I'm a systems engineer at AMD's Austin design center (where they design the latest CPUs). I'm on a team that administers the 500 2P linux desktops that the engineers use, as well as a 5000-node linux cluster that they use to compile/simulate core designs. Among the more interesting statistics:
>100TB disk space on-line, via NFS
gigabit network to everything
~1000 dual-proc K7 systems in cluster (1U)
~500 single-proc nForce2 K7 systems in cluster (4U)
~3500 dual-proc K8 systems in cluster (1U)
~50 quad-proc K8 systems in cluster (64G ram each! ::drool:
We do everything: repair broken systems, maintain the network, write all the software/scripts that are needed to efficiently run this large of an operation (trust me, it's a whole different ball game after you start managing > 500 hosts)
Big props to the rest of my *nix-using IT comrades!