Hench said:
Cpu will be a 3770K
Looking for a motherboard recommendation that will allow memory oc'ing
but in addition:
has more than 2 sata 3.0 ports (preferrably 4)
has headers for USB 3.0 on back and front.
Was considering the Asus P8Z77-V or that class of motherboard. Nothing
extreme or glamorous please
In regards to my request for more than the standard sata 6Gb/s ports:
Does anyone even foresee a time in the next 5 or 7 years where drive
will max this spec?
There are SATA III SSD drives promising 550MB/sec on reads. SATA is
bound to have some packet overhead, so this is getting close to
the max possible over the 600MB/sec cable rate. (The overhead
doesn't contribute to the users available bandwidth. Sort of
how my GbE network interface can do 117MB/sec but not 125MB/sec,
there's packet header bytes as overhead.) So I would say we're already
using SATA III pretty well as it is.
This is only an issue for SSDs. Hard drives are nowhere hear
sustaining those kinds of rates. The best hard drive I've heard
of to date, is around 180MB/sec sustained. Only an SSD can
saturate a SATA III link.
*******
If you look at the chart near the bottom of the page, Intel doesn't
seem to put more than (2) SATA III 6Gbit/sec ports on the Southbridge.
For a motherboard to support more than that, would require the addition
of a second chip. Early Marvell SATA III chips, like the 9128, seemed
to be hardware limited to around the mid 300MB/sec range or so. So
when you do find your motherboard with the >2 SATA III ports, check
the review info for the *second* chip, to determine whether it's a
winner or not. Generally speaking, the Intel (Southbridge) ports
are the good ones.
http://en.wikipedia.org/wiki/List_of_Intel_chipsets
There's a reason for Intel limiting at least some of the chips
that way.
Intel processors have two different interfaces. Low end CPU chips
use DMI (1GB/sec or 2GB/sec). High end chips use QPI and
a "full" external chipset - bandwidths there are perhaps as
high as 25GB/sec. Maybe LGA2011 would be QPI.
Now, if you strapped four 550MB/sec SSDs to a proposed
four port SATA III Southbridge, used RAID-0 striping,
the user expects to see 4*550MB/sec or 2200MB/sec speed.
Yet, if there is DMI bus in the path, that speed cannot
be achieved. Some lesser number results.
If there is QPI, there's a better chance a non-bottleneck
path can be found.
Sure, you can stick a PCI Express SATA III RAID card in
a video card slot. But then, the user will ask "how do I
run my new pair of $500 Crossfire video cards" ? If you
attempt to use all interfaces at the same time, then
you might run into that Southbridge limit. That is, if
you're using LGA1155 or LGA1156 and the thing has DMI.
If you're only running one video card, then there's a chance
to put some $500 RAID card in the second video slot.
http://ark.intel.com/products/65523/Intel-Core-i7-3770K-Processor-8M-Cache-up-to-3_90-GHz?q=3770K
(Click the "Block diagram" link. It's got DMI.)
http://ark.intel.com/inc/images/diagrams/diagram-18.gif
Intel is seldom honest about DMI. You can find it quoted
in dishonest gigabits/sec rates. In some cases they even
double-count, by adding the TX and RX bandwidths together,
which is not telling the user anything useful in terms of
bandwidth limits to be expected. When your SSD is limited
in a benchmark, it's either purely a TX limit, or
purely an RX limit. There is no point adding the bandwidths
together, like a used-car salesman might be tempted to do :-(
In any case, you're getting a DMI, and two SATA III ports
are a relatively good fit for it. Doing four SATA III ports,
would require a move to the next level of DMI (some day, if
they wanted, they could use PCI Express Rev.3 PHYs in there).
Sticking a SATA III card in a video card slot, is a way to
build larger bandwidths, in say, RAID-0 form. And if you
try hard enough, eventually you'll discover a second level
of dishonesty. PCI Express slots may have fancy rates
like 4GB/sec or 8GB/sec, but those can only be achieved
with relatively large packet sizes. Many Intel implementations,
can't actually run at 4GB/sec or 8GB/sec. Someone did a calc
for my chipset, and found slightly less than 2GB/sec is
to be expected. Using small packet sizes, means perhaps
a smaller buffer in hardware, but sometimes the issue is
sharing nicely with other devices, and not causing
real time latency problems when multiple cards are present.
So if you "aim for the really high bandwidths", you only
end up running into more problems you weren't aware of.
I wasn't aware, until a couple years after I bought my
motherboard, that it couldn't really run right up to
the PHY speed limit of the PCI Express interface. It
runs out of packet bandwidth inside.
Paul
