[Beowulf] SC|05: will 10Ge beat IB?
hahn at physics.mcmaster.ca
Tue Nov 22 07:56:20 PST 2005
> The Chelsio CX NIC is about $800. You correct in that it
> uses the new "skinny" IB copper cables and connectors. I
CX4 is similar, but not identical to IB, afaik.
> don't know the bend radius of these new cables, but the old
> IB cables are very sensitive to bend radius (i.e. make the
> radius of a bend too small and the cables are toast).
I don't think this is really a big problem. optics have pretty
bad failure modes when bent the right way, too. heck, even
gigabit has a bend spec, though it's not often attended to.
> To get about 8-12 nodes, you either need to go with a Foundry,
> Force10, or Extreme 10 GigE line card (i.e. $$$).
24pt 10G can be done on a single switch now, which necessarily
brings its price down - there were a couple examples of this at
> Quadrics was showing a new 96-port 10 GigE switch at SC05.
Myricom offers something similar.
> It uses the Fujitsu ASIC. They use 8 of the 12 ports in the line
I'm a bit puzzled by the Quadrics product, since in the booth they
basically justified it by saying they were ready to take on a bigger
> Another company is developing a new 10 GigE switch ASIC.
> Fulcrum Microsystems is developing a new ASIC with great
> performance (200 ns latency) and 24 ports. The general cost
I asked about this a bit - the latency sounds great, but if you
expect a 10G nic to cost ~10 us, it hardly matters whether the
switch is 200 or 600 ns.
> I talked to Myricom about their 10G solution. A ballpark
> price for it is about $1,200 a port, but I don't know their 10G
> switch prices yet, so it's REALLY a ballpark.
afaikt, you're not obliged to use myri 10g nics with myri's
switch (fulcrum's would work, for instance, or one of the others.)
though if you do go all myri, you get to optionally run MX
at better performance than IB...
More information about the Beowulf