Another interesting year, 2001, with several twists as one firm or the other came back from what had seemed an impossible position. As it opened, the Athlon Thunderbird ruled the high-performance arena, and with its huge performance advantage over the Celeron, the Duron was untouchable in the midrange. Intel's Pentium III was outclassed by the Athlon, their Celeron was hopelessly off the pace, and their great white hope, the Pentium 4, was in even worse shape.
Then came the Intel renaissance. New versions of the Pentium 4 started to deliver the performance that had been so conspicuously lacking in the original. April's 1700MHz P4 was not too far off Athlon performance, and in August the Pentium 4 2000 became the first Intel chip to have a really solid claim to the ultimate performance crown since the Pentium III 500 two and a half years before.
Mid-year also saw the long-despised Celeron arrive in its new, un-crippled 100MHz bus form — a vast improvement — and then the announcement of the Tualatin core Celeron which threatened to out-perform the Duron. In short, 2001 was Intel's great catch-up year, and for the most part, a year where AMD appeared to achieve little.
But just when it looked as though Intel was regaining clear leadership at last, AMD struck back with the excellent Athlon XP, easily faster than the best of the Pentium 4s. As things stood at the close of the year, the Athlon XP was clearly the most powerful CPU on the market, but the threat of the fast-improving Pentium 4 was ever-present — particularly as the P-4 is the easier of the two designs to ramp up to ever higher clock speeds. Our guess is that the fastest X86 crown will be hotly contested over the next two years or so, and that both manufacturers will possess it for short periods only. For buyers, it doesn't matter if you are interested in ultimate performance or in maximum value: fierce competition is always good news.
The Athlon XP was essentially a revised Thunderbird. There was no massive change, rather a number of incremental improvements. Better instruction pre-fetch and bigger look-aside buffers give it a modest clock-for-clock lift over the Thunderbird: no more than ten percent. The major changes, as we see it, are the reduced power consumption (which lets them run cooler and clock higher) and the long overdue thermal diode, which provides some protection against cooling system failure.
This last addition, tiny though it is, is probably the most significant of all. Until the XP, Athlons had no built-in thermal protection: while they can get as hot as 90 degrees before they fail, the power consumption of modern CPUs is such that without effective cooling it takes only a few seconds before they self-destruct. In this regard, the Pentium 4 has much the best arrangment: it self-senses its operating temperature and clocks itself back if it starts to get too hot. The Athlon XP is less advanced: it simply sends a signal to the main board, and relies on the board to cut the power. An improvement over the Thunderbird, yes, but still nowhere near as good as the Intel system.
The Athlon XP also marked the return of the unpopular PR rating system. Most commentators see it as a backwards step, as a marketing trick that is only quasi-honest. On the whole, we don't see it like that. Clock speeds, after all, have long been dreadfully misleading. Consider the several notable under-performers which have come nowhere near delivering the power implied by their clockspeed: Celeron Classic, VIA Cyrix III, Pentium 4. Or, on the other hand, ponder the over-performers which delivered much better performance than raw clockspeed implied: the 6x86 family, the K5, the K6-III, and perhaps even the Pentium Pro. (We never like the Pro, but there are those who loved it.)
In short, PR ratings honestly applied are no worse than, and quite possibly much better than actual clockspeeds. Sometimes we fondly imagine that both clockspeed and raw CPU ratings will one day become so obviously unrealistic that buyers will cease to pay any attention to them at all, and simply consider each chip on its actual merits. But we won't be holding our breath waiting for it.
| Form | Design | Manufacture | Introduction | Status |
|---|---|---|---|---|
| Socket A | AMD | AMD | October 2001 | Legacy |
| Internal clock | External clock | L1 cache | L2 cache | Transistor count |
| 1333 MHz | 266 MHz | 128k at 1333 MHz | 256k at 1333 MHz | 37.5 million |
Unlike the XP 1500, which was a paper part and very rarely seen, the XP 1600 sold in reasonable volume on the market at large. We have never been able to understand this, as the price difference between an XP 1600 and the 1700 was usually tiny — typically about $30 — and not going that small extra step for the extra 66MHz seemed hard to justify.
| Form | Design | Manufacture | Introduction | Status |
|---|---|---|---|---|
| Socket A | AMD | AMD | October 2001 | Current |
| Internal clock | External clock | L1 cache | L2 cache | Transistor count |
| 1400 MHz | 266 MHz | 128k at 1400 MHz | 256k at 1400 MHz | 37.5 million |
With this part, the Pentium 4 arrived. The 2000 was the first Intel chip to outpace AMD's best since the Pentium III 550's brief reign two and a half years before, and provided some justification for Intel's brave and very unorthodox design decision to sacrifice all else in the pursuit of high clock rates.
As a practical proposition, however, it made no sense at all. The performance advantage of the Pentium 4 2000 over the Athlon 1.4 was very small — so small that there remains some controversy as to which of the two parts really performs better — but the cost was (and still is) astronomical. The CPU itself is between three and four times more expensive than the more-or-less equal performance Athlon 1400C or Athlon XP 1600, but the real cost is even higher, as Pentium 4s were designed for, and only run decently with, proprietary Rambus RDRAM, which performs similarly to but costs several times more than the DDR that Athlons use. On the face of things, the P4 2000 ought to be a very difficult part to move. But if we cast our minds back a few years, it becomes apparent that it is, in fact, a very traditional product indeed. Up until they lost performance leadership around the beginning of 1999, Intel always priced their flagship part far higher than any rational analysis could justify. The absurdly low price-performance of the Pentium 4 2000 is simply a return to traditional Intel form. In a way, it says more about the psychology of some CPU buyers than it does about anything else.
| Form | Design | Manufacture | Introduction | Status |
|---|---|---|---|---|
| Socket 478 | Intel | Intel | November 2000 | Current |
| Internal clock | External clock | L1 cache | L2 cache | Transistor count |
| 2000 MHz | 400 MHz | 8k at 2000 MHz | 256k at 2000MHz | 42 million |
This was the obvious first-release Athlon XP to buy.
Nearly always when a new part first comes out, the top-of-the-range stepping is absurdly dear, and the cheapest one is only marginally less than its bigger brothers. This was as true of the Athlon XP as of any other CPU.
Priced modestly higher than the 1500 and the 1600, and significantly lower than the 1800, the XP 1700+ was the clear price-performance leader when the XP arrived in late 2001, and it remained the one to have for a good six months. Paired with a DDR board, it was easily faster than a Pentium 4 2000 with RDRAM, and about one-third of the price. The gap has narrowed since the XP debuted but right now (April 2002) it is still clearly the high performance chip of choice.
So far as we were concerned, the 1700 was the only XP part to have: we did not bother stocking any other speed grade for a full six months after the XPs arrived. The 1700s, on the other hand, sold like hotcakes.
| Form | Design | Manufacture | Introduction | Status |
|---|---|---|---|---|
| Socket A | AMD | AMD | October 2001 | Current |
| Internal clock | External clock | L1 cache | L2 cache | Transistor count |
| 1466 MHz | 266 MHz | 128k at 1466 MHz | 256k at 1466 MHz | 37.5 million |
The fastest CPU on earth when it first arrived, well out in front of the original Pentium 4 2000, and very reasonably priced by comparison with that spectacular budget-buster.
Still, at AU$150 more than the almost equal XP 1700, we thought it hard to justify. It took almost six months before the XP 1800 became a sensible proposition.
We have never understood why so many people are prepared to pay outrageous premiums for marginal performance differences: certainly very, very few of our customers ever do. Perhaps it is 'fastest on the block' syndrome or simple ignorance that drives sales of overpriced hardware, but we don't think so. In reality, it is almost always badly trained or unscrupulous sales staff that sucker people into paying vastly more for almost nothing extra.
After the XP 2100 came out in April '02, AMD finally saw fit to halve the price gap between the XP 1800 and the 1700 and, rather late in the day, it began to sell in reasonable numbers. With the new breed of third-generation Pentium 4 parts out now, featuring more design optimizations, twice as much secondary cache, and even higher clockings, overall performance leadership is more difficult to call than in was in 2001, but while the XP 1700 remains the best price-performance chip of all, the 1800 is in the ballpark .
| Form | Design | Manufacture | Introduction | Status |
|---|---|---|---|---|
| Socket A | AMD | AMD | October 2001 | Current |
| Internal clock | External clock | L1 cache | L2 cache | Transistor count |
| 1533 MHz | 266 MHz | 128k at 1533 MHz | 256k at 1533 MHz | 37.5 million |
As ever, this industry produces new and better products faster than we can find the time to write about them. There is a great deal of detail that needs adding to the entries above, and yet we have not even touched on the more recent models. Maybe one day we will find time to bring this up to date again, maybe not. But here, very briefly, is a run-down on the more notable CPUs of the last few years:
- Celerons. None of the Pentium 4 based Celerons are worth having. The only thing we ever used them for was notebooks, where they used too much power and didn't go fast enough.
- Athlon XP Thoroughbred: 2200 and up, most common as 2200, 2400 and 2700 parts. No faster than Palomino (original Athlon XPs like the 1700) but the production process allowed clockspeed increases, which kept it more or less competitive with the much-improved Pentium 4.
- Northwood Pentium 4. Much improved over the lack-lustre early models, and well able to compete with the small-cache Athlon XPs performance-wise, but usually way too expensive.
- Sempron (Socket A). Essentially just small cache Athlon XPs. Faster than Celerons. But then so is custard.
- Various other Pentium 4 models. Not much to say about these, as we practically never see them. There about a million models with a host of different motherboard requirements, nearly all of them quite expensive. No one ever buys them. Yes, we mean no-one — you could certainly count the number of Pentium 4s we have sold new on the fingers of one hand. If you wanted to count on your thumbs instead, you'd need both hands. We see them come into the shop from time to time for small jobs like having a Windows reinstall or viruses removed, and always walk away scratching our heads and wondering why people spend so much to buy them, as they just don't seem to be capable of delivering any real snap, particularly when you ask one to do two or three tasks at once.
- Pentium M. A truly excellent little chip, based on the old Pentium III but heavily revised. Designed as a low-power notebook part, the Pentium M blows the P4 into the weeds and would be very interesting to see running in a desktop system, as it would give an Athlon XP a very good run for its money.
- Athlon XP 2500. The big cache Athlon XP with the Barton core was easily faster than anything else around, be that the old Thoroughbred Athlon models or the Pentium 4, and remained so for a very long time. The 2500 was unquestionably the one to have: despite a very modest 1833 raw clock, the fast bus speed and the big cache were both significant. Faster clock versions (up to 3200) were quite expensive, and achieved little, unless you were gaming. One of the all-time great CPUs.
- Sempron (Socket 754). The new Durons? Excellent performers based on the Athlon 64
- Athlon 64. The fastest CPU on the planet by quite a big margin. The top models are very expensive, but parts in the 3000 to 3500MHz range are quite reasonable, and very quick.
