the red hill cpu guide

To reach its very high design clockspeeds, the Pentium 4 used a very long 20-stage pipeline. For comparison, the P6 pipeline had twelve stages, the Athlon only ten. The advantage of the longer pipeline is that it makes it possible to clock the chip faster, the disadvantage is that it makes branch mispredictions more of a problem than ever. When the processor correctly guesses what it is going to have to do next and pre-fetches the right instructions, it cooks right along; when it mis-predicts, the penalty is severe.

The 1500 was one of the two original Pentium 4s (along with the 1400). It was absurdly expensive: the debut price would have been crazy even had it been a genuine high-performance chip. As a decidedly mid-range part, equal to the lesser P IIIs and Athlons, clearly inferior to the gigahertz Athlons and quite possibly the old Pentium-III 1000 as well, it was risible.

Still, there were people who bought them — which just goes to show that it is always possible to separate a fool and his money.

The 1000MHz Pentium-III was announced a bare two days after the Athlon 1000 and promised to be a faster, better part. But the announcement was not just optimistic, it was pure fantasy. The 1000 MHz Pentium-III was nowhere near production ready. In March 2000 when Intel's marketing department first trumpeted the part, they added that they wouldn't have volume production until July, but in reality 1000MHz P IIIs didn't become readily available to the market at large until February 2001 — almost a full year from the original announcement.

P III 1000s trickled out of Intel's plants at an incredibly slow pace to begin with: pilot production in all but name. A small number of favoured OEM customers, notably Dell and IBM, got a few dribs and drabs to let them build some limited edition flagship models, but for all practical purposes the chip was not for sale.

Nevertheless, it was enough to give die-hard Intel loyalists some justification in claiming the P III 1000 as the fastest chip in the world for a time, as with its advanced cache design the Intel part would have been comfortably faster than an Athlon Classic 1000, had it been available in anything like production volume. By the time Intel's fabs started shipping in earnest, its nominal rival had long since ceased production and been replaced, first by the Thunderbird 1000, and then by the 266MHz bus 1200 C.

But despite arriving far to late to win any speed crowns, in a modest sort of way the Pentium-III 1000 became one of our favourite CPUs here at Red Hill. They were no match for the faster Thunderbirds but they were nevertheless serious performers and it was really only the traditional high Intel pricing that held them back from selling in good quantity. It was a black year for Intel and the 1000 was the only Intel CPU that we could see a rational case for buying in 2001.

Applications that work best on Intel CPUs (as opposed to other brands) were rare indeed by this time but they did exist, particularly in the audio field — and the cost of professional recording equipment and software is such that the extra few hundred dollars for a Pentium-III 1000 as opposed to something like an Athlon 1200 was immaterial. We could count the number of new P III 1000s we sold on our fingers if we used both hands, but we remember them fondly just the same.

The fastest of the original first-release Thunderbirds was priced to match. For the first six months or so of its market life the Thunderbird 1000 was almost as rare as the Pentium 1000, and for much the same reason. Intel's production problems were all but over now and you could buy a Pentium 1000 by this time, but the price of them was crazy and no-one did.

The Thunderbird was even easier to get but no-one much bought these either as the particular appeal of that four-figure number was such that AMD felt able to command a hefty premium for it. The result was that nearly everyone bought Thunderbird 800s and 900s. It wasn't until the Thunderbird 1200 came out in October 2000 that the price-performance ratio became attractive, and although this meant that TB 1000s started selling at last it was only for a short while: within another month or two the 1100 and then the 1200 C had taken over.

Illustration: how not to treat a Thunderbird. This 1000 A had a fan failure and the main board temperature sensing safety switch did not operate correctly. The Athlon XP would introduce a better overheat detection method in another year or two, but it was too late for this part. The motherboard survived the experience but the CPU was cooked within seconds.

The second version listed below was the the smallest of the C Model Athlons, and substantially faster, but very uncommon: few people were prepared to go to the expense of a brand-new 266MHz mainboard but not bother paying the few extra dollars it took to get a 1200 or a 1333.

An extraordinary debacle. In an attempt to maintain some credibility at the top of the performance tree, Intel announced the 1130Mhz Pentium-III in July 2000. As we had grown used to by this time, it wasn't available to the public in any quantity, but more of a very low-volume marketing stunt. And this time, it was a stunt that backfired. Several of the reviewers who were lucky enough to get early production samples sent to them had difficulties. Tom's Hardware, for example, were unable to complete a Linux kernel compilation without it crashing, neither were Hard OCP. (Both sites have extensive write-ups of the issue, which are worth searching out.)

The initial reaction from Intel was to ignore the warning signs, and say that there were no known problems with the 1.13GHz P III. But soon it became apparent that the problems were real, and that the poor old P6 core just couldn't be pushed any further in the then-current 0.18 micron process. Humiliatingly, Intel were obliged to withdraw the P III 1133 from sale and issue refunds to those who had bought one.

From this point on, what had long been obvious became quite unmistakable: the P-6 core had reached the end of its road. There was to be one final re-shrink of the Pentium-III but this was never going to arrive in time to compete with the ever-improving Athlon Thunderbird. From here on for Intel, it was Pentium 4 or bust.

A completely different part to the ill-fated 1133, introduced much later, and an excellent performer. It introduced the last of the Pentium-III redesigns and was manufactured on a smaller, cooler-running 0.13 micron process. There were two versions with different cache sizes: the 512k internal cache one was very highly regarded as a server chip.

The last of the 200MHz bus Thunderbirds ("Athlon A") to sell well — the 1200, 1300 and 1400 were all overshadowed by the 266MHz bus Athlon Cs and rather pointless. The 1100, in contrast, was very much a sensible option for high-performance systems and had its own short while in the sun before VIA's KT-133A chipset changed the motherboard landscape once again. The 1133 might have been more popular had it not been priced quite close to the 1200 C.

Over their design life, most CPU families have a high point, a version that in later years will be regarded as the definitive one. Most commonly this is one of the last two or three speed grades before the changeover to a new design. For the 386 it was the DX-40, for the Pentium the 166 Classic or perhaps the 166 MMX, and for the Athlon Thunderbird it was undoubtedly the 1200 C. There was a 200MHz bus Athlon 1200 too, but given the very real performance advantage of the faster bus, there was little point in buying it.

Like the 1200, the 1300MHz Athlons came in both 200 and 266MHz bus versions, and just like the 1200 the faster one was the only one to have.