List of device bandwidths
mo at ccr.org
Thu Jan 17 19:27:03 CST 2008
andre kesteloot wrote:
> Could be useful:
> André N4ICK
> Tacos mailing list
> Tacos at amrad.org
except where it's flat-out wrong (Frame relay shows 2Mbps.
while Frame runs at bearer speeds from 56Kbps on up),
grossly inconsistent (the number for Infiniband are
all-over-the-map crazy), or documents a non-thing (eg, OC-9).
but to give credit where due, it is a decent first attempt
and is considerably broader in scope. usefully so, than other
attempts i've seen to produce a similar table.
It is interesting to note that the AGP-8 video card interface
cedes no quarter to PCI-E when it comes to speed. the difference
is that AGP-8 is a hump on the back of a kludge with
very iffy "signal integrity", while PCI-E is a point-to-point link
designed with modern small-swing signalling. The point not
made is that PCI-E is a packet bus - meaning every transaction
has significant overhead which AGP-8 doesn't have. But Intel
*had* to have a counter for RapidIO, RocketIO, and HyperTransport
and PCI-Express was it. the two Rs still exist - Freescale DSPs use
RapidIO and some FPGAs use RocketIO signalling to drive pins
at warp speeds. the current Opertons use HyperTransport 2
and HT3 is on the way with much more go-fast.
one novelty is that there are PCI-Express switch chips coming and
folks building *active* cables which will allow PCI-E to be
used not just between two chassis, but as a general interconnect.
HT2 switches already exist.
we now return you to your regularly scheduled QRM.
More information about the Tacos