![](/Content/images/logo2.png)
Original Link: https://www.anandtech.com/show/237
Just months before the release of Intel's "revolutionary" step towards cost-effective microprocessors, the cacheless Celeron 266 received some of the most intense levels of media-bashing the company ever experienced. The reasoning behind the intense hatred towards Intel because of their motion to remove all cache from the cache-filled Pentium II design they so devilishly advocated just a year before was because of a trivial argument stating that the business application performance of the Celeron would be too poor to be considered a viable option. | ![]() |
Before ever seeing the likeness of Intel's first Celeron clocked at 266, most users were convinced that the "Celery" would be a huge flop in sales. |
The general prejudice of the public towards Intel's Celeron almost guaranteed it to be a failure, and in terms of overall sales, the initial Celeron processors without any L2 cache were nothing but marketing failures. However everyone makes mistakes, including the microprocessor giant Intel, and quickly learning from their mistakes Intel took no longer than a few months to outfit their once cacheless Celeron with 128KB of L2 cache running at clock speed and re-release the "flop" as the Celeron A, the "A" being tacked on to the end of the name to signify a difference between the original and the newer Celerons, and in the eyes of the public, an illustration of a well deserved letter grade on behalf of Intel.
The overclocking population was taken by surprise when the cacheless Celeron 266, albeit clock-locked, could be taken up to 400MHz and beyond depending on the specific case. This was quickly denounced as a fluke related to the fact that the cacheless Celeron didn't have any clock speed limiting L2 cache on-board to prevent it from being overclocked, and for the time, this justification stuck; until Intel dropped another bombshell on the market, the, now famous, Celeron 300A. As mentioned before, outfitted with a full 128KB of L2 cache running at clock speed, the Celeron 300A shocked the community once again with its ability to hit 450MHz (in most cases) and definitely achievable without being limited by the L2 cache.
The success overclockers received with the Celeron 300A was tremendous, however in the big picture, the overclocking population represents, at most, 5% of the entire PC processor purchasing community, making the 300A's golden secret a hidden treasure to most. The justification for the 300A's overclocking success among those that did happen to uncover the treasure was Intel's extremely high processor yields (the "yield" is essentially the ratio of processors physically produced to those that work properly) in recent times. The question was raised as to whether or not Intel could simply remark the 300A as a 400MHz part and sell it a few months later without having to produce any new processors at all. In theory it was definitely possible since the 300A's were already making their way up to the 450MHz with relatively few problems, and those that did have problems usually only needed a quick bump up to a higher core voltage and they were on their way. Wouldn't it make sense for Intel to stay on the safe side and release a Celeron A processor at 400MHz rather than ramp up production on a higher yield Celeron A at 450MHz? It definitely does make sense, but before we start making any assumptions we'll want to take a few steps back and define the Celeron.
Taking a few steps back: Defining the Celeron
What makes a Celeron a Celeron? At first, as mentioned before, it was the processor’s obvious lack of L2 cache, however since its initial introduction things have changed considerably. How important is L2 cache to the day to day usage of your computer, can you really survive without it? The answer is yes.
Although you’ll always hear someone commenting on how their processor has more cache and is therefore better, the fact of the matter is that the 512KB of L2 cache present on the current generation of Pentium II processors and the upcoming Pentium III processors is more than enough to suit our needs. Why have it then? Because, from a marketing point of view, saying you have 512KB of L2 cache while the competition is playing around with less than < that number is a better investment, especially considering that adding the extra cache does not increase the overall cost to manufacture the processor by too extreme of a degree. At the same time, Intel had no idea how the market would react to the grand "more cache at a slower speed vs less cache at a faster speed debate," therefore, to stay on the safe side, the Pentium II was outfitted with a hefty amount of cache operating at a reasonable frequency.
What is the difference between 512KB of L2 cache running at = clock speed and 128KB of L2 cache running at clock speed? In terms of performance, very little. If a CPU attempts to access data located in the cache and fails during the request (a cache miss), regardless of how much L2 cache is present and how fast it is operating, the performance of the system (overall) will not vary as a direct result of the cache size/speed. The only time performance would be biased towards one of the two processors would be in the event that the CPU succeeds in retrieving the data (a cache hit), in which case the CPU with the faster cache would prevail. The thing to note is that, in spite of the faster operating frequency, the L2 cache in both cases affects performance virtually identically in that a 400MHz processor with 128KB of L2 cache running at clock speed would perform as a generally equal competitor to a 400MHz processor with 512KB of L2 cache running at 200MHz (1/2 clock speed). This "rule of thumb" applies in virtually all cases, except for server situations, where the basic rules of computing and performance do not necessarily apply, however for the sake of simplicity, servers will be eliminated from the comparison.
In the days when the 333MHz Celeron A was the fastest Celeron processor out, power hungry users still had a reason to purchase a Pentium II, as it was offered in clock speeds in excess of 400MHz, and for those that didn’t overclock, that was a considerable difference – this was Intel’s marketing plan for the caste system among their processors: the Celeron would be limited by an absurdly low clock speed (comparatively speaking), and the Pentium II would lead the pack with a much higher (and more expensive) offering. For those that overclocked, the 300A running at 450MHz was already faster than the fastest Pentium II, making the Celeron Intel’s one "undiscovered" hit wonder.
As soon as AMD began to catch up in clock speed, Intel gave the go-ahead, and announced the resignation of the Slot-1 300A and the entry of the PPGA (Socket-370) 300, 366, and 400 processors, with the latter two making their way into the slot-1 market as well. In actuality, the only change Intel probably had to make to get the 366 and 400MHz parts out the doors of their fab plants was resetting the clock locks on the processors to support the 5.5x and 6.0x clock multipliers necessary for the two processors. From a marketing standpoint, the introduction of a 366MHz+ Celeron would mean more intense competition for AMD at the low end of the price spectrum of the market; and if the Celeron would attempt to interfere with the Pentium II’s sales, the response any salesman at a retail chain would give a questioning buyer stuck in the Celeron 400 vs P2 400 debate would be, "the Pentium II 400 uses the 100MHz bus, which is a full 50% faster than the 66MHz bus the Celeron uses." Too bad most people that don’t know that the real performance difference is negligible will let that statement ride and spend the extra cash on the Pentium II 400 system. This is why the Celeron 400 runs at 66MHz x 6.0 while the Pentium II 400 runs at 100MHz x 4.0, purely marketing reasons.
As far as official specifications are concerned, the Celeron 400A is clocked at a 66MHz FSB frequency (as referenced above), using a 6.0x clock multiplier, at a core voltage setting of 2.0v and does not feature the embedded serial id number the Pentium III "boasts" having.
The Locks
How did Intel make sure that a Celeron always ran at a 66MHz Front Side Bus (FSB) frequency and a Pentium II always ran at a 100MHz FSB? The easiest way possible, making the individual chips locked to a certain multiplier, making operation at any FSB other than its rated frequency, extremely difficult if not impossible; this is Intel’s most popular combating force towards remarking (selling processors rated at lower speeds as if they were higher speeds and charging more for them, i.e. selling a Pentium II 300 as a Pentium II 450 for twice as much) known as clock-locking. You will notice that "clock-locking" is never referred to as an overclocking deterrent, this is because, Intel never truly had any objections to users overclocking their chips, it is just those that insisted on making an unlawful profit on such experiments (i.e. remarkers) that would give Intel a bad name and therefore forced them to pursue this avenue of escape. If a consumer spent a horrendous amount of money on a Pentium II 450, and it turned out that the processor was nothing more than a 300MHz Pentium II remarked as a 450, and it was sold to them by a fly-by-night company the only group left to blame would be Intel - this was something Intel would definitely not stand for.
The other type of processor "locking" that has received its fair share of media coverage has been "bus" or frequency locking which, in theory, would involve a modification to the processor to allow it to run within only a specified range of clock speeds derived directly from the FSB setting it was run at. The humorous part of the frequency locking discussions we've all been having for quite some time now is the fact that Intel never publicly admitted to pursuing such an avenue with their processors, although Intel will continue to implement restraints on remarking with each modified processor they kick out of their fabrication plants, frequency locking is one that has never been admitted to. Will future Intel processors be frequency locked? They may be, however for now, as far as the current generation of Intel processors are concerned (including the Slot-1 Celeron 400A), there is no implementation of frequency locking in practice by Intel. The Pentium III was originally stated as being the first frequency locked processor from Intel, however with its release less than a month away, and still no official word from Intel, the last week of February should be an interesting one. Intel's current samples of the Pentium III are shipping without any frequency restraints, and a quick search among on-line vendors will reveal that OEM Pentium III processors are in fact, available and in stock, although the quality and reliability of the vendors that offer them is up to your own experimentation, at your own risk. Frequency locking may become a necessary evil in the future, but for now, you're safe, even with the Celeron 400.
Socket vs Slot: What to buy?
As most of you are probably already aware of, the Celeron 400A is available in two distinct flavors, a PPGA Socket-370 version (Plastic Pin Grid Array, for use with Socket-370 motherboards) and a SEPP Slot-1 version (Single Edge Processor Package, for use with Slot-1 motherboards). The performance, reliability, and stability of the two 400MHz parts are identical (not taking motherboard designs into account), and the only real difference (other than the physical) between the two processors lies in the price difference.
The prices of the two processors, at the date of publication, actually reflected an inverse relationship than how things should be according to Intel. The physical size of the Slot-1 Celeron makes it more expensive to produce, however with the production of the Slot-1 300A having gone on for such a lengthy period of time, and the relatively few differences between a 300A and a 400A, the Slot-1 400A processors are currently cheaper than their Socket-370 counterparts. In a few weeks time the pricing structure should reflect Intel's "goal" for the Socket-370 vs Slot-1 argument, with the PPGA Celeron processors weighing in at a lower overall cost than the Slot-1 processors, but for now, the Slot-1 chips are simply cheaper.
What about motherboards? Socket-370 motherboards will generally cost $5 - $10 less than their Slot-1 counterparts, and Socket-370 boards will also offer low-cost 440ZX based solutions at a price point up to $15 cheaper than comparable BX Slot-1 motherboards, but for the most part, once again, the price difference between Socket-370 and Slot-1 motherboards is negligible. ABIT's overclocking wonder boy, the BH6 as well as the newly release BX6 Revision 2.0 are both under the $130 mark, with the BH6 often going for around less than $110. Until the prices of Socket-370 Celeron processors can come down another notch or two, there is absolutely no reason to opt for a Socket-370 MB/CPU combo at this point. The guaranteed upgrade path to a Pentium III is another benefit Slot-1 motherboard owners will hold over Socket-370 board owners, as Intel has yet to publicly disclose their plans for the future of Socket-370.
Socket-370 and Slot-1 400A processors overclock to pretty much the same limit, the latter being a bit more friendly in that respect. Although the 400A is clock-locked as described above, it does allow for an increase in FSB frequency to obtain a higher clock speed. Interestingly enough, the overclocking sweet spot for the 400A, like the 300A, seems to be 450MHz, or 75MHz x 6.0. AnandTech's tests illustrated that clocking the 400A at 500MHz, or 83MHz x 6.0, was not nearly as stable as the system at 450MHz, and the system wouldn't boot at all at the ridiculous 600MHz setting (100MHz x 6.0). The limitation of the Celeron processor in this case isn't because of the L2 cache, rather the processor's die itself, just as the physical limit for the AMD K6-2 was around the 500MHz mark (even with Kryotech's supercooled, -40 degree Celsius Cool K6-2 system), the physical limit for the Intel Celeron A seems to be a little above 450MHz. Does this mean that settings above 450MHz simply won't work with Celeron processors? Absolutely not, it means that the Celeron, in general, as it currently stands will probably never make it up past 500MHz in a reliable manner and in mass quantities. If Intel tweaks their production of the processors a little further then things may change, but, for now, this is how it is. Cooling the processor efficiently isn't the limiting factor here, it's the processor itself that can't make it that high (generally speaking, there will be exceptions to the normal, although very few).
Performance
The Socket-7/Super7 Test System Configuration was as follows:
- AMD K6 233, AMD K6-2 350, AMD K6-3 450 (engineering sample)
- FIC PA-2013 w/ 2MB L2 Cache
- 64MB PC100 SDRAM
- Western Digital Caviar AC35100 - UltraATA
- Canopus Spectra 2500 TNT AGP Video Card (16MB)
The Pentium II comparison system differed only in terms of the processor and motherboard in which case the following components were used:
- Intel Celeron 300, Intel Celeron 300A, Intel Pentium II 400, Intel Pentium II 450
- ABIT BH6 Pentium II BX Motherboard
The following drivers were common to both test systems:
- nVidia Reference Drivers 0.48
- DirectX 6.1
The benchmark suite consisted of the following applications:
- Ziff Davis Winstone 98 under Windows 98 & Windows NT4 SP4
- Ziff Davis Winstone 99 under Windows 98 & Windows NT4 SP4
- Ziff Davis Winbench 99 under Windows 98
- Quake 2 v3.19 (for 3DNow support) using demo1.dm2 and Brett "3 Fingers" Jacobs Crusher.dm2 demo
All Winstone tests were run at 1024 x 768 x 16 bit color, all gaming performance tests were run at 800 x 600 x 16 bit color. 3DNow! support was enabled when applicable.
Windows 98 Performance
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image49.gif)
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image50.gif)
As you can see, under Windows 98, the Celeron 400 is just about as fast as the Pentium II 400, making the Pentium II 400 useless for most users. Although high-end workstations will still want to opt for Pentium II processors (in situations where the larger cache does count), most users out there, including "high-end" gamers will want the 400A over the old-school Pentium II 400.
Super7 users will find that the K6-3 is a better investment, provided that AMD can crank the chips out at a low enough cost to prevent their loyal followers from picking up a cheap 400A/MB combo, as the K6-3 will top the charts well into the release of the Pentium III. With a 500MHz K6-3 going up against a 500MHz Pentium III, the winner in terms of business application performance under Windows 98 does turn out to be AMD with the K6-3 by a very small margin, but one great enough to convince most Super7 users to hold on to their boards for a little while longer.
Windows NT Performance
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image51.gif)
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image52.gif)
Windows NT is the realm of high performing cache subsystems, it is for this reason that the K6-3 450 takes the lead over the Intel Pentium II 450 and the Celeron clocked at 450MHz as well. The 400A does provide quite a bit of competition though, especially considering it is considered to be a "low-cost" competitor.
Gaming Performance
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image53.gif)
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image54.gif)
Die hard gamers will find Intel's Celeron to be unbeatable in value and performance, AMD is held back in this area by poor implementation of 3DNow! in the latest video card drivers and therefore the reliance of most 3D accelerators on the relatively poor floating point performance of even their upcoming K6-3.
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image55.gif)
Things do change when 3DNow! mode is enabled under Quake 2's software renderer, as the K6-3 quickly closes the gap between itself and Intel's Celeron 400. The best overall gaming value is still the Celeron in spite of AMD's efforts with 3DNow! support.
High End Performance on the Low End
Using the same tests AnandTech conducts on all multiprocessor systems, the Celeron 400 was put to the test against the 300A, the 300 (with no L2 cache) and AMD's upcoming K6-3 to show how the performance standings change near the high end of the processing spectrum, with the only varying component being the amount of RAM used.
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image56.gif)
With 64MB of RAM, the 400A doesn't really distance itself from the competition all too much, however as anyone familiar with Windows NT already knows, the more RAM you equip a system withwell
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image57.gif)
the larger a performance increase you experience. The 400A slowly begins to break away from the competition (particularly the K6-3 400), a gap which grows even further in some cases with 256MB of SDRAM. The true performance differences seem to come out in Photoshop and image editing situations, while the performance under CAD/drafting and development environments between the Celeron 400A and the K6-3 400 is virtually identical. The bottom line seems to be that for image editing, the Celeron 400A is without comparison in terms of a price to performance ratio, even taking AMD's K6-3 into account and factoring in the fact that the K6-3, for Super7 users, won't require a motherboard upgrade. For all other high-end users, the K6-3 offers a similar value to the Celeron 400, so if you're currently a Super7 user, and not a graphics artist, the K6-3 is probably your best upgrade path. Otherwise, the Celeron 400A is looking pretty tempting with its sub $160 - $170 price tag, a quote which should drop to around the < $100 level within the next 3 months.
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image58.gif)
![](http://images.anandtech.com/old/cpu/intel-celeron-400a/image59.gif)
Conclusion
If you can get your hands on a retail Celeron 300A, then that is most likely your best bet as the retail 300A's can generally make it up to 450MHz with a bit of tweaking (see AnandTech's Celery Report). If you find yourself being beaten to the punch by Intel's efforts to cut off our beloved 300A supply, then the 400A makes for a fine friend. The 400A should be opted for above the Pentium II 400 or 450 processors, as the performance difference between the Celeron and its two older brothers is negligible and definitely not worth the money for most users. The Socket-370 version of the 400A is a waste of time to even bother with right now, if the price drops to at least $20 below the cost of the Slot-1 400A then it may be worth a slight consideration, but for now, you can pick up a SEPP 400A for at least the same price as a PPGA 400A, big mistake Intel. If Intel had waited until production of the PPGA chips was at a higher level, then things might have been different, but they're not, so for now, Slot-1 is the way to go.
What happens if you're a Super7 user? Luckily enough, there is a guaranteed upgrade path for you all, it's called the K6-3. If you're fine with the Super7 platform then the K6-3 will be the answer to your prayers, provided that AMD can release the chip on time. Expect the K6-3 400 to retail for a more than Intel's Celeron 400A, however don't expect it to retail more than the combined cost of a 400A and a motherboard, making a K6-3 upgrade a viable one considering it only requires, at most, a BIOS upgrade for Super7 users to achieve full compatibility.
If you're purchasing a new system, or building one for that fact, the Celeron 400A is essentially a 400MHz Pentium II at a much lower cost, and is definitely a better option than even AMD's K6-2 400. The K6-2 400, at the time of publication, remains around $10 cheaper (all price estimates are courtesy of www.pricewatch.com) than the Celeron 400A, and Super7 motherboards are generally $20 - $30 cheaper than Slot-1 BX motherboards, making the price difference between a brand new K6-2 400 system at most, $40 - $50 cheaper than a Celeron 400A system. Not too great of a price advantage there.
The bottom line is this, for a new system, Intel does have the best bang for your buck currently with the extremely aggressively priced Celeron processors, however the future of the PPGA (Socket-370) Celeron processors is too shady to make a full speed ahead decision on a Socket-370 system just yet. Super7 has a future with the K6-3, but it is one that will come to an end eventually with AMD's migration to Slot-A and the highly anticipated K7, making the K6-3 a good alternative upgrade, but giving the Celeron the thumbs up as the most cost-effective and best overall performing upgrade in its price class for everyone but Super7 users. If only all Intel chips were this affordable…