Monday, June 6, 2005: Our stress test has now been going for more than 60 hours. The systems from AMD and Intel are running steadily under full load, and a lot has happened in the Munich THG lab during this time.

After just a few hours the Intel system, based on the Gigabyte GA-8N-SLI Royal with an Nvidia nForce4 SLI chip set, failed. That was on June 4, at around 5 PM Pacific time. After a fault analysis had been conducted, we restarted the applications seven hours later - at 12:20 AM, to be precise. Between 5 PM and 12:20 AM no applications were able to run.

We should also add that at 8.45 AM another crash occurred on the Intel system. A comprehensive fault analysis of the hard drive conducted afterwards did not shed any light on the possible cause.

At this point, here is a brief note from the lab engineers. The actively cooled Northbridge of the GA-8N-SLI Royal becomes unusually hot, so in our opinion the chip set is probably the cause of the crashes. A BIOS update may provide the potential for improvement; the manufacturer promptly sent us one when we requested it. In conjunction with this update, the Intel system was restarted once again.

gets extremely hot: actively cooled Northbridge on the Intel system.

What is interesting is the sequence of events in the case of the AMD system: There was not a single crash in the same period. There has been a correction to the restart counter: many readers interpreted this as a system restart. In practice, the entry "1" stood for the start of the stress test. In order to eliminate misinterpretation, we reduced both counters by this value. Something peculiar happened with the 3D game Farcry though: after running the demo 1000 times, the system abandoned the benchmark on its own, as an analysis of the log file clearly showed.

Excerpt from the benchmark-specific log file Cooler01.log

Several readers presumed that the AMD system only ran in single-channel mode. That is not the case, however, as with newer Gigabyte boards the modules are inserted right next to each other. As a result, both systems run in dual channel mode.

We also need to explain something about the benchmark applications. In the graphical display of the live charts, there were errors in the programming at first; consequently, the bars were not visible during MP3s encoding. We have since resolved this problem (bear in mind that real-time display involves a considerable amount of programming work.)

Another issue was the extremely slow Divx compression on the AMD system. Both AMD and Intel work with the same software and the applications were started in the same order. The Windows system itself determines what priority is assigned to each application. We do not use any software to influence the priorities of the individual applications during processing.

Wattmeter: The display of the electrical current gave up the ghost after a very short time indeed.

Finally, there was also a failure with our measurement technology: the wattmeter gave up the ghost with the current reading - we plan to replace it in the near future.