The time is now right for companies to begin looking into how their applications might be able to take advantage of 64-bit features, such as the ability to address exponentially more memory, according to research firm Gartner. "Enterprises must... begin preparing for the transition to the 64-bit platform," wrote Gartner's John Enck in a recent study.
What is it?
So what is all the fuss about, exactly? A 64-bit chip is one capable of handling chunks of data that are 64 bits wide. The number of locations in memory that a chip can address is determined by how wide these chunks of data are.
Thus 32-bit chips allow for 232 addresses, or 4GB of RAM; 64-bit chips and operating systems vastly increase this, depending on the implementation.Windows XP Professional x64 supports up to 128GB of RAM and 16TB of virtual memory, which could be increased in future versions of Windows. The AMD64 chip architecture can address up to 256TB of memory in its current implementations, though this could go up to 2 exabytes in future versions.
For several decades 4GB of RAM was plenty for most conceivable uses, but by the early 1990s 64-bit CPUs had begun to appear, mainly aimed at particular types of processing, such as video rendering and processing large databases, that benefited from the larger address space. In databases, for example, the ability to load the entire database into virtual memory can mean a huge performance boost. Intel's Itanium tried to target this market, but was hampered by being effectively incompatible with existing 32-bit software.
More recently AMD has pioneered the approach of selling 64-bit chips that include 32-bit addressing, meaning they run 32-bit applications at least as well as 32-bit-only chips. AMD calls its architecture AMD64, and uses it in the Opteron and Athlon 64 CPUs. Intel uses essentially the same architecture in newer chips, calling it EM64T. IBM sells 64-bit chips using the Power architecture, adapted for the PowerPC chips used in newer Apple computers (called G5 by Apple). What these all have in common is commodity pricing, with many ending up on inexpensive desktops.
On the software side, Unix has led the way, with 64-bit implementations appearing more than a decade ago. Linux has had 64-bit versions since 1994 (a port to the Alpha chip) and versions for the newer 64-bit architectures from AMD, Intel and IBM have been around for years. Mac OS X is a hybrid, with some recoding allowing the OS and applications to address more than 4GB of memory.
Real-world benefits
What can companies do with all this? Certain vertical industries have already been using 64-bit Unix — and increasingly Linux — for years, including the government and military, life sciences, manufacturing, research and development, energy and media businesses. For them the advent of commodity 64-bit chips and 64-bit Windows just lowers cost and provides more platform options to choose from.
Certain core Windows services should see big improvements in 64-bit Windows, including Windows Terminal Server (WTS), Internet Information Server (IIS) and Active Directory. The catch is that applications will need to be recoded from the ground up, even needing 64-bit drivers, in order to take advantage of the improvements. Microsoft is giving a number of its flagship products the 64-bit treatment this year, including SQL Server, Visual Studio 2005, Commerce Server 2006, Host Integration Server 2005, Biz Talk Server 2006 and Services for Unix; more are promised for next year as the company gears up for Longhorn.
All this recoding means it will take some time for the Windows ecosystem to become significantly 64-bit enabled. Microsoft admitted as much when it confirmed in April that Longhorn would come in 32-bit, 64-bit and Itanium versions, putting to rest speculation that the OS might be 64-bit only.
Still, companies who want to stay ahead of the curve should begin testing Windows Server x64 this year, Gartner says. The firm recommends testing 64-bit IIS and WTS deployments, to confirm their functionality and performance improvements; testing 32-bit applications on the 64-bit OS to make sure the drivers work and to see which packages could benefit from transitioning to 64-bit; and kicking the tyres on 64-bit applications and drivers.
As a demonstration, Microsoft has migrated Microsoft.com and the MSN Search and Messenger applications to Windows Server 2003 x64, claiming Messenger server performance has improved 10 times since the transition.
The advantages listed so far will apply to a relatively small core of companies willing to buy into specific, often niche 64-bit applications or perhaps recode the software themselves. More controversially, it is claimed that any company can run the same old 32-bit software on 64-bit Windows and see a significant performance improvement. IBM, for example, argues that customers can get immediate benefits from migrating to a 64-bit platform with 32-bit applications.
Some analysts dismiss such talk as nonsense, but it contains kernels of truth. One improvement with 64-bit Windows involves the way user programs are allowed to address memory. In 32-bit Windows, as in some other 32-bit operating systems, a portion of each process' address space is reserved for the use of the OS, which reduces the total amount of memory user programs can address. Under Windows XP, for example, each process can only address 2 or 3GB , even if 4GB of RAM is installed.
Talkback
In your article "The importance of being 64-bit"
So what is all the fuss about, exactly? Technically speaking, a 64-bit chip is one that has integer registers that are 64 bits wide, allowing them to process 64 bits at a time. CPUs store the address of locations in virtual memory in integer registers. This means that the total amount of data the CPU can keep in its working area is determined by how wide these integer registers are.
Which isn't the case for the CPUs you are talking about. The X86 64 bit CPUs still have 32 bit integers. They have 64 bits of address space but the default integer word size is still 32 bits -- that's why it only increased the die size of AMD's original chip by something like 7%. If it really had a 64 bit integer word size the chip would have nearly doubled in size.
It is very sad that the AMD-Linux x86-64 initiative is completely hijacked by WinTel in your article. It is thus IMO a biased and incomplete mis-information.
BTW, to the previous comment, making "default" integer size 64-bit won't actually increase the die-size a bit. The ALU and registers in AMD64 chips are already 64-bit capable. It's the OS / compiler / programming langauge that dictates the 32-bit integer size now (and that is not an unreasonable default).