In regards to SM2 and Video Coding:

SM2, as redpriest stated, is a work of enthusiasm that begain when I was in college with version 1.0. Version 2.0 has Alex and I working collectively upon it. As Alex, redpriest, stated, we purchase hardware and software out of our own pockets and do not recieve support from Intel or AMD. Case in point, I just purchased for 400$ the Intel MKL library for hoisting the DGEMM and SGEMM testing currently done with my own code out of SM2 and leverage Intel's MKL and AMD's ACML. Soon, in a new release, SM2 will test FFTs, Linear Alegebra, be multi-threaded and have support for new bnechmarks from a vectorizing compiler.

In regards to media and encryption. Encryption is very tightly bound by dependencies, SHA1, SHA256, RC4, etc. The greatest optimizers achieve almost 1.5-nearly 3 IPC presently upon their algorithms. Like Alex, I had expected more from Conroe upon this benchmark with it's 4 way issue of ALU ops. In media, SIMD vector instructions are heavily utilized, in MP3 popularly with MMX while video codecs use SSE2 and soon MNI. These vector instructions will run +2x better upon the "core" arch. What is witnessed in the BLAS benchmark results of SM2 will also hold true in Media apps with SIMD vector optimizations.

my 2 cents for what its worth..