From what I understand, the noarch suffix means that it's architecture independent and should work universally. If this is the case, why should I install architecture-specific packages at all? Why not just go straight for the noarch? Are there optimizations in the x86 or x64 binaries that aren't found in the noarch binaries? What's best for high performance applications?
Folding@Home does this with their controller:
In this case the noarch "binary" is really just a control script (shell, perl, or similar - in this case it's a Python script). It is "architecture independent" because it's written in an interpreted language and doesn't care about the CPU it's running on.
You would still need a Client or Viewer appropriate to your architecture in order to perform folding or view the results (because these are actually compiled code, most likely C or C++, and care very much about the processor they were built for).
This is not to say you couldn't build an architecture-independend Folding-at-Home client or viewer, but unless you're offering to step up and implement it in a cross-platform language I wouldn't hold my breath :-)
What's best for high performance applications? It depends on the application, but typically compiled code, built and optimized for the architecture it's going to be running on.
