What About Intel’s Pentium III at 1.13 GHz?
The story with Intel’s latest paper-released hype(r)-product is not at all over yet. You certainly remember the report, stating that my test sample was faulty and concluding that other samples might be faulty as well. In fact Kyle Bennett happened to offer me a historically new joint-venture with his publication [H]ard|OCP on this, since Kyle also received a seemingly flaky 1.13 GHz Pentium III sample from Intel last month. Kyle had sent his CPU back to Intel, who in return sent an engineer over to him to find out more about his problems. Before the engineer arrived at Kyle’s lab in Texas, he asked Anand Lal Shimpi from AnandTech for his obviously working sample and me for my faulty part as well. Therefore Kyle and Intel’s engineer Gary had the chance to see three different Pentium III 1.13 GHz samples in action. I also supplied a Linux installation to run the Linux kernel compilation benchmark that I recently included in our processor evaluation suite. This kernel compilation was never able to run on my faulty 1.13 GHz sample, while it ran fine with any other Intel CPU, including overclocked Pentium III 1 GHz and 933 MHz samples.
Joint Venture of [H]ard|OCP and Tom’s Hardware!
Let me first express my gratitude to Kyle Bennett, who organized the testing of all those three 1.13 GHz Pentium III processors and who also conducted all of the testing. Without Kyle’s efforts we might never have had the chance of shedding some light into this dubious story with Intel’s latest top-crutch processor.
It turned out that the kernel compilation failed on all three Pentium III 1.13 GHz samples in Kyle’s lab, right in front of the eyes of Intel’s engineer. My CPU happened to be the flakiest of the three, as it failed in other tests on different platforms as well.
I actually had to force a response out of Intel’s PR-department, which initially seemed to prefer burying this issue rather than talking to me. These are the messages I received in a telephone conference with Howard High, George Alfs and Gary the engineer:
- Intel is concerned about the issues Kyle and I were seeing with the Pentium III 1.13 GHz and their engineers are trying to duplicate our findings.
- So far Intel could not find any software that would fail on a Pentium III at 1.13 GHz, provided it runs ‘within spec‘.
- Other platforms besides Intel’s specially modified VC820 motherboard are most likely a bad choice for use with the 1.13 GHz Pentium III, unless they were specially approved.
- Intel was indeed able to duplicate the failures Kyle and I had found when, as they put it, the CPU ran ‘out of spec‘. When I asked to give a bit more detail to this expression Intel admitted that the cooling solution supplied with the test samples were not good enough to ensure that the Pentium III 1.13 GHz is running at less than the allowed 62 degrees centigrade. My comment stating that it was neither hot in Kyle’s or my lab and that we are both more than well capable of ensuring effective processor cooling was answered with the statement that the cooling solutions used in the 1.13 GHz OEM-boxes are rather massive indeed. After I told Kyle of Intel’s strange remarks he tested the temperature of the CPUs and measured 37 degrees centigrade. I guess there’s not much to add …
- Intel still refrained from really admitting that my sample is faulty. The fact that my sample was the only one of the three that failed on Kyle’s BX-platform was disregarded as a test that was conducted on an unsupported platform.
Please don’t hesitate to make sure you get the full scoop of this joint venture between Kyle Bennett and me and go to http://www.hardocp.com now! Don’t forget to come back though 🙂
Pentium III 1.13 GHz Bottom Line … For Now …
Basically I consider Intel’s response as well as actions as rather unsatisfactory. Here are my thoughts to the 1.13 GHz situation:
- I have yet to see a Pentium III 1.13 GHz that would be able to pass the Linux kernel compilation that I am using for processor evaluations. This compilation is using gcc 2.95.2, compiling an average kernel to a size of 572 kB. All other x86-processors that I have on all other platforms are able to pass this test. Only some overclocked processors happen to fail it as well. It is a very good test for overclocking stability. For the ones of you who are unfamiliar with Linux kernel compilations I’d like to give a very brief explanation. Linux is a GNU open source project and therefore all software, including the operating system itself, comes in form of source code. To make a program out of source code you need to compile the code with a compiler. The kernel of Linux is ‘monolithic’, which means that it is one piece of software (that can load additional modules as well, but that’s a different story). To customize your Linux installation to your own needs you modify the kernel’s make files and compile the kernel. This way you produce a kernel that meets your requirements without wasting system resources or performance on unnecessary stuff. This situation makes Linux able to even run on tiny systems, as e.g. hand-held computers. Imagine you’d try to run Windows2000 on one of those …
- Intel’s VC820 motherboard happened to be indeed the most stable platform for my flaky 1.13 GHz Pentium III sample. Still I was unable to finish a complete run of Sysmark2000 or the above mentioned kernel compilation on it. In further tests I found out that Intel’s special VC820 is ‘tuned’ for mediocre performance. A Pentium III 1 GHz on the VC820 runs significantly slower than on a comparable Asus P3C-L i820 motherboard. Therefore it seems that Intel modified the BIOS of the special VC820 for very low performance to ‘buffer’ the instabilities of the Pentium III at 1.13 GHz. As a result most shipping 1.13 GHz systems will hardly be much faster than a properly adjusted 1 GHz Pentium III system, because the majority, if not all, of those systems is equipped with this very Intel VC820 motherboard.
- The fact that all three 1.13 GHz Pentium III samples (Kyle’s, Anand’s and mine) acted in the Linux compilation test as if they were overclocked processors combined with Intel’s comment that these CPUs need massive cooling solutions leads to the simple conclusion that this processor is indeed nothing else than an overclocked CPU, only certified by Intel. Let’s be honest, when it comes to overclocking of Intel processors, who would be more experienced than Intel? Intel did all the well-known things: Raising the core voltage (1.7 V to 1.8 V), hefty cooling, lowering of the temperature spec, slow BIOS settings. Doesn’t that sound familiar? Maybe we should ask Intel to publish their very own Overclocker’s Guide. I bet it will be a huge success!
Is Tom Biased Against Intel?
Recently I got a bit irritated by some 10-20 emails in which people complain about my hefty criticism against Intel, accusing me of bias. To straighten this issue out once and for all, I’d like to explain myself here.
There is quite a bunch of sensible reasons why Intel is getting a whole lot of flak from me lately:
- I consider the release of high-end processor solutions that turn out to be utterly unavailable as unworthy of Intel. In the past Intel released a product and on the very release day you could buy those products almost anywhere in the world. Today it seems as if the releases of new Intel CPUs are pure PR-gags.
- I have got to criticize Intel if I receive a product that doesn’t work the way it should. My Pentium III 1.13 GHz sample is simply not running properly and I don’t seem to be the only one with this problem. I consider this as a rather serious incident. Once more I’ve got to say that I am not used to such things from a company of the format of Intel. In the past all Intel CPUs that I’ve used happened to be rock solid. Now times seemed to have changed for the worse, thus my criticism.
- I cannot possibly appreciate Intel’s dealings with Rambus and its policy of forcing a questionable product such as RDRAM down the throat of hundreds of thousand of loyal customers. I consider the dirt that has been dug up about Intel’s deals with Rambus so far as rather disgusting. Still I have my serious doubts that we have already heard it all yet. Intel will certainly receive praise from many others and me once it drops Rambus for good. My problems with Rambus have most certainly nothing to do with any stock market issues. I don’t own any stock of neither Rambus or any other chip or memory maker. People who make allegations that I am trying to make money with falling Rambus stocks are only sad, because they are projecting their own greed on me.
- Even Intel’s recent i815 chipset received criticism from me, although it would have been the one product that was supposed to be highly praised by Tom’s Hardware. I do like the i815 and I have made it my default Intel processor platform. Still I believe that this chipset could be better if Intel wouldn’t fear that it could interfere with its Rambus dealings.
- Finally I would like to remind you of the ‘Caminogate’ 1 and 2. Last year the i820 chipset was delayed and finally one RIMM-slot had to be removed from the design of the i820 motherboards. This situation left the brand new Coppermine processors without an Intel platform for a couple of months. Later i840 turned out to be unable to run with ECC SDRAM and then there was the sad MTH-story.
I only listed five points, although I could go on about it even further. Bottom line is that in the recent past Intel has provided a long list of failures and questionable dealings. Each failure and each piece of the Rambus-affair pushed me to be very critical with Intel. I cannot help it if Intel is having the worst record of its time. Don’t blame me, blame Intel!
Finally I’d like to mention that I am aware of the fact that Intel is also providing good stuff. The ‘Coppermine’ Pentium III processors up to 933 MHz are excellent performers at an excellent price. There’s of course more, but failures, flaws and questionable deals heavily overshadow all those good things right now.
Please follow-up by reading Intel Admits Problems With Pentium III 1.13 GHz – Production and Shipments Halted.