I think you are confused about addressable memory and physical memory.
Every process in Unix/Linux, Windows, and VMS (I don't know about other OS, but they probably work the same way) is given a virtual memory space (you call it addressable memory). In 32bit systems, you can address up to 4GB. The kernel usually give user mode apps 2 to 3 GB of virtual memory space for them to run on, even if you only have 128MB of RAM in your system.
I don't know how you come up with 4GB RAM plus 1GB PCI space equals 5GB addressable memory. But the bottom line is if you have a process that needs to address more than 2 or 3GB of memory (database, 3D modeling, simulation app, etc), 64bit is the only way to go. Other than that, with all else equal, 32bit machine should be faster, at least in theory.
You say you run parallel builds. Unless each of your source file is gigabyte big (than the problem is between your keyboard and your chair), chances are you won't need 64bit addressing. Each cl.exe, g++ or whatever your compiler is, will run happly inside your 4GB RAM and each of them will have their own 2GB/3GB address space. Just in case you don't already know, you don't need a 64bit machine to build a 64bit app.
Slashdot Mirror
32bit C runtime libs also have 64bit time_t. 2038 isn't a problem of 32bit machine, it is a problem of 32bit time_t used in many libraries.
I think you are confused about addressable memory and physical memory.
Every process in Unix/Linux, Windows, and VMS (I don't know about other OS, but they probably work the same way) is given a virtual memory space (you call it addressable memory). In 32bit systems, you can address up to 4GB. The kernel usually give user mode apps 2 to 3 GB of virtual memory space for them to run on, even if you only have 128MB of RAM in your system.
I don't know how you come up with 4GB RAM plus 1GB PCI space equals 5GB addressable memory. But the bottom line is if you have a process that needs to address more than 2 or 3GB of memory (database, 3D modeling, simulation app, etc), 64bit is the only way to go. Other than that, with all else equal, 32bit machine should be faster, at least in theory.
You say you run parallel builds. Unless each of your source file is gigabyte big (than the problem is between your keyboard and your chair), chances are you won't need 64bit addressing. Each cl.exe, g++ or whatever your compiler is, will run happly inside your 4GB RAM and each of them will have their own 2GB/3GB address space. Just in case you don't already know, you don't need a 64bit machine to build a 64bit app.