Sitting and surfing the net, researching smtp proxies (I'm working on a heuristic anti-spam solution), and I get this alert on my PC indicating my processor temperature is exceeding the recommendation of its manufacturer. In this machine it shouldn't ever happen since I have this huge Zalman pure copper fan on the processor that varies its speed based on the temperature of the processor. The only way that should be possible is if I have some runaway process consuming 100% of the available CPU cycles and just punishing it for long periods of time. Even then, increased fan speed by Zalman should handle it. In fact, I could faintly hear the zalman fan now that I was looking for it -- normally I can't hear it at all.

With a regular processor, a runaway thread consuming all resources it could get would be obvious -- nothing else would run at all. With my Intel P4 Hyperthreading processor I didn't even notice it. Hyperthreading uses the unused available parts of each cycle to toss in instructions for the next cycle in queue. Most instructions don't take the full width of the processor's core so there is plenty of room left over. Windows XP actually sees this machine as a dual processor, with one processor being that "sideband". Before shutting down, I decided to have a look. I pressed ctrl-alt-delete and brought up the task manager, the sorted on processor use. I showed two processes each at 50% and holding. One was the system idle process -- windows saw one processor as sitting empty -- and the other was trillian.

Result? Instead of a hung machine, I simply had to kill that one process. As soon as I did, the fan quieted and by the time I could launch Intel's "Active Monitor" software (the tool comes with Intel motherboards) to check the fan speed and processor zone temperature it was back to normal. No re-boot required.

When I checked the logs, I found that the Zalman CPU Cooler (what a fantastic piece of equipment) had in fact been able to handle it. The core temperature had shot up but that had lasted less than 1 second before the increased fan speed was able to compensate and the temperature returned to normal even though the process was still running wild. Had I not been here (and had Zalman not handled it) the Intel software would have shut down the box of course before the thing melted.

Even though some bug in trillian made it go crazy and take over all available cycles, that bug was single threaded. It was pipelining instructions at the processor as fast as the processor would take them but it still left over the "hyperthread" -- the unused portion of each cycle -- for the chip to allow Windows to do its other stuff. To Windows, that looked like a 50% busy system even though the processor was totally hammered. The left over portion of each cycle, at 2.8ghz, is still way way way more power than anyone needs to do work or surf the web.

The implication for daily work is that on this machine -- a 2.80E gigahertz P4 -- even though a process was pushing for every single clock cycle at 2.8 gigahertz, the "hyperthread" given to me was more than enough to still have nearly ANOTHER 2.8 gigahertz devoted to doing my work. WOW.