Re: a project supervisor that overhears snippets of real tech talk between the engineers and then passes it on. Is that true?
We'll all know soon. My money is (very literally) on Scumbria being right and Paul Demone being wrong.
:-)
One of Paul's frequent arguments is that Intel has spent many man-years profiling typical legacy X86 code, and used that knowledge to design P4. Therefore P4 had to be fast on any instruction that matters, and overall very fast on existing code. But in the next post, having been reminded that commonly used barrel shift instructions are a disaster for P4, he starts arguing that only recompiled code matters.
It's going to be a long, long, time before most of the code out there is optimized to avoid P4's weaknesses. First the compilers will have to be rewritten, then all the libraries recompiled, then the production software recompiled.
Getting that through to all the custom apps used in business and government will take a long time. When do you think Visual C, Java in its various incarnations, Visual Basic, Delphi, Smalltalk, etc. will start producing P4 optimized executables?
Until then, the P4 won't be worth much. AMD isn't arrogant enough to assume that world will change to match its processors, instead it designs its processors to work with the code and tools that are out there now.
Dan |
