No Help From Intel
Intel kept completely quiet after the release of my second article. I finally received the special VC820 motherboard that was tailored for the Pentium III 1.13 GHz and the testing on this platform confirmed my accusations. My Pentium III sample was still flaky on this platform as well. Obviously I decided against publishing another article about this dreadful CPU, since I didn't want to upset even more of my unbelieving readers. In the meantime Intel had asked me to send my processor sample back, so that they could scrutinize it. Unfortunately my faith in Intel's words wasn't big enough to dare giving away the one and only proof I had. I decided to hang on to it and asked Intel to send me a working sample instead.
The Linux Story
When I was testing the Pentium III 1.13 GHz I was also using a GNU/Linux installation to run kernel compilation benchmarks. I had introduced this benchmark in our processor-benchmarking suite only recently and was therefore not too experienced with this operating system. What I knew for sure however was that my Pentium III 1.13 GHz hadn't been able to finish the compilation even once. I wanted to make sure that my Linux installation was proper, so I decided to install Linux on a new hard drive, install the latest NVIDIA drivers and then go back to testing. It took me several days to get my Linux installation running to my satisfaction and the result of those efforts was the first Linux review article about 3D-cards under Linux .
The Proposal Of A Joint Venture
In the meantime Anand from AnandTech , who had been lucky to receive an actual working 1.13 GHz P3 sample, had offered me his CPU after he had heard that Intel wouldn't be able to supply me with another sample soon enough. I declined because I considered this issue to be Intel's business, which is why I expected Intel to sort out the situation and not Anand. In return Anand offered his CPU to Kyle Bennett. Kyle accepted and asked me to send him mine as well, so that he could compare the two and possibly back up my findings. I agreed, but I wanted to do a bit more testing myself first. I also hoped that Intel would finally send me a new sample, but this hope turned out to be in vain.
Linux Brings It Out In The Open
In my further testing sessions I reached the following concluions:
- My 1.13 GHz Pentium III sample was unable to pass any benchmark runs on any platform besides Intel's special VC820 motherboard.
- On Intel's VC820 platform Sysmark 2000 crashed consistently. I was unable to finish even one run of Sysmark with this CPU and I certainly tried about 20 times. As soon as I plugged a Pentium III 1 GHz into the system the benchmark would run all the way through.
- The most consistent error I got however was with my timed Linux kernel compilation. Even on the VC820 the Pentium III 1.13 GHz was utterly unable to finish the compilation even once. All other CPUs I used finished the compilation without the slightest flaw.
- Interestingly, stress tests as Prime95 or CPUburn under Windows98 would not get my 1.13 GHz processor to fail on the VC820.
Armed with this knowledge I finally felt confident to send my sample to Kyle Bennett. Kyle had told me that an Intel engineer was supposed to visit his lab to oversee his testing. Later it turned out that this engineer was not specifically sent to Kyle, but happened to be in the area, so that he could easily visit Kyle's lab. Since Kyle was rather unfamiliar with Linux I copied one of my Linux test hard drives and put it into the shipment as well. I wanted to know if the other two 1.13 GHz processors would be able to pass this test.
Well, if you have read Kyle's article you will know what happened. All three 1.13 GHz Pentium III samples failed the Linux compilation. My CPU did particularly bad, while the other two showed reasonable stability.
Intel still hadn't felt any need to get in touch with me. Instead of that, Intel acted even worse. Last Monday, on August 21, Intel had released details about their upcoming Pentium 4 processor. In a crazy act of retaliation against my article about the faulty Pentium III 1.13 GHz processor, an Intel official had decided to suddenly take Tom's Hardware out of the loop. Almost any other hardware website reported about the Pentium 4 architecture except of us, because we had never received the information.
I complained to anybody I know at Intel, and believe me, I know a lot of guys there. Finally I had a telephone conference last Friday with Howard High, George Alfs and Gary the engineer who had been at Kyle's place. I was told that Intel was still having full confidence in their Pentium III 1.13 GHz processor, but that further testing was still being conducted. You find more details about this conference call in the recent article Latest Update On Intel's 1.13 GHz Pentium III .
I had worked through the whole Sunday night to finish my article about AMD's Athlon 1100 and in the later morning hours I had to get a few things done in town. When I came back a few hours later I had a long message from Howard High on my answering machine. He stated that last weekend Intel had finally found out that the failures Kyle and I had seen are not due to conditions that are beyond spec of the 1.13 GHz processor, but consistent. He particularly pointed out the failure I had seen in Sysmark 2000 and the Linux kernel compilation. Therefore Intel had finally decided to advise its OEMs to put the shipment of 1.13 GHz Pentium III systems on hold. Intel also halted the production and shipment of their 1.13 GHz processors.
I called him back to get into another telephone conference with him, George Alfs and Christian Anderka. These are the basic messages:
- Further testing conducted on the weekend has confirmed the instability of the Pentium III 1.13 GHz as I had stated all along.
- Intel has decided in emergency meetings last weekend that the further shipment of this processor needs to be halted. OEMs were informed.
- The Pentium III 1.13 GHz will require a new stepping, which will obviously take some time. Until then this processor will be unavailable.
- Customers who own a Pentium III 1.13 GHz will be contacted to either get some kind of replacement or a refund. Intel states that the number of end users with a 1.13 GHz system is rather small. Our own estimates are between 10,000 and 20,000 shipped systems. Intel did not quote a number.
- Intel reacted 'that swiftly' because it wants to save its brand name.
There it was. Finally Intel had done what I had forecasted almost a month ago.
There are quite a few unanswered questions lingering around:
- Why was the testing only commenced 5 days ago and not right after my initial article on July 31? Couldn't that have saved Intel a major part of the face loss it's encountering now?
- Why was I never equipped with another Pentium III 1.13 GHz sample to verify my findings?
- Why did Intel neither try to send an engineer over to my lab nor ask me to fly over to San Jose, which is right around the corner from Santa Clara? We've got another lab there. That could have saved valuable time and kept a few thousand end users from buying a system they now have to return.
- Why was Kyle only visited by a guy who 'happened to be in the area '? Does that mean that the whole situation would have been delayed even further if 'Gary' had not been around Dallas, TX?
- Why did Intel not contact me after the testing session in Kyle Bennett's lab, although I supplied not only my sample but also the Linux software that turned out to be the final piece of the puzzle?
- Why was the Pentium 4 information deliberately kept from us a week ago?
- Why was Intel's in-house validation team unable to find the flaws of the 1.13 GHz Pentium III in the first place?
- What will Intel do to ensure that its next processor release will actually work as expected?
- Is it indeed true that Intel suddenly found an issue with the 1.13 GHz Pentium III last weekend, or couldn't it be that Intel was aware of design flaws for much longer? After all there was a big PR-meeting last weekend. It seems as if the decision for the retraction was made then, but that the testing had nothing to do with it. What is Intel trying to cover up here?
I suppose we will never get the answers to these questions. I am particularly puzzled with the last question.