64-bit computing

The 32-bit era is coming to end, but what does this mean and what's so great about 64-bit? Ashton pads the bits.

The advent of 64-bit CPUs brought with it the promise of a new level of computing. The inevitable successor to 32-bit just as 32-bit succeeded 16-bit before it, we naturally assume it must better. But how is better, and how does it affect you?

The case for bits
I've said this in numerous X-Ray pieces, and it applies again here: the technology isn't new. 64-bit computing has been around since the 1960s when IBM developed its Stretch supercomputer, able to process data and instructions with 64-bits. But it wasn't really necessary back then, and even in more recent years 64-bit computing has been the domain of servers and supercomputers with real work to do rather than the pithy Windows thing desktops run.

Inevitably technology trickles down, but ironically the first consumer level devices to receive the 64-bit treatment were gaming consoles - the Nintendo 64 and Playstation 2. It didn't reach mainstream computing until AMD released its AMD64 line of CPUs, starting with the Opteron line, which ultimately spurred affordable desktop oriented 64-bit CPUs from AMD and Intel soon after.

The timing was partly driven by a limitation of the 32-bit architecture: 32-bit CPUs can only count so high. As memory became more affordable, and mainstream processors formed the basis for servers, the demand for memory rose and the humble 32-bit processor can only see up to 4GB in addressable space. This was partly satiated through the use of PAE (see 'Easy as PAE') but ultimately this was a stop-gap on the way to 64-bit CPUs and a much, much larger address space.

Indeed, this is what most people think of when they hear about 64-bit computing - raising the limit from 4GB to a mind numbingly large 16EB, or exabytes. That's over 16 billion gigabytes.

But 64-bit computing is about much more than addressable memory, it's an architecture change that requires software to change as well, and this brings its own advantages and disadvantages.

What's in a word?
This is actually a very good question when it comes to computing - your CPU thinks in chunks of data known as a word, which roughly approximates to the largest block of data it can process at one time. Traditionally a word is 16-bits, born from previous architectures and carried over as technology progressed. A double word, or dword, is 32-bits and this is the fundamental word size for a 32-bit processor. Data, instructions, registers, pointers and addresses are all managed in 32-bit dword sized blocks. Internally the processor can't cope with anything larger unless it splits them up into 32-bit sized chunks, which takes more time.

And that's where 64-bit CPUs have an advantage. As the name implies, a 64-bit CPU can use a word size 64-bits wide, otherwise known as a quad-word or qword. Data, instructions, registers, pointers and addresses can all be 64-bits in size, manipulating or performing instructions on more information at a time. Indeed, one of the key features of a 64-bit CPU is its ability to sometimes process in one clock cycle what would take a 32-bit CPU two.

Which, on the surface, would seem to imply that a 64-bit CPU is twice as fast as a 32-bit one, but a doubling of the word size is not equivalent to a faster processor. In fact in some cases, it can even be slower.

It all comes down to the data and what the CPU needs to do with it. Software that performs complex maths - especially compression, encryption, and encoding - can usually leverage the larger word sizes to process more data at once. And here, most certainly, there can be considerable performance gains. Multimedia instructions (think SSE et al) can also perform faster, as these usually operate on 128-bit data and, again, a 64-bit CPU can process these faster.

However for most tasks you do, at least on a desktop computer, the extra word size has little if any impact at all. Browsing the web or tapping away in Word isn't going to require the use of large word sizes. Which is why when you installed 64-bit Windows or Linux on your first 64-bit CPU everything seemed just about the same, because it was.

There is an exception, and it's more a indirect benefit than a function of 64-bit architecture - Windows and most Linux distributions are compiled for the lowest common denominator to ensure they can run on as many platforms as possible. This means not tailoring the binaries for particular CPUs, but using a generic base i386 instruction set. This usually means some of the advanced CPU features developed over the years for accelerating performance can't be taken advantage of, as different CPUs support different functions.

The same is also true for 64-bit versions of Windows and Linux, with the exception that even the base 64-bit processor brings with it all those years of development from its 32-bit heritage, and certain functions are standard across all 64-bit CPUs - such as SSE, MMX, register sizes and cache. Hence even when operating systems are compiled for the lowest common denominator for 64-bit platforms, the code is still likely to be better-optimised than the 32-bit versions of the same, the result which may be faster performance.

If you've ever compiled your own Linux kernel, you can see an example of this in the way Linux distributions use the 'Generic x86-64' target over the other 'Opteron' and 'Core2' paths. As a user you can re-compile the kernel and optimise it for a given architecture to further improve performance, which in some cases can provide quite a boost. It would be great to be able to do this with Windows too, but alas Microsoft isn't going to give up the source to users anytime soon.