為什麽AMD輸了64-位 戰爭：過低地估計了OS 和 SOFTWARE 的 工作量
A DECADE AGO AMD released the first Opteron processor and with it the first 64-bit x86 processor.
AMD's lavish New York launch for the Opteron processor was far more than a product launch, it was AMD showing it could reproduce the success of the Athlon K7 chip and take the fight to Intel by developing ground breaking new features and not just one-upping its rival on some benchmarks. The firm's Opteron processor brought 64-bit computing to the commodity x86 chip market, along with an on-die memory controller and the Hypertransport bus all in one product.
Back in 2003 AMD's workstation and server processor range barely existed. The firm dabbled in two socket systems with the Athlon MP, and while second tier system builders such as Appro embraced it, AMD really didn't have a server chip that could entice the high volume server manufacturers - HP, Dell and IBM - to use its kit, something that Opteron should have changed.
AMD's Opteron chip was a 64-bit processor at a time when 64-bit processors were consigned to exotic silicon such as IBM's PowerPC, Sun's UltraSPARC and Intel's Itanium. It was Intel's Itanium chip with its IA-64 architecture that served as a lightning rod to spur AMD's designers not to simply go their own way but to build on top of the existing 32-bit x86 architecture.
In effect AMD's AMD64 was positioned by the firm as being 64-bit extensions rather than a completely new architecture, with the firm touting existing 32-bit compatibility as a big selling point. AMD's decision, even at the time, was sound reasoning and the benefit of hindsight further justifies its decision not to drop 32-bit x86 support.
Intel's Itanium had already been out for the best part of two years, with a radical instruction set that promised a lot on paper but simply didn't deliver at the silicon level. The firm's projected sales never materialised and even in 2003 Itanium looked to be little more than an anchor around Intel's neck, and while Intel has had the resources to drag that anchor for decades, AMD realised that what customers really wanted was a fast 32-bit x86 chip with the promise of 64-bit capability for certain niche applications.
AMD's AMD64 architecture brought the usual band of instructions, registers and at the silicon level, more cache than previous AMD processors. The firm sold the primary advantage of 64-bit capability as being able to address more than 4GB of RAM without the need for physical address extensions. While companies like Dell now configure £280 laptops with 4GB of RAM, back in 2003 it would be rare to find a two socket server with more than 8GB of RAM, because there was no point in configuring more memory if the CPUs couldn't address it.
The problem with AMD's 64-bit architecture is that the firm underestimated the work it needed to do in order to get developers to port operating systems and software to make use of its architecture. Operating system developers such as Microsoft and the various Linux distributions lagged behind at launch, while in the case of Linux the kernel supported the AMD64 architecture, but Linux distributions initially did a poor job of providing supporting software packages and libraries.
However it was Microsoft that really left AMD up a creek, especially in the workstation market. Although the firm produced a Windows XP 64-bit edition, the operating system could be classed as little more than in beta state, with driver support being best described as poor.
Inadvertently, AMD's Opteron chip and its AMD64 architecture helped x86 chips take on established server chip vendors such as IBM and Sun on performance, features and software. The irony is that it was Intel that took advantage of AMD's trailblazing AMD64 architecture and ran away with the server market.
Perhaps the greatest accolade for AMD was Intel's decision to license AMD64 and re-brand it as x86-64 on its Xeon processors. Intel's decision to go with AMD's 64-bit technology instead of its own IA-64 architecture served as an admission that Itanium and its complex instruction set were unsuccessful in convincing the market to develop for its own products.
AMD's Opteron processor launch was a major event because it introduced three key technologies to a single product at one time. The firm's AMD64 extensions was clearly the headline technology but the firm also moved the memory controller off the Northbridge and onto the same silicon die as the processor core and introduced the Hypertransport bus.
With AMD moving the memory controller on-die latency was significantly reduced and it also removed complexity for motherboard manufacturers, another technology that Intel would adopt many years later. AMD also did away with the use of fully-buffered DIMMs, something that considerably reduced overall system power consumption.
It should noted that AMD's decision to put the memory controller on-die meant that the processor was tied to a specific memory subsystem. This meant the firm had to bet on a memory technology and hope it stayed competitive for the lifetime of the architecture, which in the case of server chips would be in the region of three years to be cost effective for customers.
AMD's work within the Hypertransport consortium yielded a high bandwidth bus that it used to both as a front-side bus replacement and, in multi-socket Opteron systems, to link multiple processors. With Hypertransport, AMD's partners were able to build exotic four and eight socket servers at a time when processors were still single-core units and display scaling that systems shod with Intel's Xeon processors could only dream of.
Given Intel's dominant server market position, AMD's Opteron processor could be classed as a business failure. However to judge AMD's Opteron chip simply on sales figures would be to do the firm a great disservice, as it is hard to find another launch that packed so many substantial innovations into one product.
Out of the three major technologies introduced with the Opteron processor, it is relatively low cost 64-bit computing that consumers at all levels and including rivals are enjoying today. If AMD had managed to get software and server vendors to push Opteron servers harder, then it is very possible that the firm wouldn't be in the financial situation it finds itself in today.