It’s been a very long time since I’ve written an Editorial for AnandTech, mainly because there hasn’t been a need to.  But with the recent course of events surrounding the launch and later the recall of the 1.13GHz Pentium III, it is definitely time for a few comments from yours truly. 

Success is something that can be measured by a variety of factors, some consider wealth, material possession, and stature to all be measures of it, while for others, personality, integrity and honesty, especially when faced with adversity are the true measures of success.  The only true point of agreement most of us can come to is that in order to be successful, one must undoubtedly work very hard for it. 

Intel has definitely achieved a visible level of success.  The computer industry quite possibly wouldn’t be the same if it weren’t for the advancements that Intel has brought to the arena.  Their Pentium Pro (P6) processor was the first x86 CPU that could truly be positioned against the RISC (Reduced Instruction Set Computing) competitors and finally made the high end market take x86 processors seriously.  The Pentium II was the first x86 processor to bring a high speed L2 cache down to the desktop market, the CPU also paved the way for a number of variants to be produced.  The resulting Celeron gave new meaning to the value market segment, offering performance that was far from low-end, yet garnished a price tag that was definitely far from high-end.  There isn’t a doubt about it, Intel has definitely worked hard to get to where they are today, and the industry as a whole has definitely benefited from their presence. 

Throughout the majority of 1999, Intel had very little to worry about in terms of competition.  While the AMD K6-2 and K6-III processors were great for Socket-7 motherboard owners, most users that were looking to purchase a new system would have to get quite a bit of convincing before journeying down the Socket-7 path.  As more and more incompatibility problems began to surface between the two main Socket-7 chipsets with AGP support (ALi Aladdin V & VIA MVP3) and some of the most popular graphics chipsets (NVIDIA TNT2), the headaches associated with owning a Socket-7 system quickly became not worth it, especially with the Celeron positioned at a very affordable level.  Not only were AMD’s flagship solutions plagued with more incompatibility problems than their Intel counterparts, they were also the victim of sub-par FPU performance even when compared to Intel’s “low-end” Celeron processor.  AMD’s 3DNow! instruction set yielded some improvement in this category, unfortunately it wasn’t enough to give AMD the edge over Intel. 

Intel used to completely dominate, especially under games as can be seen from the above chart in our February 1999 review of the Pentium III

Intel was clearly enjoying the life of a success, with no major competition present in the industry, and a secure grip on the performance crown in value, desktop and performance market segments.  Unfortunately for Intel, this paved the way for AMD to come in and take quite a bit of the lime light with their Athlon processor that was released in August of 1999. 

On September 27, 1999, just under two months since AMD’s Athlon stole the performance crown from Intel, the next shocker came in the form of the delay of the elusive Camino chipset (i820).  The BX replacement we all had extremely high hopes for had been delayed “until further notice” and all of the sudden Intel’s recently introduced 133MHz FSB Pentium III ‘B’ processors had no platform to run on other than the i810E which was far from a performance platform. 

As we put it in our review of the i820 chipset, the delay was “a huge hit for Intel because this [was] the first time they’ve truly screwed up a chipset launch. Sure, there have been delays before, but never have they come within a few days of releasing a product and had to cancel the launch due to a severe problem with stability.”  At the time, we discounted it as simply a mistake on Intel’s part, and everyone makes mistakes at one point or another. 

Then came the issue regarding the i820’s virtually exclusive support for Rambus DRAM as a memory platform.  The lack of a real performance benefit courtesy of the i820’s RDRAM implementation left many users confused as to why RDRAM was priced between 3 and 5x the price of SDRAM yet offered a very small if not non-existent performance improvement. 

Following the Rambus fiasco, including a disgust many shared regarding Intel’s attempt to force RDRAM upon the market as well as a disappointment with the business practices of Rambus, the next major blow to Intel’s stature was incurred: a recall of their Memory Translator Hub chips.  The MTH was the only point of salvage for the i820 chipset, simply because, it was what allowed for PC100 SDRAM to be supported on i820 motherboards, unfortunately while incurring a very serious performance penalty because of the translations required by the chip.  Needless to say, at this point, users began to turn to VIA in order to supply them with chipsets to use their 133MHz FSB Pentium III processors with. 

The fourth disappointment Intel saw towards the beginning of this year was the eagerness of motherboard manufacturers to adopt platforms based on VIA’s KX133 chipset, an Athlon platform.  The Athlon was gaining momentum as well as market share, two things that Intel did not expect to see happen. 

Disappointment number five came in the fact that the price delta between the Pentium III and the Athlon began to increase to the point that Intel was not only losing chipset sales, but CPU sales as well.  A recent figure from an unnamed motherboard manufacturer indicated that approximately 55% of their motherboard shipments featured VIA chipsets, that’s an incredible number if you consider that just a year ago there was virtually no reason to go with anything other than the Intel BX chipset. 

Not only did prices of Intel processors begin to rise, but the supply of their processors seemed to drop.  Prior to Intel’s downward spiral, AMD would be the one we would accuse of “paper launching” processors, since you could never find a newly “released” AMD CPU until after its launch.  Intel’s policy was exactly the opposite, upon the introduction of a new CPU, systems based on that CPU would be available the very same day. 

Since the release of AMD’s Athlon, things have changed.  Slowly but surely the roles of the two companies have reversed, now, Intel is the one being accused of “paper launching” processors while AMD CPUs are readily available and definitely affordable.  These “paper launches” were at their worst with the release of the 1GHz Pentium III (March 2000) before the 850, 866 and 933MHz Pentium IIIs in an attempt to compete with AMD’s 1GHz Athlon that was released just days before.  What began to make the community characterize Intel’s CPU releases as “paper launches” was the fact that you couldn’t actually go out and buy a 1GHz Pentium III whereas, by the end of the month, the Athlon was already available in speeds from 500MHz up to 1GHz in 50MHz increments. 

The trend seemed to continue as Intel released the 1.13GHz Pentium III at the very end of July, while the 1GHz Pentium III parts were just finally becoming available in decent quantities, four months after the “release.”  The 1.13GHz processor was intended to give Intel the clock speed over AMD who had only been able to reach 1GHz with their Thunderbird.  Remember that in the OEM market, performance comes second to clock speed, it’s easier to sell a 1.13GHz system over a 1GHz system because the 1.13GHz system is “obviously faster.” 

We received a sample of Intel’s 1.13GHz Pentium III a couple of weeks before its launch on July 31, 2000, and during that period the processor effortlessly completed our usual suite of benchmarks and we readied the review for publication Monday morning.

Click to Enlarge

That day, a number of other online sites published reviews of the 1.13GHz Pentium III processors, seemingly without any problems, with the notable exception of Tom’s Hardware Guide.  Instead of a review of the processor, Dr. Pabst put together an editorial discussing Intel’s recent history of “paper launches” as well as the fact that Dr. Pabst’s production sample 1.13GHz Pentium III failed to work properly. 

Based on his experience, Dr. Pabst concluded that Intel had a serious production problem and was thus unable to reliably produce the Pentium III at 1.13GHz, his justification came with the fact that by underclocking the CPU to 850MHz the processor exhibited no signs of instability. 

At the time, we didn’t think much of the incident, other than assuming that Tom’s Hardware simply got a bad part.  We’ve received bad parts in the past, none from Intel, however we have had experience, especially with motherboards, where a production sample fails to work properly. 

There are a handful of people whose opinion highly respect in this community of ours, among those is Kyle Bennett of HardOCP.  I had the opportunity of meeting Kyle for the first time in person at this year’s Computex 2000 in Taipei and gained an instant respect for his ability to tell the truth, regardless of the circumstances.  As simple as that may be, you’d be surprised how often such a simple trait is absent from people these days. 

In any case, Kyle apparently had very similar problems with his 1.13GHz Pentium III sample and instead of writing a judgment on Intel’s production process, simply reserved to refrain from publishing any sort of review until he could get a reliable part.  After seeing that Tom had problems with his sample, Kyle came forward and mentioned that he had similar problems. 

Both Tom and Kyle included me on a CC of their email conversation regarding their experiences with the 1.13GHz Pentium III, so I decided to contact both of them.  I presented to both Tom and Kyle the facts that I saw, basically our 1.13GHz sample worked flawlessly in the tests we ran and I found it very odd that two Intel production samples failed in the hands of two individuals whose testing methods have always seemed to be sound. 

I offered Tom my working 1.13GHz sample so that he may get some benchmarks and complete a review of the processor.  After declining, I made the same offer to Kyle who accepted.  Kyle then contacted Tom and mentioned that he’d like to get a hold of Tom’s CPU which also failed most of the tests and demonstrate this in front of an Intel engineer. 

Armed with our 1.13GHz sample, Tom’s sample and Kyle’s sample, the Intel engineer and Kyle went at it.

You can visit Kyle’s description of what tests each CPU failed, but the basic conclusion was that our 1.13GHz sample ran through all of the tests perfectly, with one exception.  Tom provided Kyle with a hard drive formatted with Tom’s standard Linux test suite, which included a Kernel compilation test that Tom normally runs in order to test the stability of overclocked processors.  According to Tom, all normally clocked CPUs he has benchmarked have passed this test flawlessly, while most overclocked CPUs fail. 

It was on this test that all three chips failed, a test that has never failed on a normally clocked CPU but has failed on overclocked processors.  Why would a normally clocked CPU fail a test that only overclocked CPUs would fail?

There is a fine line between “overclocking” and what AMD/Intel do to increase the yields on their CPUs, but as long as they work, increasing the core voltage of a CPU in order to hit a higher clock speed is fine.  Overclockers do it on a daily basis and run their systems harder than a lot of your average home users with very few problems. 

It seems like this time around Intel may have pushed a little too far, since only two of the processors that made it out as review samples seemed to fail horribly, however it wouldn’t be surprising if most of the review samples failed the Linux kernel compilation test.  The fact that the three processors, ours, Kyle’s and Tom’s all failed the Linux kernel compilation test, a test that is failed normally by processors with poor margins, seems to indicate that maybe Intel was taking a bigger risk than they should’ve with releasing a 1.13GHz Pentium III.  Is clock speed really that important?

The kudos here goes to Dr. Tom Pabst of Tom’s Hardware that pieced together the puzzle so early in the game, and brought forth the proper test to prove it most recently.  While even we were skeptical initially because of our flawless experience with the CPU, it is clear that the 1.13GHz Pentium IIIs weren’t up to par with what we were used to from Intel and what we have come to be used to from AMD. 

For the first time in a while, Intel falls underneath this section.  Feeling the pressures of AMD, it seems like Intel was coerced into thinking that releasing a 1.13GHz Pentium III, although not completely ready, would give them something to compete against the upcoming 1.1GHz Athlon. 

It’s about time that someone realized that clock speed isn’t everything.  In AMD’s case, the Thunderbird core is perfectly capable of hitting 1.1GHz and thus AMD released a 1.1GHz Athlon.  If you remember, the first Thunderbird cores were demonstrated at 1.1GHz last February. 

Intel put quite a bit on the line by “releasing” a processor that wasn’t quite ready, they broke one of the oldest rules in the book, they sacrificed quality for quantity.  And now they’re forced to recall their 1.13GHz CPUs, not a smart decision. 

We have really seen the tables turn for Intel in the past year, but the company simply has too much momentum behind them to simply collapse right now.  They have definitely hit a number of bumps in the road, quite a few of them created by their own mistakes, although a number are due to the recent success of AMD processors. 

The Pentium 4 will prove to be an interesting chip for Intel, because out of the box it won’t immediately save Intel, but it’s a start.  We already know what the performance of the Pentium 4 will be, but unfortunately we can’t share that data with you just yet. 

It seems like Intel got a bit too wound up in the spirit of competition, let this be a lesson learned to both Intel and AMD.  According to Intel, the Pentium III 1.13GHz won’t be shipping in quantity for at least a couple more months.  While this date may change, it’s basically an attempt to play it safe on Intel’s part. 

As it has always been said before, once bit, twice shy, let’s hope Intel doesn't pull this a second time.