[Beowulf] Athlon64 / Opteron test
bill at cse.ucdavis.edu
Thu May 13 11:48:33 PDT 2004
> In order to do the test, we have no doubt about the OS: Red Hat
> Enterprise 3, but we are a bit confused about the harware of choice:
Just FYI, I run RHEL-3 on both Itanium 2, P4, and Opteron and haven't
noticed any practical differences.
> As far as we know, Opteron has two main differences:
> - A wider memory interface (128 bit in front of 64)
> - A larger L2 cache memory (1 Mb)
Athlon FX, and opteron have a 128 bit memory interface.
Athlon FX, and opteron have a 1 MB L2 cache.
Some athlon 64's have 1 MB L2 cache, some have 512 KB l2 cache.
> Which is the theoretical maximum performance gain using full 64 bit
> architecture in front of 32 bit, taking into account that our
> computations are done in double precision floating point using really
> big matrices?
I didn't quite parse that. "32 bit" and "64 bit" have basically nothing
to do with double precision floating point. Accelerated double FP has
been common in "32 bit" chips for quite some time.
In any case the best way to decide on nodes, is to take all the nodes that
meet your needs (run the needed compilers, OS, and applications), fit
any weight, heat, space and heat constraints, then take the application
that justifies the cluster purchase and measure both performance and
Buying nodes on "theoretical maximum performance" is much like
buying a car based on what the maximum RPM of the engine is.
> Is it nowadays the 64 bit solution using Linux ready for production?
> If this is the case, which problems may we have to deal with in order to
> compile and run our code in a full 64 bit environment?
Well written code should "just work" with or without a recompile.
Well written means never assuming that a pointer and an integer is the
same size (by assigning one to the other), or assuming (with pointer
arithmetic) that a pointer is 4 bytes. Often porting an application to
a new architecture exposes bugs in the original code that affected the
original architecture as well.
> Which is the most mature solution: AMD Opteron or Intel Itanium?
I've run both without noticing a lack of maturity. What exactly do
you mean by mature? If you mean OS, Compiler, Application availability
I'd definitely say opteron. If you mean units shipped, again I'd say
opteron. If you mean something else I'm not sure.
Computational Science and Engineering
More information about the Beowulf