The Intel Celeron has been with us for over two years now, and throughout that time, it hasn't changed much at all.  First introduced in 1998, based off of what was known as the Covington core, the original Celeron had no L2 cache.  The next incarnation of the Celeron featured Intel's first on-die L2 cache, with 128KB of it to be exact; this was the Mendocino core that carried us through 1999. 

The next update to the Celeron core came in March of last year, with the Coppermine128 core.  This core didn't improve performance noticeably, it simply added SSE support since it was essentially a Coppermine Pentium III core with half of its L2 cache disabled (256KB / 2 = 128KB). 

While the processor may have gained new instructions and went from being built on a 0.25-micron process to Intel's most advanced 0.18-micron process, it has generally remained true to its original form. 

Although the Celeron hasn't changed much since its induction into the value PC market segment, the applications that it's forced to run have changed.  If you'll remember, back in 1998 and 1999 the performance difference between a Celeron and a Pentium II/III wasn't too great at all.  Clock for clock, the Celeron was at most 10% slower than its older brother.  However, as time moved on, that performance gap grew considerably. 

It is now common knowledge why the Celeron is performing considerably lower than the Pentium III, Athlon and Duron processors.  The Celeron's clock speed has gone from an introductory 266MHz to just recently hitting 766MHz; however, all this time, the Celeron's Front Side Bus has been stuck at 66MHz and in most cases, the memory bus is stuck at the same 66MHz as well. 

What does this mean?  Well, the processor can continue to crunch numbers as it gets faster internally, but its connection to the "outside world" (memory, hard disks, graphics card, etc…) occurs through what is known as the Front Side Bus (FSB).  In the case of the Celeron 766, the CPU was operating at 11.5 times the speed of the FSB.  Regardless of how fast the CPU is, if the data can't get to it as fast as it is processing it you're going to be bottlenecked, which is the symptom that the Celeron was exhibiting. 

In spite of this, and in spite of the fact that the Duron was simply trampling all over the Celeron in the performance benchmarks, Intel's Celeron was still outselling AMD's Duron.  The fact of the matter was that, independent of its performance, the Celeron had something the Duron didn't have - a value platform that OEMs and system integrators could use in sub-$1000 systems.  This value platform was the i810E; with its integrated video and audio, it made the perfect companion for the Celeron.

Until recently, the Duron didn't have a similar claim to fame, thus limiting its success in the retail market.  The enthusiast community recognized that the Celeron was crippled by its 66MHz FSB and appreciated that the Duron could be used as a value solution offering high-end performance.  With the release of the VIA KM133 chipset and the SiS 730S, the Duron is given the same opportunity to succeed at the Celeron, and thus can finally threaten the Celeron's retail sales. 

In response, as we predicted in our recent roadmap articles, Intel is giving the Celeron what it has desperately needed, a 100MHz FSB.  This makes the Celeron 766 the last 66MHz FSB Celeron, and the Celeron 800 the first 100MHz FSB Celeron, which is being announced today.

The Chip

The Celeron 800 remains unchanged from the previous Celeron processors.  It is still manufactured by disabling half of the cache on low-yield Pentium IIIs that have bad cache blocks, thus giving the Celeron half the cache as well as making that cache half as associative as the Pentium III’s cache.  This ends up hurting the Celeron severely as it, clock for clock, is unable to outperform the Pentium III, even when both are clocked at the same FSB/memory clock. 

The explanation for this is simple, the less associative a processor’s cache is, the lower hit-rate it has, and the lower the hit rate, the more times the CPU has to go to main memory to fetch needed information.  For more information on cache mapping techniques, check out our one page guide here.  

Celeron 800 Specifications

·        0.18-micron Coppermine128 core

·        800MHz clock speed – 8.0x clock multiplier

·        32KB on-die L1 cache running at core speed

·        256-bit Advanced Transfer Cache – 4-way set associative 128KB on-die L2 cache running at core speed

·        Advanced System Buffering

·        370-pin FC-PGA Socket-370 GTL+ CPU interface running at 66MHz

·        1.70v core voltage

CPU Specification Comparison
	AMD Duron	AMD Athlon			Intel Pentium III		Intel Celeron
Core	Spitfire	K7	K75	Thunderbird	Katmai	Coppermine	Mendocino	Coppermine128
Clock Speed	600 - 800 MHz	500 - 700 MHz	750 - 1000 MHz	750 - 1200 MHz	450 - 600 MHz	500 - 1000 MHz	300 - 533 MHz	533 - 800MHz
L1 Cache	128KB				32KB
L2 Cache	64KB	512KB		256KB	512KB	256KB	128KB
L2 Cache speed	core clock	1/2 core	2/5 or 1/3 core	core clock	1/2 core	core clock
L2 Cache bus	64-bit					256-bit	64-bit	256-bit
System Bus	100 MHz DDR (200 MHz effective) EV6				100 - 133 MHz GTL+		66/100 MHz GTL+
Interface	Socket-A	Slot-A		Socket-A Slot-A (OEM only up to 800MHz)	Slot-1	Slot-1 Socket-370		Socket-370
Manufacturing Process	0.18 micron	0.25 micron	0.18 micron		0.25 micron	0.18 micron	0.25 micron	0.18 micron
Die Size	100mm^2	184 mm^2	102mm^2	120mm^2	128mm^2	106mm^2	153mm^2	106mm^2
Transistor Count	25 million	22 million		37 million	9.5 million	28 million	19 million	28 million

Testing the Chip

We have split the performance section of this review into two separate areas.  The first set of tests will illustrate the performance of the Celeron 800 in our standard CPU test bed.  The second set of tests will illustrate the performance of the Celeron 800 in a simulated system configuration which will be very close to what you will be able to find in retail/mail order stores. 

The two sets of tests are targeted at two different audiences.  The first set is intended for those that intend to custom build their own setups using one of the compared processors.  The second set is intended for those that will be shooting for the lowest possible price with a configuration without primary concern for performance, mainly the OEMs and system integrators as well as those looking to purchase a retail system based on one of the compared processors. 

Kicking off the performance tests, we have SYSMark 2000.  The first thing we notice is that the Celeron 800 isn't lingering below with the rest of the Celerons.  It's 100MHz FSB gives it the advantage it needs to offer, for the first time in quite a while, respectable performance. 

Granted, the Celeron 800 is still 9% slower than the Duron 800, however it is no longer trailing AMD's "slowest" Duron offering.  In fact, the Celeron 800 is approximately 6% faster than the Duron 600 which is very impressive for a CPU that has half the FSB bandwidth and 64KB less cache than the Duron. 

Business Winstone 2001 doesn't change the picture too much at all.  While SYSMark 2000 is focused around measuring system performance while running a single application, Business Winstone 2001 is more representative of a power user's machine where multiple applications are running at once. 

In this case the Duron 800 is almost 14% faster than the Celeron 800 which happens to be around 4% faster than the Duron 600.  More importantly, however, is the comparison between this new 100MHz FSB Celeron and the 66MHz FSB 766MHz part.  A 4% increase in clock speed (766 > 800MHz) resulted in a 6% increase in performance, indicating a better than linear increase in performance. 

Can you begin to see how much the Celeron was being limited by its 66MHz FSB?

Again the Duron 800 holds a healthy lead over the Celeron 800, this time of just over 8%.  And once again the Celeron 800 is able to outdo its older siblings by offering performance superior to that of the Duron 600. 

In comparison to the Celeron 766, the same 4% increase in clock speed this time results in a 9% increase in performance.  In content creation applications, the Celeron must have really been hurting with its 66MHz FSB. 

Gaming performance changes the picture slightly.  Games such as Quake III Arena are more FSB dependent than our business and content creation tests from earlier, and thus the Duron gets the major advantage here. 

The Celeron 800 still manages to provide a 24% increase in performance over the Celeron 766, in spite of only boasting a 4% actual clock speed increase (proving our theory regarding FSB dependency in games like Q3A). 

However that 24% increase in performance is only enough to bring it up to the speed of a Duron 600 whose superior L2 cache subsystem (read our explanation on the Duron's L2 vs Celeron's L2) and 100MHz DDR (effectively 200MHz) FSB give it the edge in games like Quake III Arena.

Bumping up the resolution a bit to 1024 x 768 x 32, where performance would normally be determined solely by your graphics card, the Celeron is still only able to offer performance close to that of a Duron 600. 

The Celeron 800 does come out 14% faster than the 66MHz FSB Celeron 766.

UnrealTournament, historically a more memory bandwidth intensive game paints a very similar picture.  The Celeron 800 and Duron 600 are offering equal performance, both about 10% slower than a Duron 800 however.

The scene at 1024 x 768 x 32 doesn't change too much since the benchmark is not quite video card limited just yet. 

MDK2 is very similar to Quake III Arena in the manner in which it illustrates solid performance.  And again we see that the Celeron 800 and the Duron 600 are almost identical in performance, both being over 10% faster than the second "fastest" Celeron at 766MHz.

As video card limitations kick in, the Celeron 800 and Duron 600 are perfectly stuck at 99 fps. 

And to round off our gaming performance tests, Expendable continues to support the same results we have seen throughout the test suite: The Celeron 800 is approximately as fast as a Duron 600. 

In spite of the Celeron 800's 100MHz FSB, and the performance gains it has shown thus far over the original Celeron 766, a faster FSB frequency can only feed the processor more data, it cannot change the processor's architecture.

The Duron, being a direct derivative of the Athlon, can flex its powerful FPU muscle in SPECviewperf and continue to outperform even the fastest Celerons.  So while the Celeron may have caught up in performance to the Duron in most home/office applications, it is still not a value priced, high-end solution like the Duron can be passed off as. 

The System Test

Testing the processors in a configuration you would find in a retail or OEM system paints a much different picture.  Now that the processors have to rely on the integrated video and potentially sub-par memory controllers of their platforms, the performance difference between the Celeron 800 and the Duron 800 is next to nothing. 

For most do-it-yourselfers that will be building their own systems this doesn't matter, since you would most likely be pursuing a different platform for your processor if performance was your goal. 

But for those of you that will find yourself scouring the store shelves for a pre-built system, the Celeron 800 and Duron 800 will be indiscernible from one another other than in price. 

Things get even worse for the Duron here.  As we showed in our KM133 review, the integrated video of the KM133 chipset is hindering its performance considerably as the Celeron 800 is actually able to pull ahead by a decent amount. 

Luckily the SiS 730S manages to save face as it offers performance that is identical to Intel's 815 platform that the Celeron is tested on. 

The largest performance difference we have seen between the Duron 800 and the Celeron 800 in these OEM/retail system benchmarks is the 3% gap we see here. 

For those of you that are in the market for this type of a pre-built system, you cannot expect to get much in terms of gaming performance.  But for those of you that are interested in seeing what these low-cost setups can do, the performance is determined almost exclusively by the integrated graphics of the particular platform.  This gives the Savage4 3D core of the KM133 chipset the advantage and the 8 - 12 fps lead over the competition here. 

Because all of the above platforms are UMA based (the integrated video uses system memory for its frame buffer), the performance drops considerably when simply increasing the resolution to 800 x 600.  However the standings remain the same. 

UnrealTournament, not being as torturous on the integrated video as Quake III arena brings about a closer performance range from all of the offerings as it is mainly affected here by memory bandwidth (which is compromised in a UMA based system). 

When you're dealing with a sub $800 system, you don't expect to see the 100 fps performance figures we're used to reporting on, it's back to the days of 30 fps goals here. 

Finally, a CPU utilization comparison while playing back a DVD stream shows that DVD playback is more platform/video core dependent than it is CPU dependent, as long as the proper features are implemented.  In the case of the SiS 730S, its support for hardware motion compensation and iDCT give it the tremendous advantage here while the KM133 lags behind severely.

The i815's integrated video does not offer the same support as the 730S however its driver support is significantly better than the KM133's, giving it the middle of the road in terms of performance. 

Final Words

We have said it for a seemingly endless period of time, the Celeron needed this 100MHz FSB upgrade much earlier.  But sometimes, when you're making products for an incredibly large population a sacrifice has to be made in favor of preserving profits and that sacrifice often is performance.  This is the sacrifice Intel made by keeping the Celeron as a 66MHz FSB part, however it doesn't really cost any more for them to produce a 100MHz FSB Celeron, the main reason for doing this was to keep the Celeron as far away from the Pentium III's performance level as to not threaten Intel's flagship product. 

For those of you that are building new systems from scratch, even the new 100MHz FSB Celeron 800 isn't able to come closer than offering 90% of the performance of the Duron 800.  Not to mention that it is a more expensive chip, making it a less attractive option than a Duron, especially considering the maturity of current KT133 based motherboards. 

While we concluded our first review of the Duron with the statement that it had restored the idea of a low-cost processor performing like a high-end speed demon, the Celeron 800 unfortunately doesn't meet that characterization.  As the SPECviewperf performance analysis illustrated, the Celeron 800 is still classified, at best, as an entry level desktop PC processor whereas the Duron could very easily pass as a workstation solution.

But those that were planning on using a processor for such tasks already knew that the Celeron wouldn't be the answer.  We proved early on that an 850MHz Celeron with a 100MHz FSB wouldn't be able to offer the same type of performance as a similarly clocked Duron in most high-end applications, so there's no reason to continue to argue the point. 

In terms of business and home/office application performance, the Celeron 800 is definitely much more competitive than any other Celeron ever was, simply because of its 100MHz FSB.  In these types of applications, the Celeron 800 is slightly faster than a Duron 600, and often approximately 10% slower than an equivalently clocked Duron.  In games, the Celeron 800 performs identically to the Duron 600, and again approximately 10% slower than the Duron 800. 

For OEMs and System Integrators, provided that we are talking about using processors on their appropriate value platforms (KM133/730S for the Duron, i810E2/815E for the Celeron), the Celeron 800's performance is indiscernible from the Duron 800.  If you happen to be walking down the isles of a retail computer sales store, chances are that the value positioned Duron systems will be using one of the two aforementioned chipsets, bringing the Duron's performance down to the level of the Celeron, making the only reasons to go one way or another with your purchase: price, system configuration, and upgrade path. 

Stepping back again to take a look at the big picture, Intel just gave the Celeron exactly what it needed to remain competitive, a 100MHz FSB.  It's definitely not too little too late, although we would have definitely liked to have seen it a bit earlier.  The Celeron has the retail advantage of have a much more mature value chipset that it can be paired up with, it's AMD's job to nullify that advantage if they are going to break into the market that the Celeron has a very tight grasp on.