Imagine an excellent
processor that's compatible with all the newest
boards and can do most anything except
overclock its
FSB. Now imagine the
polar opposite.
To be more exact (and perhaps a bit less cruel), the original Celeron (Covington core, thanks to g026r for pointing out that it wasn't Deschutes) was thought up around the time that the Pentium II got a 100MHz FSB and dropped the old Klamath core. If they made a chip that was basically a crippled PII, it would cost less to make, and they could sell it at about the same margin. What it would be was basically a crappier and cheaper product; Pentium, Mark Negative One. The principle of the bargain product was and is pretty well-understood, and it seemed logical to apply it to a CPU. The FSB was set at 66MHz, to cripple it some.
That's about when the train derailed.
Rather than simply cripple the core and let the thing be, they chopped off the entire cache; cache takes size, and the smaller the die was, the more they could make. Unfortunately, slicing off ALL of the cache had catastrophic effects on the processor's speed, as was soon discovered. Shortly after, Intel hurriedly stuck 128kb of onboard cache onto the new model. The fancypants new Celeron chip with on-die cache was dubbed the Celeron A, which used the Mendocino core. The first model out was the 300A. This is where Intel made a really big screwup. The 300A turned out to be pretty much exactly as powerful as the PII, at literally a quarter of the cost. What's more, it overclocked easily to 450MHz even though the multiplier was locked. That's right; a 33MHz increase in the frontside bus. Oops, indeed.
The 333MHz and 366MHz (and 400MHz, the fastest CPU that can run on a Slot 1 motherboard with a 66MHz FSB, as g026r points out) SEPPs following the 300A are worth noting, but only so that there's no hole in the history. Unremarkable Celeries that didn't overclock nearly as well as the 300A. The PPGA 366, however, is a different case. PPGA stands for Plastic Pin Grid Array, and it utilized the Socket 370 format. The PPEG overclocked much better than the regular old SEPP Slot 1: 366 to 550 (thanks to Zerotime for the correction) as opposed to 366 to 433 on average. Additionally, it had 256kb of full-speed cache. People actually argued that the Celerons were better than the Intel's Pentium line, as opposed to arguing that the Celeron is better than a stick of celery.
The Celeron 566 used the PIII Coppermine core, or at least almost all of it. The FSB remained at a sad and pathetic 66MHz, and the cache was down to 128kb again. The reason for the latter was that the regular Coppermine used a 256kb cache, and the Celerons were bargain computers. This core was called the Coppermine 128, and the CPU was called the Celeron II. It had most of the advantages of a PIII. Various core changes and upgrades that are now pretty much requirements for a processor went into the Celeron II: think Pentium going to Pentium MMX in terms of performance. This did not change the cache. With overclocking, it ran like a donkey. Without, it still ran like a donkey, the chief difference being that the latter would be a deceased donkey.
It took longer for the FSB to hit 100MHz; the equivalent of reaching the CPU Bronze Age, nowadays. The 800MHz was the first Celeron to have a 100MHz FSB, which is odd when you consider that it came out at the exact same time as the 766MHz and cost almost the exact same.
Intel's last P3-based Celeron was based on the Tualatin core. It had a 256KB L2 cache and was a notable improvement over its ancestors; of course the Pentium III Tualatin itself was better. As usual.
The Tualatin was rather short-lived, and the real next step in the history of the Celery is the Willamette (old Pentium 4)-based Celeron. Often referred to as "Half a Willamette," the P4-based Celeron suffered from a meager 128KB L2 cache and an underpowered FSB. The Pentium 4's FSB is basically quad-pumped-- the 533MHz FSB of newer models is literally a 133MHz done four ways. The Celeron has a 266MHz FSB. Do the math. Basically this means that we got to start all over again at 66MHz. (Shudder) Anyway, like all the Celerons before it, the Willamette Celery could overclock from 66MHz to 100MHz, giving it a clock speed several hundred megahertz higher than before, and much better performance.
However, even when overclocked to about 2.26GHz, a Pentium 4-based Celeron would continually lose to the Athlon XP and the Pentium 4 in performance testing. The reason for this is simple-- when the task required a high clock speed and nothing else (media encoding), the Pentium 4 would win even at slightly lower clock speeds due to a non-castrated processor design. When the task required rendering, the Athlon XP won, as the XP excelled at 3D rendering of most kinds. It is for this reason that the modern Celeron is regarded as a painfully obvious attempt to appeal to the casual computer user. While some arguments have been made that it's the best possible choice for a value Socket 478 processor, this is true only because there's no other value Socket 478 processor.
On a related note, AMD's counter to the Celery was the Duron, which had more L1 cache, a vastly more efficient L2 cache, and a different base design. The Celeron, while generally agreed to be inferior to the AMD Duron, is used in many bargain PC packages, especially by HP. Users who don't know any factors in a computer's speed but clock speed continue to take Celery processors. The fact that one can generally count on a random graph in a benchmark to have a pathetic-looking line practically on the x-axis labeled "Celeron 700MHz" or something along those lines is of little consequence to those who don't have any idea what a benchmark is. (The funny thing is, they often have to expand the scale a ways downwards so Celerons fit. No, really.)
Recently, Intel has introduced the Prescott-based Celeron, designed as a replacement for the Northwood Celeron. Given that Prescott, ehm, had some problems, one would expect this "Celeron D" to suck even more-- but gasp, it didn't! Since the integer pipeline in Prescott was lengthened from 20 to 31 stages, Prescott had a whole bunch of shiny features (including a BPU somehow improved even further) designed to prevent stalls, clogs, and various other problems one generally associates with toilets. Now, the P4 Celeron has always had a mostly empty pipeline due to its small cache and narrow FSB, so these improvements were REALLY major for it-- suddenly it was competitive with the Athlon XP! Now the K8 Semprons existed, of course, so it wasn't quite the king of budget yet, but it's still worth noting, because it's the first time since the Tualatin that the Celeron hasn't sucked.
In case you think the Northwood/Willamette Celeron really wasn't THAT bad:
http://www.anandtech.com/cpu/showdoc.html?i=1927
JDooty1234: Err, you don't need a particularly powerful heatsink to overclock a Cu128 Celeron. After all, you did say a
Golden Orb managed it. The Celeron 600 came out a few months before the
Duron, so it wasn't a bad choice if you bought it before June 2000. If you bought it after then, well, is it worth anything that nearly all Duron 600s can make it to 900MHz, and perform far better than Celerons? (A Celeron 850 is equivalent to a Duron 700.) Durons were also much cheaper. If you got it in June of 2000 or afterwards, well, you made an uninformed decision and paid the price.