Listen up, children! this is a story of graphics cards and GPUs, of sensors and tempratures, of ignorance and miracles.
After this pompous intro you will probably find the subject of this blog entry disappointingly dull. Yes, we all know temperatures are important – and not only for the hardcore overclockers. I am one of the paranoid nerds who has all system stats displayed on a separate screen. This is actually not entirely true since my secondary screen is often used for other functions as well – but nonetheless, I have conky running at all times. In addition to CPU, memory, hard disk, process stats and logs, it displays some 12 temperature readings and fan speeds. Overkill? Sure, but better safe than sorry, eh?
It appears that all my precautions have been in vain however and my graphics card is the culprit. It is an NVidia GeForce 8800GT bought two years ago. I noticed rather early that it used to run hot, but attributed it to the cheapo blower-style fan which it has by default. At some point my temperature readings appeared to be way off: While ambient temp sensor readings seemed to correspond well with other temps (CPU, mobo, etc.), GPU sensor was clearly way off. Idling at 85C and peaking at max load somewhere around 115C it appeared to be malfunctioning. Or at least that was my conclusion after being unable to pinpoint any specific cause. Since keeping the card under high loads for prolonged periods did not seem to elicit any artifacts, crashes or physical damage I was convinced that my conclusion was correct.
That was up until yesterday when I got the crazy idea of disassembling the fan. No, not just taking it off the card – but taking the top cover off of it. I was indeed in for a surprise. Out of reach of my usual cleaning methods (canned air, vacuum) there was an enormous clot of dust. It was located in the narrow gap between the fan and copper tubing which rendered it invisible from the outside and firmly in place out of reach of my cleaning tools. Otherwise the fan was clean – apparently my cleaning had payed off at least to some extent. Without any hesitation I removed the thick dust clot blocking air passage from the fan. Lo and behold, temperature readings dropped from 85C on idle to 60C (!), and, even more amazingly, under full load the card rarely reached over 75C compared to the previously mentioned 115C (!!!).These are drops of 20C and 40C, respectively.
Thus, it would appear that the temperature readings from the sensors were in fact correct. This also implies that I have been running my GPU at temperatures up to 120C continuously for up to 4-5 hours a day playing Fallout 3 and Assasin’s Creed over a period of two years. How this did not fry my GPU is indeed a mystery.
So, the morale of this miraculous tale is that you cannot clean your computer enough – dust will get into the most peculiar places no matter what you do. Unless you want to test the quality of your costly equipment in practice you better get to work with that canned air and screwdriver.
