nothing beats good old ASM when you do HPC on a single node. :D
Don't know you if you're trying to be ironic here, but actually above is mostly true. For example, if you're up to getting any kind of performance from your x86 node, you simply have to deal with SSE (either through inline assembly, or using compiler intrinsics). And even if you are dealing with something that is much nicer for programming for performance than x86, like CUDA for example, knowledge of corresponding assembly language (PTX in this particular case), could be in my experience very helpful from time to time.
Fortran is great, but the job market?
Fortran job market is doing very well, at least in my experience, that is in turn mostly in the domains (computational electromagnetics, neutron transport, molecular dynamics) that are probably involved much longer with the HPC (I use this term in kind of its "traditional" meaning, which would be closely related to scientific computations) than what is the case with quantitative finances. There exist many scientific Fortran codes that are still actively maintained, and lots of new code get written in Fortran too. And to further perceive how vibrant is Fortran community, you could for example take a look into which is the only language besides C that is possible to use for CUDA programming - it's Fortran, in the Portland Group implementation (
http://www.pgroup.com/resources/cudafortran.htm). Now, don't get me wrong -
C++ certainly has its own strengths, but languages like C or Fortran are much more meaningful for numerical/parallel programming, so I'll stick with my (outsider, admittedly) impression that
C++ is heavily overused by quantitative finances community, at least for these particular purposes.