The aim of TBB, which I've described in previous coverage of Intel's multicore compilers, is to make it easier for coders in C++ to express task-level parallelism. TBB works by abstracting parallelism above the level that most programmers are currently used to, especially those coders who use POSIX or Windows threads. Coders can use the TBB template library in conjunction with the platform's native threads, and the different pieces of the project (task scheduler, mutex locks, atomic operations, containers, etc.) can be used independently of one another.
Intel clearly wants to position TBB as the standard tool for writing multithreaded code, as opposed to OpenMP, and Windows and POSIX threads. The company claims that the project as a whole is committed to processor-, compiler-, and OS-independence, and they've launched a new site for the open-source version: threadbuildingblocks.org. The site has all the typical open-source project aspects, like a CVS, forums, mailing lists, and so on, and Intel is currently in the process of adding engineers to it. TBB 2.0 currently runs on non-Intel hardware, like the G5, and on operating systems from Solaris to Linux. It also works with multiple compilers, including Intel's own in-house compiler and gcc.