scooter said:
this may seem like a stupid question, but how does the new FX-60 chip by AMD stack up against any of these? it's the first dual core FX line chip; basically two FX-55 (2.6GHz) on the same die.
i was trying to use this site as a very *rough* estimation:
http://www.systemshootouts.org/processors.html
You pay a premium for the FX designation as the the clock limitations are
removed (related to the multiplier I think). So you generally buy an FX so
that you can overclock.
If you're in this class of machine, you're better off with an Opteron than
the consumer chips because you can then put two, three or four of the
dual-core opterons together in one system. You can't do that with the
consumer X2, FX and A64 chips.
There was a benchmark shootout between the quad PowerMacs against
a quad opteron and I think that the Opteron won. In some things by a
little and in some things by more than a little. I don't recall whether or
not the PowerMacs won some of the categories. At any rate, you are
talking about some serious compute horsepower.
One other small issue is that to do a dual core duo with 64 bits when
those chips come out, you're going to need some kind of hardware to
do cache arbitration. The Opteron has on chip hardware to do this
(you can buy them with one, two or three channels I think) and this
is why you buy Opteron chips to do multi-CPU systems. The FX, A64, X2
chips don't have this so you don't have multi-CPU systems with the
consumer chips.
I assume that the PowerMac stuff has hardware to address this given
that IBM probably has done multi-CPU systems for quite some time.
So there would have to be some hardware to do this if you wanted to
paste multiple core-duo chips into one system. If you don't have
cache arbitration, then you'd have to slow down the system somehow
so that one processor didn't do a read from cache for a memory location
that another processor just updated. I suppose that you could just
lock the memory location but then a processor would have to wait for
the memory to be unlocked.
At any rate, it's clear that Opteron and Power already have this problem
solved. I don't know if there's an existing solution for Core Duo on the
same issue. Note that Core Duo does some really interesting cache stuff
between the two cores in the chip. But I'm pretty sure that they can't
do this off chip.