Intel builds in more built in obsolescence

FOR QUITE SOME TIME, Intel's had a reputation for being the "business" chipset to use if you wanted to run a Pentium 4, Xeon, or were the other purchaser of an Itanic system. Its products are regarded as very stable solutions, and many companies find comfort in the idea of buying a chipset built by the same company who designed the CPU it runs in, based on the (logical) premise that no one knows how to make the two work better than the people responsible for both. With the exception of i820 and the Rambus disaster, in fact, Intel has produced some of the best chipsets ever released for the various Pentium platform(s), and the company does deserve recognition for that fact.

While consumers have definitely benefited from Intel's high design standards, Chipzilla has another tendency which has been less-than-appreciated by customers-and-manufacturers alike—their chipsets and sockets practically define the term "planned obsolescence", forcing manufacturers to continually switch or update motherboard designs, RAM types, or socket form factors. End users, meanwhile, are faced with a dizzying array of chipset choices, with a new army of them marching out quickly, each "better" than the last.

Oddly, however, reviewers have been remarkably slow to pick up on this trend. When the P4 first launched on its Socket 423 platform, the reaction of the enthusiast community was decidedly negative. Readers were, in fact, repeatedly warned away from purchasing any P4 on the platform, and advised to wait for Socket 478. Of course, the technical community's general hatred of Rambus, the P4's high price, and its decidedly lackluster performance didn't help matters any, but the fact remains, many enthusiasts rejected the P4's Socket 423 platform on the ground's that it was "not-upgradable." The P4, after all, had launched at 1.5 GHz, and it was known that the Socket 423 platform would top out around the 2 GHz mark—so people were advised to stay well away.

So, when Socket 478 eventually comes out we see some significantly positive press for the new format—specifically, how it's the "real" P4 format to buy, its upgrade potential, longevity, etc.

Only one problem with that….it isn't true. Let's look at the math on Socket 478, just over a year after its release.

Socket 423 debuts at 1.5 GHz and ends at 2 GHz, for a 500 MHz increase-above-maximum, or a 25% gain. Socket 478 debuts with a 400 MHz FSB at 2 GHz….and terminates at 2.6 GHz on a 400 MHz FSB. 23% gain for that platform there. Socket 478 moves on to a 533 MHz FSB, but according to Intel, many of these 533 MHz-capable chipsets don't have support for Hyper Threading, or, even if that can be added via BIOS, do not officially support the new CPU's heat and voltage requirements. So, i845E heads to the dumpster, and i845PE comes in. The i845E ends up a real loser chipset in this regard—it lacks DDR333 support, and it tops out at 2.8 GHz—barely higher than the i845D chipset it replaced.

So, now we have the i845PE, which adds support for the 3.06 GHz P4 and Hyper Threading. Only problem is, the writing was on the wall for i845PE almost as soon as it appeared. Not only is Intel transitioning to dual-DDR-based platforms, but the P4 itself won't be on a 533 MHz bus much longer. In the second quarter of this year Intel is expected to release a 3.2 GHz P4 running on an 800 Mhz FSB—and it's a guarantee they won't want too many 533 MHz CPU's running around. Hammer's HT system runs at 800 MHz, after all, and the 533 MHz P4 won't look very good comparatively to those blue boys in marketing. It's conceivable we'll see the Celeron's jump up to the 533 MHz bus, but given the anemic performance of THAT particular chip, its unlikely to ever be a serious upgrade option for anyone already running a high-end 533-based P4, at least not in its current form.

Various reviews across the 'Net have shown the Celeron 2 GHz overclocked to 3 GHz…and losing benchmarks to Athlons and P4s clocked at half to two-thirds that speed. Given that i845PE will likely check out at the 3.4 or 3.6 GHz mark, max, that leaves the i845PE with a 22% increase in performance, at best—potentially, the number could be as low as 13%, if Intel chooses to break cleanly between the two bus speeds. This would be an unusual move for Chipzilla, but if 533 MHz P4's can't compete against Hammer and 800 MHz P4's can, you can bet the 533MHz flavor will make a very swift exit.

Even Intel's workstation chipsets are given similar treatment. Intel's long awaited, dual-DDR chipset codenamed Granite Bay has a lifespan measured in weeks and lacks support for both DDR333 and a faster FSB. Soon we'll have Springdale and Canterwood which change the goalposts yet again.

Now, given that Intel is in the business of making money (and chipsets are big business) it's a guarantee that Chipzilla wants a profitable chipset business—and that means making lots of chipsets. Fair enough. But there's also such a thing as paying attention to the desires of your customers, and making motherboards upgradeable, an important factor when making a purchase.

Despite the fine words and implications of most computer magazines and hardware technical sites, Socket 478 is no better an upgrade path than Socket 423 ever was. In fact, to date, it's been about as good—or as bad, depending on your point of view. For all the pounding Socket 423 took for its non-upgradeability past 2GHz (often discussed in the most indignant and damning of terms) you'd think those same publications would've pounced on Socket 478 for the same problem and, in fact, pounced on it more fiercely —some of these folk appear to be past masters at reading tea leaves, and capable of reading the most sinister of motivations into a company's decisions.

Though I've purposefully avoided mention of the other CPU company, the upgrade-path situation with AMD-based solutions is worth examining in brief. For starters, KT133 buyers are largely out-of-luck—the 200 MHz FSB option ended with the Athlon 1.4 GHz, leaving you with a CPU option 28.5% faster than the one you could've bought when the platform debuted. Not a bad solution, but you won't be breaking that 1.4 GHz barrier.

Once we make the jump to KT133A, however, it's a whole new ballgame. If you own a KT133A motherboard, you have the option (depending on your board manufacturer) of upgrading from an Athlon 1.2 GHz all the way to an Athlon 2600+. If we accept AMD's model numbers as comparative, that's a massive 54% jump. Even if we drop to a straight MHz-to-MHz comparison, the 2600+ is a 44% jump above the Athlon 1.2 GHz. Of course, KT133A boards lack DDR support as well as other, newer features that have appeared in the intervening period—but strictly comparing CPU to CPU, they can still compete with the newest and most powerful solutions AMD has to offer. If your KT133A board can overclock to a 166 MHz FSB (and many can), you've got even more options, with the 2800+ and its followers available to you.

It's quite an accomplishment that AMD was able to engineer the Socket A platform so well that it could survive through a 300% increase in processor speed, a near-doubling of its FSB (with a potential 400 MHz bus coming before the platform is phased out), adequately support the increased power and voltage requirements of these new technologies, and do so all while supporting at least four different revisions of the CPU architecture itself over the course of its lifetime. Perhaps what makes all this so surprising, is that Intel can't seem to keep its chipset capabilities synched to its RAM speeds or CPU power for more than about four months. Business reality and money making is one thing, but it's more than a little pathetic when the so-called "little guy" in the market designs a form-factor that, in terms of longevity, makes the market giant's look like something built from Lincoln Logs and Tinker Toys.

The bottom line is that Intel makes good chipsets — maybe even great chipsets. It's even possible that they, as the manufacturers of the Pentium 4, make the most stable and compatible P4 chipsets you can purchase. If these are the most important features to you, that's a great thing. In terms of upgrade potential, however, Intel platforms suck wind. Some of the third party chipsets offer slightly better potential, but only slightly.

Hopefully, next time, the technical community as a whole, both paper and web, will recognise that fact — and not hammer one platform for its lack of an upgrade path, while blithely failing to recognize an identical failure in that platform's successor, simply because the numbers behind the word "Socket" haven't changed.

Aren't we supposed to know better? In fact, isn't it precisely for this reason that review sites were formed — so that consumers wouldn't be led down the garden path by marketing departments with their own agendas to pursue? µ