IBM uses slower processors in their cluster machines. This keeps the heat generation down and allows them to pack more CPU modules into a smaller volume. So the individual processors are actually cooler than the Nehalem Xeon processors which we, Bull, use in our cluster machines.
Bull machines are water cooled and use fans which SCREAM when the heat builds up. I visited an installation that our crew had installed at a French nuclear computation facility last October. At the time the Bull Tera 100 cluster was the fastest in Europe and the 6th fastest in the world. Before we could enter the computer room we had to put on industrial grade ear muffs to protect our ears.
Not exactly accurate. I “owned” IBM’s worldwide HPC marketing for five years (still in the biz; different co. now). Blue Gene, Linux clusters, etc. can all ‘scream’ in a big way; depends heavily on # of procs./cores, individual proc. speed (and you’ll see IBM, HP, et al use the latest/greatest from Intel & AMD pretty much across the board), proper interconnect (BG uses a proprietary fabric; IB is the fabric of choice for clusters), other factors. Cooling is essential for any supercomputer, several ways to effect that (including, as you suggest, water cooled doors....but those tend to be the exception vs. the rule).