kpf, first I think that aixman was mistaken because he did not account for the fact that longs continue to be 32-bit under AMD64 (see my previous post). They use a different variable for 64-bit, longlong.
So even code under 2GB should run faster under a proper 64-bit compiler because of the extra registers (reduced memory accesses). Right now, Intel has the best compilers, hands down (and they don't do AMD64). We have a bit of compiler work to do.
Could you elaborate what exactly is the benefit of the extra adress space for the users of AMD-64?
This becomes a really big thing for programs which use more than 2GB of space (under Windows, or 3.5GB under Linux). Currently, either the application has to be optimized for Intel's PAE architecture - which is a bear for the programmer, and introduces runtime overhead - or worse, it has to disk thrash. Even with disk swapping, the whole application space hits a wall because of Windows limitations on how much space it can automatically use on the disk.
Here is the rub (and controversy): Currently there are no desktop applications which can use these huge address spaces. However, there are two catagories of desktop applications which are bumping up against the limit: High-end gaming and graphics development (things like Photoshop, Illustrator and video editing). The key to Athlon64 is a few compelling programs from this catagory - then the P4 will grind to slo-mo while the Athlon64 briskly sprints along. Will key game makers and Adobe come on board?
On servers, databases are already across the 4G line and require 64-bit machines to hold their working set of data in memory. That is one of the reasons that AMD released Opteron first: The applications are already compelling on servers.
Market Data 
Markets 
AI
