.2 --- SATA III chip
You can see a similar kind of approach here. A card with an x4 interface
on it (won't fit in an x1 connector). A switch chip. Connects to two
storage chips. The chips are Marvell 9128. At least, that's the information
I could find elsewhere.
http://www.newegg.com/Product/Product.aspx?Item=N82E16816115077
Now, the thing is, first consider the 9128. If that is what is being used
on that card, it has a single x1 Rev.2 lane, at 500MB/sec. That's sufficient
to support one SATA III drive at close to full rate. So you'd put SSDs on port 1
and port 3, and leave the other ports unused, if attempting to attain a high
benchmark result. If something other than a 9128 was being used, like an
actual storage chip with an x4 interface, then some of the glossy claims
for the card (by its manufacturer) might be true.
http://hothardware.com/articleimages/Item1470/Marvell9128.jpg
It doesn't matter in this case, whether the x4 or larger slot was Rev.1 or
Rev.2. The bottleneck is the x1 lane on the 9128. There's really only
enough bandwidth to test one SATA III port flat out, on each chip.
PCI-Express x4 Rev.1 --- switch chip --- PCI Express x1 Rev.2 --- 9128 -- 2xSATAIII
--- PCI Express x1 Rev.2 --- 9128 -- 2xSATAIII
As well, if you read the reviews for the card, it's a PITA to use.
(The Asus U3S6 was also a PITA, in case you were thinking of Ebay.)
There is a BIOS flasher for the Highpoint card. Like a lot of storage
cards, there needs to be an Extended INT 0x13 BIOS routine,
to allow it to boot. They include a "Reallocate EBDA" setting, and
that becomes an issue when a motherboard has multiple controllers,
multiple BIOS add-in modules, and there isn't sufficient low
memory space for all of them to load. Examples of things
that don't "squeeze down", are things like the video BIOS on
an NVIDIA video card, which may continue to use 64KB of ROM
space during POST. There is very little space down there available,
for lots of boot options at POST time.
http://wiki.osdev.org/Memory_Map_(x86)#Extended_BIOS_Data_Area_.28EBDA.29
Anyway, that was an attempt on my part, to show that x4 cards do exist,
that can provide sufficient "plumbing", to maybe squeeze 400MB/sec+ out
of an SSD for testing. But it still might not be enough to match a
proper motherboard SATA III port (~530MB/sec). You really need to find
review articles, where the reviewer goes into details, to see if it gives
the desired results or not.
So now, it's time to benchmark the RocketRaid card.
http://www.bit-tech.net/hardware/2011/01/04/high-point-rocketraid-640-review/1
The 9128 appears to be a lemon. Which means searching for another
card with a similar design concept (combining x4 lanes to give 1GB/sec
available on older motherboards, so you can get at least 500MB/sec Rev.2 x1 lane
for a storage chip). The 9128 is barely squeaking by the SATA II
performance level.
http://www.bit-tech.net/hardware/2011/01/04/high-point-rocketraid-640-review/5
These things wouldn't be necessary, if they made storage chips with their own
x4 interface, rather than x1, but we can't expect miracles. I'm just
amazed some times, at the lack of choices in these areas. And also,
the pricing on some of the chips. There have been storage chips in
the past, that would have made excellent material for add-in card
design, but the chip price started at $100 or so, which leaves
them completely out of the consumer market.
I think it's going to take a bit of research, to find an
add-in card actually worth owning, and running in an LGA775
motherboard. Too many bottlenecks, and potential "lemons" out there.
If your motherboard is one of the ones with two Rev.2 video slots,
then the second slot can work with more modest SATA III cards. Like
in this diagram. The lane is going to cap things at 500MB/sec minus overhead,
so still can't be as good as some SATA III Southbridge ports. Maybe you
get 400MB/sec+ with a thing like this, rather than seeing 530MB/sec
if your SSD actually can do that. Since a 60GB 520 series does
about 420MB/sec, you'd be getting roughly in the ballpark, but
not with enough headroom to really know for sure the add-in card
wasn't still the limitation on reads.
PCI Express x16 Rev.2 ----- PCI Express x1 Rev.2 ---- One SATA III port at
Storage Chip close to full rate
HTH,
Paul
