64 bits of heaven

The new "big news" in the desktop market this year is of course that we are finally moving from 32 bit to 64 bit processors. It has been on the cards for some time - Intel announced the stupid sounding Itanium project with HP years ago but it is really only in the last year that it has really arrived in the desktop market.

Apple are currently crowing about how they have the first 64 bit desktop machine, the imaginatively named 'G5', although history students may remember a product line powered by the DEC Alpha before it was willfully destroyed. Intel are busy touting their upcoming 64 bit offerings but are in a bit of a panic since the upstarts at AMD have stolen their limelight with the Athlon64.

In the Amiga market, we have waited almost ten years to move to the 32 bit PPC processors, the G3 and G4, but in typical fashion, before they are even out, we are hearing people clamor for the move to the 64 bit G5 - ironic since Amigans spent many years defending elegance over brute force, when our old custom chipset and integrated motherboards in the classic could still beat the steam roller approach of the x86 boards.

To control something you first have to understand it, and so before we get all excited about the G5 for the AmigaOS4 family (and potentially any 64 bit processor for AmigaOS5), let's first make sure we all understand the "itzy bitzy'' issue.

Computers are effectively number crunchers. They move numbers around, manipulate them, check them and use them to map out digital space. Humans use decimal because we started off counting on our fingers (many of us still do) but computers don't have fingers. They have transistors, which can be in two states, on or off. This is perfect to support the binary number system, which uses two values 0 and 1.

To count up to big numbers in decimal we have units, tens, hundreds and thousands, columns in a big number that go up by a factor of ten each time you move one column to the left. Binary is the same. Big binary numbers have columns as well but being binary or base 2, instead of decimal or base 10, each column goes up by a magnitude of 2, thus we have units, twos, fours, eights, sixteens and so on.

The first mainstream processors were the 8 bit processors, which meant that they had 8 columns in their binary numbers, meaning a maximum unsigned number of 255 (11111111). Obviously there isn't a lot of mathematics that you can do in a number range of 0-255. To get around this, sums had to be broken up into chunks which made for slow performance. The move to 16 bit and then 32 bit processing made a huge difference in this respect because it meant that bigger sums could be done in smaller steps, and in most cases just a single step. The maximum 32-bit number in decimal is 4,294,967,295 (2^32-1), which is big enough for almost all standard mathematical operations. The maximum number for a 64-bit processor is so huge that there is an argument that it may not be worth the move from 32 to 64 bits just for calculations.

However, this isn't the main reason for wanting to move to 64 bit processors. As I mentioned before, processors also use numbers to map out their digital space, giving an address to each byte (which is still only 8 bits). The biggest 32-bit number can only map out 4,294,967,296 bytes, which comes out to just over 4 Gigabytes. In other words, if you think of the processor as a robot doing lots of work inside a closed room, that room is limited to a certain size, its address space and it can only access what is in that room. For a 32-bit processor with a 32-bit address space, that means the room can only be just over 4 gigabytes in size.

In the past, when we were lucky to have 64MB of RAM in our computers, no one cared, but as is always the case, what seems a distant dream one year rapidly becomes a restrictive barrier. Computers are now shipping with 512MB and even 1GB as standard. It won't be too long before we smack up against the 4 GB limit.

So what though? We store our content on big hard drives and don't need to bring it into memory. That may have been true in the past but digital content is becoming bigger. Why play a game at 640*480 when you can play it at 6400*4800? Why have 8 bit color images when you can have 32 bit images? Why have to stream audio from a physical drive when you can just keep it in memory and do it from there? Why can't I have instant switching between my 20 running tasks rather than having the hard drive turn into a Geiger counter as it swaps data in and out? Just like an increasing salary, requirements will grow to fill up the new capabilities - one reason why the increase in computer performance seems to be out of step with our sense of increase in computer experience.

The biggest advantage in the move to 64 bit processors is that this 4 gigabyte address space for memory is suddenly exploded apart, which means more memory on the computer and more content being stored in main memory where it can be accessed many times faster than it can be when it has to be pulled from a hard drive.

Of course just because this is the maximum size of the address space doesn't mean that it has to be used and indeed the current Athlon64 only offers a 40-bit hardware bus, restricting the address space size to 1,024 GB. There are many reasons for this, but the flexibility is there to open up those other 24 bits in a future revision.

The operating system and the application software still has to be written to take advantage of the 64-bit processor. Apple may claim a 64 bit desktop computer but Panther 10.3 is still only a 32 bit OS. Sure developers can write their own custom code to take advantage of the new processor capabilities but, like any extension, this will remain the exception until the OS itself supports, encapsulates and presents these new capabilities; then they will become the norm. Most of the 64 bit processors (G5, Athlon, Opteron) can run 32 bit at full speed, which is how Apple can make the claims they do.

One of the exciting advantages of the 64-bit address space however lies in a possible move to what is called orthogonal persistence. Consider the current case, where we have a 4GB limit for main memory whilst we have hard drives now pushing 120GB. Traditional operating systems separate the static (persisted) state of content from the dynamic state. Content has to be loaded into memory, each time to different places in memory, with the application code and operating system having to manage the difference between these two states. With that 4GB address limit gone, we can use a 64bit memory management unit to remove this separation. In essence, content doesn't have to be loaded or saved. It is always there, just like in the real world because it is always at the same place in the address space.

In the end, the true advantages of the 64-bit processor lie in the future. At the moment, we have to remember that a processor is just one part of a computer and there are many places where performance can be enhanced. PCI Express is one exciting development whilst integrating the Northbridge into the processor itself (as in the AMD64 range) is another. Of course the increase in the number of instructions that can be done per second as well as the deeper issues of mapping real life application to machine code are also very important to the overall equation.

Amigans shouldn't be too eager to get onto a G5 processor. For a year or so it will just be an expensive and underused poser item. It is more important that we do the real work of completing the move from the 68k to the PPC and setting up a foundation for the new Amiga Generation 2 technology. Remember, the other platforms HAVE to have these expensive processors to offset their Operating Systems. We don't.