I am very disappointed that apparently the performance in
- not simd, not floating point
- not recompiled every time you get a new CPU
has actually gone down
. I mean WTF. I spend a lot of time in single-threaded non-media things, just starting from linking large C++ programs.
The only good thing I can see is that at least price/performance has been picked so that the price for the same performance if you do use all cores is still better. I still expect to have a dual socket system with registered RAM with Bulldozer as my next virtualization system. But ... meh.