[gh02t]

Indeed, it always surprised me that TBB didn't have more traction in the scientific community. The Intel compilers and MKL are pretty popular in the HPC community as is, so it seems natural that TBB would get more attention.

IMO, the reason is probably because of exactly what you said - the programming model is a bit different from the standard "slap an OMP pragma on it" and requires deliberate design to target TBB. From my small amount of experience with it though, it's actually pretty easy to use and can often be much more approachable than the more down-and-dirty OpenMP shared memory approach.