by FastSite
06/19/2000 | 12:00 AM
Well, Intel Celeron family turned out one of the most rapidly developing processor families. Celeron owes its birth to Intel's new strategy, which implies designing different CPUs for different market sectors, and since the day it appeared its has already been based on three totally different cores. However, all of them had one similar feature: the system bus frequency, which was equal to 66MHz. In fact, this seems to be the main bottleneck that prevents the performance of Celeron processor from rising to the level of Pentium III. And although the processor frequencies keep growing as well as the amounts of data processed, Intel doesn't care to give up the 66MHz bus and continues manufacturing Celerons with a 66MHz FSB. <%BANNER[article]%>
And so, on June 19, 2000 the microprocessor giant announced two new CPUs from this family supporting 633, 667 and 700MHz. And of course they also require a 66MHz FSB. Unfortunately, the so long-awaited switch to 100MHz system bus by Celeron processors is very unlikely to happen at all, at least Intel's plans for the nearest future don't even mention this. The only thing, which can somehow change this sad situation, is AMD with its low-end Duron processor. This CPU differs only by the cache size from its elder brothers and provides a much higher performance than Intel's competitor. Well, we'll see whether AMD manages to arrange a very clever marketing policy and to make Duron a popular processor. If AMD succeeds, Intel will have to change its views of the Low-End market, and in the meanwhile this is all we have. Take a look at the specs of Intel Celeron 667, which we were lucky to get for testing:
So, we see that this is none other but one more processor (like all the previous Celerons) on Coppermine128 core, differing first - by the multiplier and second - by the Vcore.
As for the multiplier, it seems just enormous. We have never seen a multiplier equal to 10x before. The maximum achieved by Intel Pentium III, for instance, was 8.5 by an 850MHz CPU intended for a 100MHz FSB. Besides, there are no mainboards, which could allow setting 10x multiplier neither with the jumpers, nor in BIOS Setup. However, this doesn't mean that 667MHz Celeron has a limited sphere of application. All Intel's CPUs have locked the clock multiplier in the processor core that is why Celeron 667 as well as all other Celerons and Pentium III processors simply doesn't care about the clock multiplier set on the mainboard. So, you can easily forget about choosing the clock multiplier when you install your Celeron onto the mainboard: the processor knows everything itself.
As far as the core voltage is concerned, it is a bit higher compared to all the other models (starting from the Celeron 633 it will be equal to 1.65V), there is hardly anything to be surprised at. This increase is caused by the growth of the CPU working frequency and now it became the same as of other Coppermine based processors.
All the other peculiarities of Celeron 667 are determined by the Coppermine128 core and have already been described in our Intel Celeron 566 Review, for instance.
Just in case, we would like to remind you of two Celeron's peculiarities totally different from Intel Pentium III and influencing the performance greatly. The first one is dramatically low FSB frequency, which results into extremely low processor bus bandwidth: 533MB/sec, which is twice as low as that by elder Pentium III. And bearing in mind the today's CPU speeds, this bandwidth turns one of the major system bottlenecks. As a result the systems built on 66MHz processors can't boast high performance.
The second difference is a twice as small L2 cache, which is only 128KB. So, the hits appear less probable and hence Celeron processor has to turn to the slow system memory more frequently. Really, associative cache memory working algorithm requires splitting L2 cache and system memory into equal number of parts, so that to make each part of the cache memory responsible for the corresponding part of the system memory. In the first place, this should speed up data search within the cache when the CPU addresses a certain field of the system memory. Since Celeron and Intel Pentium III can boast absolutely the same L2 cache construction, L2 cache of Celeron CPU is twice as small and the size of its certain sectors is also twice as small: four 32byte lines against eight 32byte lines of Pentium III. This is exactly the reason for the data to appear in Celeron L2 cache with smaller probability than usually, because Celeron's cache gets stuffed much quicker than that of Pentium III.
But things are not so bad as they might seem. Due to the fact that Coppermine128 processor core, as well as Coppermine core, is manufactured with 0.18 micron technology and hence is designed to support frequencies very close to 1GHz, Celeron processors keep pleasing our sight with perfect overclockability. For example, as we have already written in our previous article, Celeron 566 very often can be overclocked by 1.5 times, namely up to 850MHz. Our Celeron 667 also didn't disgrace itself and proved cool at 66MHz FSB as well as at 100MHz FSB frequency. So, since the clock multiplier of Celeron 667 equals to 10x, we got a CPU working at 1000MHz! As a result, even before the official sales of Intel gigahertz processors had been started (note that 1GHz Pentium III should appear in retail only in September), we got the chance to test a Celeron working at this super frequency. No matter that 1GHz is considered a certain symbolic barrier: it was a thrill to watch the CPU working at this frequency.
Well, let's pass over to practice. All the CPUs were tested on the following system:
Bearing in mind the results shown by younger Intel Celeron models, the performance of Celeron 667 doesn't strike as something extraordinary. Well, we were a bit upset that this CPU failed to surpass even the youngest of the normal Coppermine family, Intel Pentium III 500E. However, it was Intel's decision to provide its Celeron with a too slow processor bus working at 66MHz only. As for overclocking, Celeron 667 overclocked to 1000MHz didn't look like a brake at all working almost as fast as Intel Pentium III 866.
The situation repeats. Slow system bus and a half as large L2 cache don't let Celeron 667 catch up even with Intel Pentium III 500. And a 1GHz Celeron doesn't look that impressive in SYSmark unlike the previous test: its performance turns even lower than by Intel Pentium III 800EB. The thing probably is that SYSmark involves more resource-consuming applications that is why a CPU with a smaller L2 cache loses the competition.
As you can see from the chart, the FSB frequency is the main factor influencing the performance in this gaming test run with high level of detail and at high resolution. In Quake3 in High Quality mode the memory bus bandwidth is of great importance because the graphics card requests the data for rendering from there. That's why a 66MHz Celeron 667 falls behind all its competitors in this test. In our system on i440BX based board the memory also worked at 66MHz. That is why it is no longer surprising that a Celeron overclocked to 1000MHz shows almost the same fps rate as all other processors featuring 100MHz FSB.
Here the influence of the processor bus bandwidth is even more evident. No doubt, the today's games can't be satisfied with 66MHz. Comparing the performance of the non-overclocked Celeron 667 to that of the Celeron 667 overclocked to 1000MHz by means of increasing the FSB frequency to 100MHz, we will see that that the performance gain obtained by increasing the system bus frequency by 50% makes over 45%.
And again the situation leaves much to be desired. Intel Celeron 667 is lagging behind all Pentium III models on the Coppermine core. As for our "handmade" 1GHz Celeron, it proved even slower than Intel Pentium III 733 in this test.
Setting a lower resolution in Unreal Tournament won't have any real effect on the performance: Celeron 667 is again defeated by Pentium III 500E, and Celeron 1000 appears just a bit faster than Intel Pentium III 667.
The performance in Expendable depends greatly on the memory subsystem. So, Celeron 566 is almost 16% behind the Pentium III 500 CPU. However, setting the FSB frequency to 100MHz turns more efficient than in the previous cases: Celeron 1000 gets between Intel Pentium III 733 and Pentium III 800EB.
Here is another proof of the already said things. Low system bus frequency of Intel Celeron doesn't let it show its best. A Celeron overclocked to 1GHz performs quite nicely, however, it still looks less impressive compared to the epoch of Mendocino based Celerons and Katmai based Pentiums. Nevertheless, you should bear in mind that Mendocino had a much better cache architecture than Katmai, and the today's Celeron and Pentium III processors have an absolutely similar L2 cache, of a different size though. That is why you should forget about Celeron as a CPU offering high performance for a penny. Now Intel Celeron is just a Low-End CPU, none other than that. Even if you overclock it, it will remain that.
Speaking about Intel Celeron 667, we should point out that despite a relatively high frequency, it still failed to show something extraordinary in any of the tests run: it didn't beat the youngest Intel Pentium III on Coppermine. Therefore keeping in mind that in some shops its initial price is higher than that of Intel Pentium III 500E and even 550E, it hardly has a shining future ahead. Having left the FSB frequency equal to 66MHz, Intel automatically deprived its processors of any hope to succeed.
As for the chance to overclock this processor to 1GHz, impressive value appears the only nice thing. The performance of 1000MHz Celeron turned out even lower than that shown by Pentium III 733, which means that an overclocked to 750MHz Intel Pentium III 500E will easily leave Celeron 1000 far behind.
Well, we think that AMD's Duron positioned as a solution for the Low-End market and lacking any "artificial brakes" may appear a much better choice to make in terms of price-to-performance ratio. The only thing that may change the situation in the low-cost processors sector is the integrated Intel Timna postponed till 2001. Besides the processor core, this CPU will also have the North Bridge with integrated graphics, which will allow assembling very cheap systems.