Dr. Dobb's Journal April, 2005
When it comes to CPUs, the big news last year was that Intel stumbled—and AMD did not. The result of that is probably a permanent loss of market share for Intel. When the year began, most analysts predicted that AMD would not survive the year. Now it's an entirely different ballgame, with AMD actually leading in high-performance desktop and gaming systems.
The CPU market used to be Intel, then everyone else. It's still Intel on top in sales, but in performance and price/performance, AMD has become a real competitor with enough financial stability to stay the course. All that in a single year.
AMD has also solved most compatibility problems. In AMD's defense, most incompatibilities in the past were not due to the AMD CPU, but to the support chipsets. That problem pretty well ended when NVIDIA jumped into the picture. The combination of AMD CPU, 64-bit processing with backward compatibility, NVIDIA support chipsets, and NVIDIA graphics has brought AMD real dominance in the very high-end gaming market; and NVIDIA's importance in the gaming world ensures that there won't be compatibility problems either. AMD has a solid position among high-end gamers, and that won't easily be lost: The newest game machines are cool, and just about all of them are based on the AMD Athlon 64.
Intel bet heavily on its Prescott chips, which were intended to take them up into the 5 GHz and beyond realm. Prescott extended the 20-stage processing pipeline of its predecessor to an unprecedented 31 stages, but the chip fell far short of its clock-speed targets and the added complexity delivered no advantages on most software. Instead, the fastest Prescott chips weren't much faster than the already good enough Northwood chips they were to replace, while the heat problems were much worse.
At clock speeds below 3.0 GHz, Prescott is actually slower than Northwood. They break even at 3.0, and when you get to 3.4 and 3.6 GHz, Prescott outperforms Northwood dramatically with some software. However, there's a high cost in heat management.
Some numbers: At 3.4 GHz, Northwood dissipates as much as 89 Watts of energy. Dropping the speed to 3.2 saves about 7 Watts. Prescott, at 34 GHz to 38 GHz, dissipates up to 115 Watts—significantly more.
At Chaos Manor, Prescott systems barely faster than the previous Northwoods had real heat problems, even though they were built using the same Antec case/power supply systems as the Northwoods. The moral of this story for system builders is to use Northwood if you can still find the chips, and stay down at 2.8 GHz. Alternatively, use AMD/NVIDIA, where the sweet spot is a Socket 754 nForce3 ASUS K8N motherboard and an under-3-GHz Athlon.
Intel bit the bullet and in 2004 abandoned the development of Tejas, a core that was to feature an even longer pipeline, and is now concentrating on its multiple-core technologies. Given Intel's resources, it's a safe bet the company will come out of this all right. Still, Intel lost considerable ground to AMD, and that loss—potentially billions—will show up in profit statements for years to come. Prescott brought in considerable revenue despite its problems, so the losses aren't instantly obvious; but they're quite real. On the other hand, competition is good for consumers, and for that matter, for the industry, and the combination of Intel's stumble and AMD's flawless move into 64-bit territory has assured that there will be competition for the foreseeable future.
Note that AMD gained mindshare and marketshare by pushing 64-bit technology into the mainstream, while Intel stubbornly kept pushing Itanium as its only answer for 64-bit computing. Intel has now relented, offering 64-bit Xeon and Pentium 4 chips, but it still doesn't recognize that Itanium is dying. As I write this, Microsoft has dropped Windows XP for the Itanium; limited sales and, presumably, support remain at the very highest end, but since no major vendors are shipping Itanium workstations, the question is moot. Microsoft says, "We believe we can best serve our customers' needs with Windows Server 2003 Standard x64 Edition, and Windows XP Professional x64 Edition, respectively."
The moral of this story should be obvious. Microsoft makes a lot of mistakes, but abandoning the legacy market isn't one of them. Intel could have learned from Bill Gates.
Intel never announces anything until ready, but it's pretty clear that it will concentrate on dual- and multiple-core CPUs, and the trend to 64-bit capability is already well established. Look for the dual-core 64-bit CPU to become standard, even though there won't be many 64-bit applications. Honing the 64-bit dual and multiprocessing versions of Windows XP will be a distraction for Microsoft, but it's a safe bet that when Longhorn finally ships, it will have improvements to those features (as well as being able to manage most 32-bit software). It's also a safe bet that Microsoft will have a 64-bit Office suite, if not this year, then next. Whether that will have significant advantages over Office 2003 is another story.
AMD is also developing the dual-core CPU, and it may be a close race with Intel. The key here is heat management. Intel has had the resources to continue development in this realm, and may have an edge on AMD here. On the other hand, the Athlon64 was originally designed to support one or two cores, so AMD may get better results sooner. This is all complicated by both AMD and Intel moving away from the use of GHz as the measure of CPU success. In the practical world, systems that quietly run with lower power dissipation are preferable to systems that can fry eggs and demand you wear noise-canceling headphones when you use them, at least for the things most of us do. Some gamers and mad overclockers will put up with conditions most of us aren't ready for; unfortunately, they often get a lot of attention from marketer types as well as the press.
AMD may have a leg up on delivering multicore designs because of HyperTransport (http://www.hypertransport.org/), a serial interconnect bus that works as well on-chip as off. Two cores can communicate using one on-board HT link, another to a second CPU chip, and still have a third link to the rest of the system.
Because it can afford to offer something for everyone, Intel will have three separate dual-core desktop CPU families in the next two years. First to arrive will be dual-core chips based on Prescott, followed by dual-core processors using the more efficient Pentium M core from its Centrino laptop products. There will also be dual-core versions of the server-derived Extreme Edition chips for the speed freaks among us. The latter will feature raw speed and may require heroic measures for cooling; the other will have dual core chips running no, or not much, faster than the present generation.
The book of the month is Broken: The Troubled Past and Uncertain Future of the FBI, by Richard Gid Powers (Free Press). If this book doesn't scare you silly, nothing will. Among other things, he tells the story of how the FBI arrested Zacarias Moussaoui, a man with no flight experience who paid for access to a 747 flight simulator, whose visa had expired, and who was known by French intelligence to be a member of al Qaeda; then released him without searching his computer because FBI supervisors determined that action against Moussaoui "might appear to unfriendly observers to be racial profiling." Lest you think this a lone incident by one imbecile in a supervisory position, there are plenty of others—enough to show the entire Bureau is permeated with an attitude inimical to security operations. It's entirely possible that the new quick fix Intelligence Act and reorganization won't do much good.
The first computer book of the month is David Pogue's iMovie 4 & iDVD: The Missing Manual (O'Reilly & Associates). Mac users will need this book if they plan to do much with those programs; all the O'Reilly "missing manual" books have the subtitle: "The book that should have been in the box," and that's generally correct.
The second computer book of the month is Susan Snedaker and David H. Bendell's How To Cheat At Managing Microsoft Windows Small Business Server 2003 (Syngress). One of the subtitles is "The Perfect Guide if 'System Administrator' is NOT your primary job function." This ought to make it pretty clear who this book is intended for. It's one of those books you may not need, but if you do need it, you need it very badly.
Finally, while I have recommended Robert Bruce Thompson and Barbara Fritchman Thompson's Building the Perfect PC (O'Reilly & Associates), I find it has never actually been named "computer book of the month." It should be. If you contemplate building a PC—whether to replace your TiVo, hold a LAN party, play high-end games, write your book, or for Aunt Minnie to answer her e-mail—this is the book. If you contemplate building a PC but aren't sure you ought to, this book will tell you what you need to know for a sane decision, as well as walk you through actually building the machine. Fair warning: I wrote the preface for this book. I was paid with a free copy.
The game of the month is still Everquest II, which has better graphics and user interface than Everquest. I have not given up my original Everquest accounts, but my characters there have been languishing while I spend what time I have for such things in EQ II. It's easy to play, easy to get into, and while a bit more fun if you can get up a good group, there's a lot worth doing solo as well. Sony has pulled off another grand coup with this one. I still have some problems with the crafting system, but EQ II has had the most rapid bug and feature updates I have ever seen in an online game. They're listening, and I'm having fun. After considerable thought, I've decided that EQ II is also the "game of the year for 2004," with Medieval: Total War as a close second and Rome: Total War coming in third.
DDJ