x86-64
While Intel continues to extend and enhance the performance of the x86 ISA well in to the later stages of its life, with enhancements to its NetBurst CPU core and subtle changes to the ISA itself with things like SSE and the upcoming new instructions in Prescott (rumoured to be tightly linked to the new HyperThreading implementation, itself a branch off the core ISA), it refuses to take it to the next big step.Moving to 64 bits is the next logical progression for an ISA that's served the computing industry for close to 20 years.
64-bits applies not only to the maximum addressable memory size or the data size the processor can operate on, the attributes most people apply when thinking about the bit parts of a processor, it applies to a whole swathe of processor features. 64-bit native data operation allows you to store a 64-bit memory address, giving access to that all important > 4GB memory size limitation of the traditional x86 ISA implementations. While the 36-bit addressable memory size 'hack', available on processors that implement PAE, gives you 64GB of addressable memory, 4GB has been the reasonable limit on x86 hardware since the introduction of the 386.
The addressable memory limit rises to 264 bits (18,000,000TB), but the process address space (the memory size limit that a process executing on the processor can address) is only 48-bit, giving 248 bits of available memory for a process. Not quite 18 million terrabytes, but a bit bigger than the usual 32-bit (4GB) limit on current x86.
x86-64 removes that limit on x86 for the first time. 64-bit also allows greater data precision in calculations. While current x86 can do double precision floating point calculations on 64-bit data types, with only 32-bits of available processing power per clock, 64-bit data operations take longer than 32-bit calculations to complete as the processor uses multiple registers across multiple clock cycles to perform your calculation. Indeed, current x86 processors store floating point types internally as 80-bit precision numbers, but the clock cycle penalty still exists. x86-64's main benefit in terms of data precision therefore lies in the integer processor on the CPU, not the floating point unit.
The current physical implementation of the x86-64 ISA, Opteron, also has a range of functional parts crucial to the performance of the processor, so let's talk about those.
