64-bit age is upon us
So AMD is launching a brand-new, earth-shattering CPU that's never been seen before. We'll, that's actually quite far from the truth, if we're totally honest. We'll start off with a discussion centering upon the technical prowess and abilities of the new 32/64-bit processors, why they're here and what makes them special. We'll then discuss the FX-51 CPU in isolation and follow that with a look at just how well it performs against standard 32-bit competition, for that's what a 1P Opteron-based CPU is bound to do best at right now. We'll refer to the FX-51 interchangibly as the 'Hammer'. It's derived from the same technology that goes into manufacturing an Opteron 940-pin CPU.
64-bit computing
The single most striking and obvious factor surrounding the new Hammer series of chips is in supporting both 32- and 64-bit computing without having to resort to emulation, which is currently the case with Intel's very own 64-bit Itanium line of CPUs. In essence, what AMD has sought to create is a processor that's capable of satisfying a number of lucrative markets without having to, like Intel, design specific CPUs that considerably eat into a firm's research and development budgets. We've mentioned that the K7-series has worked well for AMD, so why the need to change to a hybrid CPU ?. To answer that, we need to look at exactly which benefit 64-bit CPUs may confer upon a cross-section of users, or, to put it another way, what limitations do the current 32-bit CPUs impose. Let's not forget that the Hammer, Opteron, FX-51, or whatever you wish to call it, is essentially a x86 CPU at heart. Additional registers push that up to 64-bit mode, so it's bound to do well, ceteris paribus, using present 32-bit applications.
Some may argue that a 64-bit processor should be just that. Get rid of the archaic x86 and start anew. Some of the compelling reasons for shifting to a 64-bit system lie in data management. The present x86 CPUs, via the use of a 32-bit system, can address up to 4,294,967,296 Bytes of system memory directly (2 to the power of 32). That's 4GB to the likes of you and me. That's also why you see consumer-level x86-based chipsets sporting no more than 4GB of RAM; the CPUs just can't handle it (excluding the slow virtual memory). 4GB of memory may sound like hell of a lot and Intel may make disparaging remarks with respect to likely consumer-level RAM usage over the next few years, but the 4GB limit will become a limit sooner rather than later. AMD probably expects it to be sooner. The Opteron / FX is touted as a true 64-bit CPU, but it can only address virtual memory to 48 bits and physical memory to 40 bits. That's still plenty for a first-generation hybrid design. The present problem is in DRAM modules themselves. Have you ever looked at the price of 1GB registered sticks?. We'd urge you to sit down and grasp something firmly before you do. High-end consumer-level PCs, through various manufacturers, now ship with 1GB of DDR333 / DDR400 memory as standard. 1.5GB can't be too far off and 4GB+, whilst seeming like a crazy amount to most home users, will be crossed at some time in the near future. Plenty of aspiring Speilbergs own and operate digital camcorders that often output DV footage to 10GB+. Wouldn't it be nice to play around with that from system memory?.
What's 64-bit good for ?
We alluded to the Opteron / FX-51 being a multi-purpose CPU. Take scientific calculations (here's where the 64-bit double-precision floating-point work can be conducted in a single clock cycle), risk modelling, huge database and CAD work. If one has the ability to pump a huge dataset into system memory, memory that isn't handicapped by a size constraint, the overall efficiency is bound to increase. Sure, system memory is significantly slower than on-die storage, but disk storage, be it top-class SCSI, is significantly slower than system RAM. Intel may agree with the above assertions but it maintains that there are other methods of increasing the RAM storage space (increasing RAM to 40-bits addressing, for example) without having to run a 64-bit CPU. Enterprise and server-class users will no doubt be happy that AMD now has the ability to ship 8-way 64-bit CPUs, ones that will support present 32-bit applications natively and quickly. Unix OS's have used 64-bit CPUs for some time now (from HP, IBM and a few others) but they're high-end workstation / server-only solutions. Potential, mid-sized users will not need to wait if they're presently using 32-bit applications. What's more, Microsoft is currently readying a 64-bit Windows XP edition and Linux 64-bit is well on the way, too. Windows 2003 Server already supports the Opterons' NUMA (non-uniform memory access) design.
How has it been done ?
The hybrid nature of the FX-51 / Opteron allows it to run present x86 apps. with a greater efficiency than the current K7-series XP Barton; we'll come to how a little later. What AMD has done, it seems, is to keep a proven design (K7, 32-bit) and add a number of registers (a location on the CPU that's used as temporary storage for calculations) that are only active in 64-bit mode. The x86 XP Barton, for example, contains 8 general-purpose registers (GPRs). This has been increased to 16 on the Opteron series of CPUs. A further 8 128-bit registers (double-precision accuracy, incidentally) have been designated for SIMD (SSE2/XMM) work. These 8, working in both 32- and 64-bit modes, allow AMD's hybrid chips to use SSE2 support, thereby making use of a number of workstation-class applications that currently employ this performance-boosting technology. 'Optimised for Intel' may well be a slogan from the past. Intel won't complain, it'll get licensing costs from AMD. The over-riding question is whether these extra storage registers are enough. Only time will tell.
A multi-purpose CPU needs to be able to meet the exacting demands of a number of target audiences. 64-bit computing, by removing some of the limitations inherent in the present x86-32 designs and thereby appealing to the lucrative workstation and server-class markets, will do little to arouse the spirits of consumers who just wish to play the latest games at hither-to unseen speeds. 64-bit recompilations / compilations of present and future games, though, should make them quicker, by dint of the the extra storage registers on offer and efficiencies gained in recompilation. Think of the FX / Opteron as a jack of all trades. It may not reach the dizzy heights of high-end unix servers or even the 32-way Itanium boxes Intel ship, but it should begin to serve a wide cross-section of the power PC public.
