Plug Into Processors

SPARC and Alpha processors.

The argument is well-known: PCs have become so fast that not even Microsoft can slow them down anymore. Customers don't necessarily have to buy PCs at the top of the performance curve. And, with the current economic uncertainty, there's been a great deal of reluctance to replace systems as they reach three or more years of use.

A recent report from International Data Corp. (IDC) doesn't cast a rosy future. The company estimates that PC sales will rise only 1.1 percent this year and that the holidays will be disappointing. That's down from its June expectations of 4.7 percent growth for the year.

Within the desktop-PC market, the real shrinkage has been on the high end as buyers favor value (read: cheaper) platforms. What qualifies as a value computer in 2002 is a 1.2-GHz Celeron with 128 MB of SDRAM and 20 GB to 40 GB of storage. Most high-end servers didn't have that kind of power just five years ago. That has made the 2.4-GHz desktop with 512 MB of memory, and bells and whistles like CD-RW, Gigabit Ethernet, wireless networking and so forth,a tougher sell.

"It all goes back to the theory there's more than enough power to do what we want," says Dean McCarron, an analyst with Mercury Research, a semiconductor market research firm in Phoenix. "The burdens e-mail and spreadsheets place on the system haven't changed, so as time has gone on, the less expensive systems have offered that capability."

"The problem here isn't a hardware problem, other than in the sense the hardware is too good," McCarron says. "What we need to spur higher-end hardware sales is something that uses it," he says. "Traditionally, we've required system shifts, where Microsoft would come out with a new OS that would bring a system to its knees, and everyone would go out and upgrade. Microsoft, to its credit, did a good job with Windows XP and didn't break everything."

What's needed is a next-generation OS that uses 3D graphics, as Microsoft hinted at with Chrome several years ago before shelving it, or some unforeseen killer app that drives upgrades. Of course, Intel thinks otherwise and wants the market to think it should upgrade now.

"People really still need performance, and they need it in the CPU and every subsystem," says Gordon Dolfie, director of product marketing for the Intel reseller products group. "I think the drag on productivity from people's older systems will be what compels people to upgrade, as soon as their company sees profitability growing again."

That's why Intel and AMD are putting more and more effort into the subsystems, and not just the CPUs. Today, CPUs are running at 2 GHz, and beyond, but still have the 10-year-old, 33-MHz PCI bus moving things around internally. Memory runs at 266 MHz at best,and more often at 133 MHz. The IDE hard-disk bus, first introduced in 1986, has updated to get to 100-Mbps transfer speeds, which still isn't enough for some high-performance applications.

That's why the next few years will see updates to PCs that haven't been seen in ages. Intel has been dying to bury the legacy I/O for some time. It wants to see an end to COM, PS/2, parallel ports and even the old PCI bus. It favors USB, 1394 FireWire and PCI Express, formerly 3GIO.

PCI Express will be backward-compatible with PCI cards, but it will start out with 2.5 Gbps of throughput, well beyond the capacity of PCI, and will eventually reach 12 Gbps of throughput. The cards will be hot-swappable, something unthinkable in today's PCs.

Intel is also working on a new disk interface, called Serial ATA, which will considerably crank up storage performance. The current ATA-100 interface is parallel, which means a whole new cable for Serial ATA, in addition to much faster performance. Instead of the flat, wide 80-pin cables we've come to know (and hate), Serial ATA drives will use a small plug similar to the PS/2 plugs used in today's mouse and keyboards and offer throughput ranging from 200 Mbps up to 1.5 Gbps.

AMD has cast its lot with HyperTransport, a new high-speed, high-performance open-interface link that can move data internally at up to 12.8 Gbps. Two HyperTransport products are on the market,one you may be selling, and one that may be in your kid's room. Santa Clara, Calif.-based nVidia, Microsoft's partner in developing its Xbox video-game console, came up with a chipset based on HyperTransport, which is used in the Xbox. Several months after the Xbox shipped, nVidia entered the PC chipset market to compete against Via Technologies with a product called nForce.

The nForce chipset is currently the only HyperTransport product on the market, though that will change when AMD ships its 64-bit CPUs next year. HyperTransport will run 50 times faster than PCI, providing 6.4 Gbps of data throughput initially, compared with the 133 Mbps of PCI. HyperTransport will replace the EV6 bus used in the Athlon, so AMD's 64-bit chips, code-named ClawHammer, will have a 1.6-GHz bus,a huge leap over the 266-MHz bus on the Athlon.

Another development is in memory. Rambus has pretty much been a failure; even Intel is giving up on it. The new initiative, it seems, is DDR-II, the second-generation of double data rate memory. DDR-II RAM will run between 400 MHz and 667 MHz, much faster than the 200 MHz to 333 MHz of today's memory, with lower latency for faster performance and less power consumption. Both Intel and AMD are expected to support DDR-II.

The combination of these new technologies,64-bit, DDR-II, Serial ATA and PCI Express,is what's needed to create high-end enterprise servers that can replace the hundreds of single servers in deployment now, says Carl Palmieri, CEO of Computer Resolutions, a VAR in Bridgeport, Conn.

64-Bit Battles Ahead

When AMD introduces its 64-bit chips next year, things could get interesting. Marketing arguments aside, there's an empirical, technological issue to consider between 32-bit and 64-bit processors.

A 32-bit processor can only address a maximum of 4 GB of memory. Even a year ago, the idea that could become an issue was probably ludicrous to most, but it's quickly becoming a concern, particularly on the server side where, increasingly, applications are being run on the server.

While the chip can address 4 GB, AMD says its own internal tests of Microsoft's 32-bit OSs, even Windows XP Server, seem to max out at around 3 GB of memory, so the OSs aren't even making the most of the limits.

A 64-bit chip can address 16 exabytes of memory. To put that in perspective, an exabyte is 1,000 petabytes, a petabyte is 1,000 terabytes and a terabyte is 1,000 gigabytes. Put another way, 64-bit has four million times the capacity of 32-bit.

Sounds impressive, but vendors still need to convince some VARs. Leo Rodriguez, head of strategic services for DataVox Technologies, an IT services and solution provider in New York, thinks 64-bit technology will eventually become significant, but it's hard to tell which chip vendor has the winning strategy.

"Sixty-four bit has the promise of the horsepower required, but I can't say it's a winner until it's proved," Rodriguez says. "It's going to really depend on the leverage they have with their distributor channels and who builds them into their systems."

AMD will have a whole family of 64-bit chips based on its Hammer family. The server products are called Opteron, while the desktop products are called 64-bit Athlon, though that may change. The 64-bit chip is based on the x86-64 extensions that AMD designed, which keep backward compatibility with 32-bit x86 applications.

So far, Intel's only stated 64-bit strategy is on the server, with its Itanium processor. It's reported to have a 64-bit desktop chip, code-named Yamhill, sitting in the labs and ready to roll if AMD does, indeed, get some traction, but the company won't confirm or deny the rumor.

AMD's desktop product, ClawHammer, is expected to ship to OEMs in the first quarter of 2003, with systems shipping in late Q1 or early Q2. Opteron products are on track for availability in the first half of 2003, and the mobile product is on track for mid-2003.

AMD stresses that the Hammer family is both 64-bit capable and 32-bit capable, and even in 32-bit mode, the chips will still run faster than any pure 32-bit product. The message to customers is they have a forward migration path.

"There will be some people who buy Opteron technology to run today's apps and that's all they will use it for, but others will run today's apps and will seamlessly be able to move to 64-bit apps, at no additional cost, in the future," says Robert Fuller, director of field marketing for AMD.

Analysts think that could help AMD break out from its also-ran status, if the stars align properly.

"AMD's decision, given where AMD is in the marketplace, makes a lot of sense," says Nathan Brookwood, president of Saratoga, Calif.-based Insight64, a semiconductor market research firm.

Still, 64-bit won't take off until 2004, Palmieri predicts, because of the lack of an OS. "Next year, 64-bit is more academic and flash," Palmieri says. "It's not going to be a big thing because the software is not there."

But Intel has the resources and the position in the industry where it could come out with Itanium, a whole new architecture that required a lot of work on the parts of other industry participants, and has a shot at getting everyone to rewrite their software and change the way they are thinking, Brookwood says. AMD doesn't have that kind of clout.

"AMD has no place in the server market, and for it to come out of nowhere and say 'we're going to be the new high-end guys' would have fallen on deaf ears," Brookwood says. "So AMD's product makes some sense on the desktop, at least in a few years, and at the low end of the server market."

The big issue is getting support. AMD has earned a respectful place in the market with Athlon after years on the tail end of the performance curve, and has managed to pick up some wins in the consumer, and even business markets. It recently secured a deal with HP for the Athlon to power Compaq's D315 business computers, a rare but much-needed corporate win.

But no OS vendor has committed to Hammer. Microsoft and a number of Linux vendors, including Red Hat and SuSe, have agreed to explore the possibility of porting their products to AMD's 64-bit product, but have made no firm commitments.

What's Ahead?

What will the future PC look like? Most likely, the PC of 2003 will finally be legacy-free, without COM, PS/2 or printer ports, using USB 2.0, 1394 FireWire, 333-MHz DDR memory, Serial ATA drives, processors faster than 3 GHz and processor buses ranging from 533 MHz to 800 MHz. Servers will use the PCI-X specification, while desktops and servers will come with PCI Express, and we'll begin to see more Gigabit Ethernet connections. Wireless Ethernet compatibility will become more ubiquitous in all systems, especially laptops.

AMD admits it hasn't been as active in communicating with VARs as rival Intel and wants to make up for that with the rollout of Hammer. "VARs have been saying AMD needs to get out to the marketplace more," says John Crank, Athlon brand manager for AMD. "The volume is going to be turned up dramatically." That means a global campaign and targeting sectors where Athlon has been making gains, namely at the government level.

Intel also plans to make a lot of noise in 2003, specifically about its HyperThreading technology, which will be introduced in the 3-GHz Pentium 4 and all successive generations of chips, Intel says. Intel will also throw its weight behind DDR-II, Serial ATA, USB 2.0 and FireWire.