Nowadays most computers use 32-bit processors (such as the Intel Pentium) running 32-bit operating systems (such as Windows XP, Mac OS, Unix or Linux). Some years back, desktop computers used 8-bit microprocessors (such as the Zilog Z80); then came 16-bit chips (the Intel 8086 and Motorola 68000). These bit numbers describe the length of the instruction word the CPU can handle in a single clock cycle. The next step in this evolution is the 64-bit CPU.
Intel has been the microprocessor industry's 800-pound gorilla from the beginning. The company began 64-bit development in 1991, and the first systems with its 64-bit Itanium CPUs shipped in 2001.
Unfortunately, Intel developers early on opted for an architecture that's completely different from the common x86 (also known as IA-32) standard. The resulting platform has to resort to an inefficient emulation mode to run 32-bit applications.
The industry-leading vendor had stunningly misread what the market wanted, and the lack of true 32-bit compatibility caused the Itanium to languish. Approximately 5.3 million servers were shipped worldwide in 2003, and of those, 4.67 million (87%) had the 32-bit x86 architecture, according to analyst Mark Melanovsky at research firm IDC. Itanium CPUs were in just 19,000 servers.
A breakthrough came in April 2003, when Advanced Micro Devices Inc. in Sunnyvale, Calif., introduced its AMD64 platform and the Opteron series of 64-bit server CPUs. Unlike the Itanium, the Opteron chips could run 32-bit applications quickly and efficiently in addition to handling new 64-bit instructions. AMD's move led to faster, more cost-effective servers that didn't need to wait for the development of 64-bit applications.
AMD followed up the Opteron in September by announcing the Athlon 64 processor family for desktops and mobile computing. In 2003, some 35,000 Opteron-based servers (almost all of them dual-processor) were sold—nearly double the number of Itanium systems.
In response, Intel announced in February that within a few months it would ship new versions of its Xeon server CPUs (code-named Nocona and Prescott) that could handle 64-bit applications and operating systems. The new capability is being called Intel Extended Memory 64 Technology.
Analysts note, however, that the new Xeons aren't expected to offer the integrated memory controllers or HyperTransport links (a chip-to-chip interconnect technology that operates at memory speeds) of AMD64 chips. Intel's new CPUs are expected to be compatible with AMD's 64-bit instructions.
Why 64 Bits?
There are two major reasons why you might want to use a 64-bit CPU. One is the ability to use massive amounts of memory. Using data in high-speed, solid-state memory is significantly faster than getting it from disk, but there are limits to how much a machine can store in RAM. Running on a 32-bit processor, for example, Windows 2003 Server can handle a maximum of 3GB of RAM, and even Unix systems top out at 4GB. The AMD64 platform can address 4 petabytes of physical memory, and a 64-bit CPU can potentially address up to 18 exabytes.
The second advantage of the 64-bit chip is its ability to handle larger floating-point numbers, which are often used in scientific and engineering calculations. While 32-bit processors can only handle floating-point calculations with values up to 232 (approximately 4.29 billion) unless they resort to software emulation, 64-bit chips can directly use numbers up to 264 (about 18.45 billion billion).
Seattle-based Cray Inc. is building a massively parallel processing supercomputer, nicknamed Thor's Hammer, for weapons research by the National Nuclear Security Administration at the Sandia National Laboratory in Albuquerque. The $90 million machine, scheduled for installation this summer, will use 10,368 clustered Opteron CPUs along with 240TB of disk storage and 10TB of high-speed RAM inside its 108 compute node cabinets.
Although a 64-bit CPU can handle twice as much information at a time as a 32-bit CPU can, it won't be twice as fast—programs won't start in half the time, for example. As an end user, you wouldn't notice a difference in speed most of the time. The difference comes into play with harder-working servers that may have to deal with hundreds or thousands of users, storage areas and process streams simultaneously. In such cases, using CPUs with bigger pipes may mean you need fewer servers overall. That translates into added long-term efficiency.
Other 64-Bit CPUs
There are many other 64-bit CPUs. For example, 64-bit Reduced Instruction Set Computing (RISC) CPUs include the UltraSparc family from Sun Microsystems Inc. and, most recently, the IBM PowerPC 970 (which Apple Computer Inc. calls the G5). Others include Hewlett-Packard Co.'s PA-RISC family, processors developed by MIPS Technologies Inc. and its licensees—including Toshiba Corp., Silicon Graphics Inc. and Digital Equipment Corp. (now part of HP), which pioneered the now-abandoned Alpha architecture. In 2003, Melanovsky says, servers using those processors accounted for 9.8% of the server market.
A Look at Six Microprocessors
>Intel Itanium 2 | Intel Xeon MP (32 bit) | AMD Opteron | AMD Athlon 64 FX | Sun UltraSPARC III | IBM PowerPC 970 FX | |
---|---|---|---|---|---|---|
Primary use | Servers | Servers | Servers | Desktops | Servers | Servers, workstations |
Maximum clock speed | 1.5 GHz | 3 GHz | 2.2 GHz | 2.4 GHz | 1.2 GHz | 2 GHz |
Maximum RAM supported | 64GB | 4GB | 8GB | 8GB | 8TB | 8GB |
Run native 32-bit apps? | No | Yes | Yes | Yes | No | Yes |
Level 1 cache | 32KB | 8KB | 128KB | 128KB | 96KB | 96KB |
Level 2 cache | 266KB | 512KB | 1MB | 1MB | 8MB | 512KB |
Level 3 cache | 6MB | 4MB | None | None | None | None |
Internal bandwidth | 6.4Gbit/ sec. |
4.8Gbit/ sec. |
Not available | 6.4Gbit/ sec. |
2.4Gbit/ sec. |
Not available |
See additional Computerworld QuickStudies