In fact, not so much time has passed since April 1998, when Intel announced their intention to divide theprocessor market into segments and introduced the first processor of the Celeron family, aimed at sub-$1000PCs. However, during these two years Intel Celeron underwent a lot of radical modifications.

The first Celeron processors working at 266 and 300MHz were none other but the ordinary Pentium II CPUswithout L2 cache. As a result, their performance in business applications could hardly be called satisfactorywhile in games Celerons without L2 cache suffered a total fiasco. The only remarkable thing about these CPUswas their extraordinary overclockability. Since they were deprived of external L2 cache microchips, which didn'tallow increasing the frequency of the ordinary Pentium II much higher than the official nominal, almost all IntelCeleron 266 processors got easily overclocked to 400MHz. However, these processors still failed to become verypopular because of their relatively low performance, so that Intel had to think of a way-out.

And they found a way-out pretty soon. It was a new Celeron version working at 300MHz and built on Mendocinocore. The mass production of these processors started in August 1998. They could boast an L2 integratedon-die cache working at full CPU frequency. In other words, the situation turned out quite paradoxical:although new Celeron processors featured a smaller L2 cache than Pentium II (128KB against 512KB), itworked at full CPU frequency and not at half the CPU frequency like Pentium II. Besides, L2 cache latencywas a bit lower than that of its "elder" brother due to the fact that it was integrated into the processorcore. So, this made Celeron CPUs perform almost as well as Pentium II did. Moreover, they got much moreoverclockable as well. Take for instance, Celeron 300A built on Mendocino core, which was generally recognizedthe best CPU, because it could be overclocked to 450MHz. All in all, Celeron had been a really serious competitorto Intel Pentium II before Pentium III came out.

However, Intel Pentium III could boast a new set of SSE SIMD-instructions for floating point operations. Celerondidn't support them and hence from then on started falling behind the CPUs aimed at more expensive PCs. Besides, thegrowing processor frequencies allowed processing larger amounts of data which turned Celeron's 66MHz system bus intoa tangible bottleneck. While Pentium III CPUs were available with 100 and 133MHz system bus, Celeron had retained its66MHz bus since the very first day.

Very soon Intel switched to 0.18 micron technology in Pentium III manufacturing, enhanced the CPU core and providedits Pentiums with L2 on-die cache working at full processor frequency, the same as by Celeron processors. The new corewas called Coppermine and featured a 256KB L2 cache, which was not only larger than that of Celeron but could also boasta lower latency. And again expensive Pentium III CPUs managed to make a significant jump forward having left low-costCelerons far behind.

Well, this was how matters stood in the beginning of the year 2000, when Intel was planning another Celeron familymodifications.

According to Intel's plans, Celeron had to acquire a new Coppermine core and its manufacturing had to be moved tothe same technology as that of Pentium III, namely, to 0.18 micron. Unlike Pentium III, L2 cache of Celeron processorswas supposed to become twice as small: only 128KB. However, the system bus was expected to finally turn something otherthan 66MHz.

But unfortunately, all these projects were only partially put into practice. As Intel had promised, a newCeleron with the Coppermine core and 128KB cache came out but its system bus frequency remained equal to thesame old 66MHz. So, let's sum up everything we know about the new Celeron. Here is a specs list for thenew Celeron processor based on Coppermine core (it is also known as Celeron 2):

• Coppermine128 core, manufactured with 0.18 micron technology;
• 533/56/600MHz working frequencies (8/8.5/9.0x clock multipliers);
• 32KB L1 cache (16KB for data and 16KB for instructions);
• 128KB L2 integrated on-die cache working at full core frequency (256bit Advanced Transfer Cache);
• Advanced System Buffering;
• SSE SIMD-instructions set;
• 370pin FC-PGA Socket-370 CPU interface;
• GTL+ system bus working at 66MHz;
• 1.5V Vcore.

Judging by the characteristics given above, we have every reason to conclude that the manufactured Celeron processors onCoppermine128 core will work at the frequencies starting from 533MHz. However, there are also 533MHz Mendocino CPUs in themarket as well. In order to distinguish between them Intel marks new Celeron processors as Celeron 533A. As for the fastermodes, they are all based on Coppermine128 core.

So, what do we get if we compare a new Celeron 2 to Intel Pentium III? In fact, Intel Celeron on Coppermine128 core is noneother but the same Intel Pentium III with an electrically disconnected half of L2 cache. In other words, it means that theremaining 128KB of the L2 integrated on-die cache work in an absolutely the same way as the entire 256KB L2 cache of IntelPentium III. This allows us to completely disprove a very widely spread opinion that the cache in new Celeron processors isslower and its latency is much higher than that of the cache in Pentium III.

That's why we should consider the system bus frequency to be the main difference between Celeron and Pentium III.Unfortunately, new Celeron processors feature the same system bus as the old ones: 66MHz. Intel seems to have given upthe idea of equipping its low-cost CPUs with a 100MHz bus because the reduction of L2 cache size didn't lead to asignificant performance drop as they had expected. So, Intel simply couldn't put up with the fact that its cheapprocessors would work as well as expensive ones therefore they had to think of some "slowing down mechanism" forthem. And they found it. All they had to make was a system bus a bit slower than required for the today's needs.Since the working frequencies of the modern Celeron CPUs equal to 533MHz and up, the amount of data processed bythe processor is considerably larger than can be transferred through a 66MHz system bus, which top bandwidth is533MB/sec. As a result, the processor can simply stand idle waiting for the new data to be submitted or to betransferred to the system memory. That's why you shouldn't be very enthusiastic about the performance of IntelCeleron processors: though built on the Coppermine core, they will hardly prove as cool as you expect them to.

Besides that, Celeron on the new core can boast a couple of other small differences from the today's PentiumIII. First, the new Celerons (like the old ones) don't have the processor serial number, one of the most vexedissues concerning Pentium III functions, which created a great stir some time ago. It looks as if Intel decidedto totally give up processor serial number support, because there will be no serial number in Willamette as well,as to our sources. Second, Celeron's Vcore is slightly lower than that of Pentium III processors on the same Copperminecore: only 1.5V. You shouldn't be surprised about it, it's just because of a lower system bus frequency. That is why thereis no need to worry when you increase the core voltage up to 1.6-1.65V (the Vcore of Pentium III) during overclocking.

And now let's figure out what makes new Celeron processors on Coppermine core better than the old ones built on Mendocino.The first and the most important peculiarity of the new processor is a different L2 cache architecture. Although L2 cache isof the same size by both CPUs, new Celerons, as well as Pentium III, possess the so-called Advanced Transfer Cache. In fact,it means that its data path between the cache and the processor core is four times as wide (256bit of a new Celeron comparedto 64bit of the old one) and its latency is much lower (four times as low as that of Pentium III, for instance, built on Katmai core).Theoretically these architectural differences between L2 caches of Intel Celeron processors should guarantee a much higherperformance of the new models.

Another advantage of the new Celeron based on Coppermine128 core, which is worth mentioning, is the support of a newset of SSE SIMD-instructions. These new instructions allow increasing the efficiency of all calculations in 3D games andfor streaming sound and video processing. Although there are not so many applications using new SSE instructions rightnow, they keep multiplying little by little.

Of course, we should also mention some other enhancements of the new Celeron besides more progressive technology (0.18micron compared to 0.25 micron). It will be also provided with a slightly different package, although theywill also be designed for Socket-370. The old Celerons were manufactured in Plastic Pin Grid Array (PPGA) package, wherethe die was hidden under a special metallic cover. However, the new ones used Flip-Chip Pin Grid Array (FC-PGA) where theprocessor die remained free. In the first place, this ensures a better chip cooling, because in FC-PGA package the cooleris pressed directly to the processor core. So, taking into account more progressive manufacturing technology, better coolingand lower voltage, we can say that new Celeron processors should be able to easily reach 1GHz and up. As for the Celeronprocessors on the Coppermine128 core available in the market right now, they can also boast much lower heating and workingtemperature and hence a much better overclockability. Take for example, Celeron 533A based on Coppermine128, which dissipatesonly up to 17.1W of heat while the Celeron on Mendocino core dissipates up to 28.3W.

As you probably remember, one of the main problems connected with Intel Pentium III on Coppermine core was the absenceof corresponding support by mainboards. And what about the new Celerons especially bearing in mind that some mainboards donot support their clock multipliers (starting from 8x and up)? Frankly speaking, there is hardly anything new compared toFC-PGA Pentium III. The mainboard should support Coppermine in its BIOS, allow 1.5V Vcore and support FC-PGA CPUs, whichpinout is a bit different from that of PPGA. As far as the clock multiplier is concerned, it belongs to the CPU functionsbecause it is locked in the processor. That's why even if the mainboard doesn't allow setting the required multiplier, theCPU will still work fine. Nevertheless, old Socket 370 mainboards, mostly those built on i440LX, i440EX, i440ZX and i440BXchipsets won't support new Celeron processors, because a couple of pins in Socket 370 pinout have changed their functions.

Well, it's high time we looked at the performance of a new Intel Celeron CPU. For our tests we chose a 566MHz Celeron.Unfortunately, we didn't manage to overclock it to 850MHz by means of increasing the system bus frequency to 100MHz,because the system was very unstable in that case. However, in our testing system we used a VIA Apollo Pro133A basedmainboard. As you know, this chipset allows setting the memory bus to 66MHz as well as to 100MHz independent of theFSB frequency. So, we tested Intel Celeron 566 twice: first at 66MHz memory bus frequency (this helps to model the workof the CPU with mainboards based on i440LX/ZX/BX) and then at 100MHz (this setting corresponds to the case with themainboards on i810/i810E).

The testing system was configured as follows:

• Intel Celeron 333 (overclocked to 500MHz by means of the FSB frequency increase up to 100MHz),
  Intel Celeron 366 (overclocked to 550MHz by means of the FSB frequency increase up to 100MHz),
  Intel Celeron 533,
  Intel Celeron 566,
  Intel Pentium III 500
• Gigabyte GA-6VX7-4X mainboard
• Creative 3DBlaster Annihilator graphics card
• Creative Sound Blaster Live! sound card
• IBM DJNA 372200 HDD
• 128MB PC100 SDRAM by SEC

As usual we started our tests with the performance of our CPUs in office applications:

Well, the first test brought the first disappointment. As you can clearly see, a new Intel Celeron 566 doesn't showany upsurge in performance compared to the old Intel Celeron built on the old Mendocino core. Moreover, a 550MHzCeleron with a system bus overclocked to 100MHz easily surpasses the new 566 one. Actually, we should blame a 66MHz systembus for this disgraceful result of Celeron 566, because the processor bus bandwidth makes only 533MB/sec, while the CPUsworking with 100MHz FSB transfer up to 800MB/sec.

Well, the situation is absolutely the same as in the previous case. The only difference is a greater lag shown by Celeron566. This time it managed to fall behind a 500MHz Celeron with a 100MHz system bus. SysMark 2000 benchmark includesapplications, which are much more sensitive to the processor bus bandwidth. Namely, these are the applications forvideo and sound editing and speech synthesis.

Now let's watch a new Celeron in games:

The main bottleneck in this mode is the graphics card that's why all the systems built on completely different CPUsperformed practically on the same level.

Here the processor bus bandwidth exerts a much greater influence over the performance. No doubt, the today's games cannotdo with 66MHz FSB. By the way, this test shows that even a 66MHz memory bus is also not enough. The performance gain obtaineddue a 100MHz memory bus makes over 10%. No wonder, really, since AGP 4x can transfer up to 1.06GB/sec while a 66MHz SDRAMprovides only 533MB/sec data transfer rate, which is the major weakness of the system in 3D graphics.

The results depend greatly on the system memory working frequency. However, even with a 100MHz memory bus Intel Celeron566 can hardly catch up with Celeron 333 overclocked to 500MHz with a FSB increase up to 100MHz.

This diagram almost coincides with the previous one. Unreal Tournament is a very resource-consuming game that's why thegaming performance of a CPU in this game depends on the whole bunch of various factors. Nevertheless, it clearly shows thata 66MHz bus is a real brake of the today's system.

The memory subsystem fastness tells greatly on the performance in Expendable. This is the main reason for the oldCeleron 533 on Mendocino core with a 100MHz memory working frequency to surpass the new Celeron 566 with the memory workingat 66MHz. It gives us every right to state that synchronous chipsets such as i440BX/ZX can't ensure the appropriate levelof performance with a 66MHz processor bus.

Here is another proof in favor of the already stated things. Having refused to provide new Celerons with a 100MHz FSB,Intel simply ruined all their architectural advantages. Moreover, low system bus bandwidth makes the use of elder Celeronmodels absolutely inefficient, because it doesn't allow the CPU to exchange data with the memory and other devices fastenough, which turns out the worst bottleneck of Intel Celeron based systems.

Conclusions

Unfortunately, the shift to a new Intel Celeron core didn't lead to any improvements. Intel provided this new core witha powerful "brake", which locked the CPU's architectural advantages. The only positive thing about Celeron is its low price,which makes it suitable for really low-cost systems. However, even in this case you should bear in mind that together withthe mainboards on i810, i810E and VIA Apollo Pro133A chipsets, which allow clocking the memory to 100MHz for the CPUsfeaturing 66MHz FSB, the performance gain may appear quite significant. But Intel Celeron should very soon acquire areally powerful competitor deprived of the listed drawbacks. This new CPU aimed at Low-End market sector is the upcomingAMD Duron with allegedly higher performance. It may actually force Intel to change their point of view, but on the otherhand Celeron also has every chance to remain just an overclocking solution.

So, Intel Celeron based on the Coppermine128 core can be regarded only as a continuation of the low-cost CPUs family.It doesn't have any worthy practical advantages except SSE support. However, we should admit that although we failed tooverclock our Celeron to 100MHz system bus, it is still quite possible due to a more progressive 0.18 micron technology.In this case a new Celeron processor will be a really nice solution even though Intel Pentium III on Coppermine core showsa much higher performance increase if overclocked.