There are some rare occasions in life when a confluence of events occurs that serves to provide a sum benefit that far exceeds the total of its components. Or, as a friend of mine is fond of saying: Sometimes great ideas arent released, they escape!

This is the case with Intel Celeron CPU.

Intel is a giant company and, as such, is sometimes a bit slow to react to emerging trends. Along comes AMD, first out of the starting gate with a 233Mhz CPU (the K6) and the knowledge that there was a huge untapped opportunity in the sub-$1000 PC market.

But, simply having a great product at a great price is not always enough. It was the willingness of some of the big players, such as IBM, to move to AMDs product line that really got the ball rolling for AMD. Suddenly, Intel woke up one morning to find that this upstart (not really, AMD has been in business 30 years) had taken a big bite out of the CPU market.

What to do? Like the car maker who finds that luxury automobile sales have slowed, and comes out with a stripped-down economy model, Intel produced the Celeron. The Celeron is Intels attempt, and a rather successful one, to get back in to the "economy" PC business.

This is the first in a series of reports on the success, or lack of, in overclocking the Celeron CPU. It will appear on a regular basis as long as these pups can be stretched to the performance limits. This first report will be longer than most, as I will be covering some basics and background that wont be necessary in future installments.

You Master Celeroners (hey, I think I may have coined a phrase there!), can feel free to skip this section. Well cover a little history, some basics about the CPU as well as the genesis of this historic tweaker opportunity.

As most of you know, the Pentium 2 series of processors are a slot design encased in a plastic cartridge, containing off-die, half-processor speed Level 2 cache on board. This is a rather expensive CPU to produce. In order to cut costs and be competitive in the burgeoning "Cheap PC" market, Intel removed a few frills from this design and came up with the Celeron. Among the changes: No plastic cartridge and no L2 cache; in other words, a stripped down P2. Further details of the release of this CPU can be found in Anand's Excellent report.

Celeron 266

The original "cacheless" CPU came in a 266Mhz as well as 300Mhz variety and was almost universally derided as a "crippled" chip. It was quickly nicknamed the Celery as a way for the "Techno-Sophisticates" to show their disdain for the product. I was among those who laughed at this CPU and what I thought was a lame attempt by Intel to get back in the game. I couldnt help flashing back to the 486SX days and feeling that they were, once again, trying to play us for suckers.

Well, as usually happens, the Geeks of the world united and began tweaking this new product. It was soon discovered that the lack of L2 cache resulted in an interesting phenomenon. These chips took to overclocking better than any previous CPU in history!

I can remember my sense of achievement at discovering that many of the AMD 5x86-133 CPUs (the ADZ stepping) could be successfully overclocked to 160Mhz (4x40). Yet, this is only a 20% performance increase. Shortly after the introduction of the 266Mhz Celeron reports were coming in from all corners of success at 400Mhz. This is a jump of better than 50%!

While I was certainly interested in this kind of performance, my natural bias against Intel caused me to dismiss this as a fluke and nothing to get too excited about. Personally, and as a reseller, Ive been a long-time staunch supporter of the "alternative" chipmakers; as a matter of fact, I hadnt had an Intel CPU in my own system since my 386 days.

So, to my later regret, I was slow to recognize the potential, both for performance and profit.

Not long after the introduction of the cacheless Celerons, Intel produced the "A" version of the chip with 128K of L2 cache residing directly on the die of the CPU. It came first in 300Mhz (300A) and then in 333Mhz (333A). The second of these is not worth buying. More on that later. Once again, Anand was right on top of this release with his excellent report.

Celeron 300A

This was a performance-enhancing feature both for the fact that the chip now had cache and the fact that this cache was operating at full chip-speed. But, many in the "Geek" community assumed that this would reduce, or even, eliminate the overclockability. Add to this the fact that Intel has clock-locked these chips, and many of assumed the run was over.

Suddenly, we had a dirt-cheap alternative that, in many cases, performed on par with its big brother, the Pentium 2, 450Mhz. Think about it, when set up at the clocked locked multiplier of 4.5 on a 100Mhz FSB, for a fifth of the price (currently) one could have 450Mhz performance!

As you can see from these graphs, this performance/price ratio is simply the best bargain we have ever seen in the PC world:

Courtesy AnandTech, 1998

Courtesy AnandTech, 1998

There are actually 14 distinct sSpec numbers for the Slot One Celeron CPU, but most are not currently in production. Here is the information for the flavors of the 300 and 333:

Source: Intel Corporation

You will note that the CPUID and core stepping are identical in all cases. That's why the assertion that the Retail Boxed CPU's are better overclockers makes no sense. It's also why, as I noted earlier, it's not a good idea to run out and buy the 333Mhz flavor since the only difference between it and the 300 is the multiplier. Since it is rare for a 300 to hit the 500Mhz mark it is also rare for the 333, since it is clocked-locked at a multiplier of 5.

But, there was still one minor problem. In order to overclock to this level, many times the CPU would need an increase in the core voltage.

Most Slot One motherboards auto-detect CPU voltage and set it without user intervention. So, what happens to those who are stuck with a chip that requires this goose in the voltage? Well, some smart Geek somewhere figured out that by taping specific pins on the CPU, one could trick the motherboard in to setting the voltage higher. But, this is a tedious procedure, not for the faint of heart, and is limited to only .2 increments.

Up to the plate steps Abit with the BH6 and its soft menu feature that lets you, the end user, control the voltage in increments as low as .05. Now, all the ingredients are in place, and now, I finally woke up to the potential in this combination.

It started with a couple of customers e-mailing to ask if I wouldnt mind testing some out at 450Mhz on the BH6 prior to the sale. Word spread and I soon found myself offering this combo at my web site. As a side note, I myself crossed back over to the Intel side for the first time in years; this was just too much for a mere mortal to resist.

During the testing of these "450Mhz Combos", I was discovering information that I thought Id pass on to the community, so I started filing regular "Celery Reports". These caught the eye of our gracious host here at AnandTech, Anand Lal Shimpi, the man himself, and he asked me if Id like to do up these reports for publication. Of course I said YES! Well, on to the meat and potatoes.

I test each CPU under initial identical conditions prior to "burning in" my combos. I do an initial test run using the same Abit BH6, same RAM (single stick 64MB Micron, CAS3) and the same video card, a Trident 975, 4MB AGP. For those who are interested, this is a very fast 2D card that consistently scores better than 5 in Final Reality. Not very good in 3D, but for a business system, it’s a screaming bargain. I use an old 270MB Quantum hard drive (actually have 3 set up identically, so I can burn multiple combos at once). I use an old hard drive because I assume that these setups will be going in to a variety of situations with a variety of drives and that if it’ll work with this clunker, it’ll probably work anywhere.

I run a quick test at 450 using Business Winstone 97. Why such an old version, you might ask? Because I’m not testing for performance, only stability. This works perfectly for that purpose. If it fails, I bump the core voltage up a notch until it passes. If the CPU won’t do 450 or requires more than 2.2v core, I set it aside for later evaluation. After passing the first test on my evaluation board, I move the chip to the motherboard with which it will be mated.

I put each one through 6 full rounds of Winstone and 2 hours of Final Reality. By using these two tests, I’m covering both the 2D business users and the 3D gamers. As an aside, for those who haven’t used it, Final Reality is not only a great test, it is visually stunning. One glitch at current voltage, and it’s bumped up a notch, until it’s completely stable. Because I believe that stability means no excessive measures, I don’t call a chip good if it requires more than 2.2v core to stabilize. I like a little breathing room.

Each report will consist of two parts. First will be the results from testing since my last report, and second will be cumulative totals. I will also include any new discoveries, such as what I’ll shortly be revealing here for the first time.

This week I tested a total of 24 Celery sticks. These were all SL2WM (OEM), all Costa Rica. Before I move on, let me re-address a couple of myths. First, there is no difference in overclockability between Costa Rican CPUs and Malaysian CPU’s. If this were true, there would never be a failure with a Malaysian chip because the Costa success rate is very, very high. Second, the same applies to OEM versus Retail. They are exactly the same stepping and core. You do, however, get the longer warranty with retail.

I had two separate 12 unit batches of chips. Using the white label on the back as our guide, the first were week 46 and the second were week 45. Of the first, I’m further dividing in to 2 batches of 6; in a minute you’ll see why.

The first half of the first batch were a disaster (that may be too strong a term, they worked, just not at 450; getting a little greedyJ ). Four of the six CPU’s flatly refused to do the 450 dance and the other two required 2.3v to run stable. This is a very low percentage based on past experience. Of the second half of this batch, all did 450 quite easily at 2.2v or less.

Of the second group of 12, all did 450Mhz and eleven of them did it at default voltage of 2.0! I began to wonder why there was such a dramatic difference in performance both within the same week and between the two different weeks. I’ve previously reported encountering difficulty with CPUs that were mounted on a darker green shade of PCB substrate, but I now realize this is only part of the equation.

I now believe that while we have all been looking at the white label on the back of the CPU to divine the information necessary to determine our potential success, we have been looking in the wrong place! If you look at this graphic of the label, youll see that what we thought was the production week is listed as 34.

OEM CPU label
Ive now come to the conclusion that this is actually the packaging week and that the true production week is found on the front, upper-right of the chip.

Ive also discovered that a specific set of symbols is very critical in determining ahead of time if your odds of hitting 450 will be good or bad! I will try to illustrate what appears on the front right of the CPU and what the goods ones should look like:

(Symbol) NM2 9839

94V-0

or

(Symbol) RU M1

94V-0 9843

The first one the symbol looks sort of like a 4 leaf clover in an eye.

The second one the symbol has the word COMPEQ inside an oval. The "RU" is actually reversed and tilted to the left on the CPU.

Note the numbers "9839" and "9843". The first was from the batch labeled week 45 and the second was from the bad half of the group labeled week 46. I now believe that this is the true production week of the CPU.

Now, heres the stunner. The chips containing the first set of symbols were the ones that all performed and nearly all required no voltage increase. The second set of symbols were the dogs. I decided to pull out the CPUs Id had in the past that hadnt performed and, sure enough, they all had the second symbol, regardless of production week! For additional verification, I borrowed a half-dozen of variety two from my local distributor, and these were weaklings also.

I would also like to note that the deadliest combination seems to be the darker green PCB and the "COMPEQ" chips. To the best of my knowledge this has never been noted anywhere on the web!

So, after this weeks testing, the total is now 77 Celerys. Out of this weeks total of 24, 20 did the 450 dance, but 2 required 2.3v core to do it. This gives us a success rate of 83.3%, which is below the norm.

To-date, of the 77 tested, 68 were successful for 88.31%. Discounting this recent lousy crop, we were well above 90% You can bet Ill be avoiding the "COMPEQ" symbolled chips from now on!

Many rumors have been flying that the more recent slot version 300As have not been as overclockable as previous versions. I dont believe this to be true, unless Intel has switched production to the second symbol Ive outlined above. I have no idea what these two different versions mean, only that we need to stay away from the second one.

It will be interesting to see which CPUs are still in inventory, whether it be the "Cloverleaf" or the "COMPEQ". Stay tuned for future developments.

Russ Stringham, Owner
CompuCheap