Bert said:
Is I/O performance strictly determined by chipset selection, or do some
makers manage to consistently get better I/O performance through their
own clever designs?
For an Intel-based MB, what's the best choice for building a system in
the next 6 months or so?
Are you referring to "disk I/O" or some other interface ?
The chipset here does OK, getting 500MB/sec on the SATA III
interface of the latest Intel chipset. I think this
Intel chipset has two SATA III ports and four SATA II ports.
(The B3 stepping Southbridge, is the one that fixed the
"degrading SATA II port" problem.)
http://www.bit-tech.net/hardware/2011/02/24/ocz-vertex-3-ssd-preview/3
AMD SB850 has six SATA III ports, but it's possible the hub
bus, only has enough bandwidth for a four drive Vertex III RAID 0
array.
If Intel were to put SATA III on all six ports, then the
DMI would be over-subscribed. In that regard, both AMD and Intel
might have the same hub bus limits (and both hubs are
very similar to x4 PCI Express Rev.2 or 2GB/sec total).
http://en.wikipedia.org/wiki/Direct_Media_Interface (Intel's hub bus)
"The original implementation provides 10Gb/s each direction (using a x4 link).
DMI 2.0 (introduced in 2011) doubles the transfer rate to 20Gb/s with a x4 link."
20Gbit/sec * 8/10 (coding) * 1/8 (bytes/bit) = 2GB/sec (minus packet headers overhead)
http://en.wikipedia.org/wiki/Comparison_of_AMD_chipsets
"Note 3: A-Link Express III is essentially PCI-Express 2.0 x4 lanes
giving 2 GB/s bandwidth." -- used on SB850 and others
The hub bus could always be bumped up some more. That's to give
you some idea what the upper limits would be.
Due to the bidirectional wiring, and the full duplex operation,
many array to array transfers will be non-blocking in terms of
the I/O. So if you set up a rwo drive Vertex III RAID 0, and
a four drive SATA II RAID 0, you could transfer from one
array to the other quite comfortably at 1GB/sec or so.
Paul
