Pentium Redux

Last week’s business article (TidBITS-248) drew several questions and lots of great mail. This article responds to some of the issue raised.

Quantities — Last week I stated with undue confidence that there were about four times as many Power Macs in existence as Pentium-based systems. I based this statement in part on information that says Power Mac sales are still outrunning Pentium sales (at least through the dealer channel – direct sales are harder to verify), but that Pentium sales are ramping up quickly. Pentium sales are definitely slower than Intel expected, and weren’t helped by the PCI glitch Intel suffered earlier this summer.

Since last week, I’ve done considerable research to uncover trustworthy numbers on this topic. An Apple technical briefing in August first raised the idea with numbers similar to what I quoted. Unfortunately, my best information comes from Pythaeus, who saw specific research from market research firms like IDC and InfoCorp, who don’t give data to non-paying customers. Hence, I can’t provide many references that people can cite to show their PC friends.

However, in the Oct-94 Macworld (pg. 41), Patrick McKenna writes, "analysts report Pentium sales lagging behind those of the Power Mac," and mentions the speculation of one analyst that many PC users recently switched to 66 MHz 486 systems, resulting in less need for another performance boost.

The current problem is that no one believes Pentium sales will continue to lag behind Power Mac sales for long, and the companies who use the PowerPC chip don’t want to have their promotion of the current situation thrown back at them later.

Native vs. Optimized — Another common comment concerned the issue of "native Pentium applications." First, at least three applications have been optimized for the Pentium to date. None of the current three are mainstream applications, but rumor has it that Adobe is rewriting Photoshop for Pentium-optimized compilation.

Second, as Eric Schlegel of Microsoft mentioned in email, the term "native" isn’t applicable, since the Pentium does include x86 code and is thus not "emulating" when running older applications. I think that this point clouds the real situation. I see little point in using a Pentium if you only end up running it as a faster clock speed 386. Still, it is more correct to ask "how many Pentium-optimized Win32 applications have you seen?"

Few applications use the 486 as anything but a fast 386 either, though that’s partly because the main advantage of the 486 is not new instructions (there aren’t many), but instead reduced clock cycles for many instructions and the addition of an on-chip cache (this enabled the core CPU speed to increase with less concern about the external bus clocking).

This is similar to the situation with the 286, since most DOS applications were written to work with the 8088 for years after the release of the 286. In other words, few programs took advantage of the extra capabilities of the 286, much less the incipient 386. Crudely speaking, each successive chip has had more instructions without those capabilities being used by application software (though other features, like caches, are used, of course). This predicament was one of the major motivations behind OS/2 – to force software into the new age of the 386.

Fundamentally, these are the issues:

• You can discount Intel’s Pentium SPECmark ratings by 10 to 20 percent to account for the fact that real software isn’t optimized for the Pentium. I think this is significant.
• Intel has lost control of the x86 standard. Now that AMD and Cyrix have come out with their own versions of "Pentium-like" chips (which aren’t clones), the problem compounds. Do you recompile for each chip? Or none of them? So far the answer has generally been none. This has long term ramifications for the Intel world.
• Intel can’t move forward without recompiling. By that I mean that they can’t take full advantage of new capabilities and full performance without recompilation for each generation of chip. That’s not to say that recompilation won’t be required for the PowerPC 604 or 620 to make best use of those chips, but Apple seems to have enough control over developers to ensure that recompilation takes place if necessary.

The Industry Says… — I’m not alone in thinking the optimized applications issue is a significant one for Intel. Consider the following quotes regarding the performance of the 486DX4 versus the Pentium. In 07-Sep-94 NewsBytes, Steve Gold wrote that the Apricot 486DX4 PC "is faster than more than half the Pentium-based PCs available from the likes of Compaq, Dell, Gateway, IBM, NEC, and AT&T." This information came from a "real world" test by BAPC (Business Applications Performance Corporation).

Several Computer Shopper articles make much the same point, and the Sep-94 issue of PC World says, "NEC’s DX4-100 outperforms six Pentium-60s and -66s. Ambra’s 486DX2-66 outperforms two Pentium-60s. At one time a PC’s processor gave buyers a rough but accurate guide to a system’s price/performance, but that simple indicator no longer works. Performance levels and prices, too, are all over the map."

Finally, in the Sep-94 issue of Computer Shopper, microprocessor guru Michael Slater writes, "semiconductor economics and typical PC user needs favor the DX4. The DX4 is significantly cheaper to make than the Pentium, and on integer programs that have not been optimized for the Pentium – which includes the vast majority of software in use today – it provides comparable performance. The cost of building a system around the processor is also lower."