Warning: if you’re not a computer g33k don’t bother reading this post, you just won’t understand!
In December, I started playing a little more of the game Colin McRae Rally: Dirt. However, I decided that my beast of a machine (refer here, here, & here) wasn’t performing quite fast enough for maximum off-road rally driving pleasure. Thus, my recent overclocking obsession came to be!
Basically, I overclocked my Q6600 B3 revision Intel Quad Core CPU from the stock 2.4GHz speed to 2.6GHz. However, after countless hours upon hours of running CPU stress testing (specifically, Prime95 v25.6 build #2), and comparing to temperatures measured by other people with almost exactly the same hardware configurations, I came to the conclusion that I am unlucky to be one that has received a CPU chip that just runs at a hotter temperature than the average manufactured chip. Upon exhaustive research through overclocking forums on the net, I discovered that lapping (making something flat via sanding) the Q6600 Core 2 Quad CPU is generally a very beneficial thing to do for increasing the thermal transfer of the CPU cover to the heatsink. I suspect the manufacturing process that Intel use to fix the CPU cover to the cores causes the cover to bend slightly. Consequently, the outside edges of the CPU cover (and a section in the middle) are slightly raised from the rest of the cover, leading to big air gaps that require filling with thermal paste in order to transfer heat half reasonably. Since it is the outside edge that is most raised, I can see why my CPU heatsink created a big air-gap when it was installed. This definitely contributes to a hotter CPU, which means less overclocking is possible before you risk overheating it.
I used the razor blade with torch behind it technique to determine how flat both surfaces really were, but I found that the razor blade wasn’t rigid enough to provide consistency for a good idea of how much curvature existed. The heatsink seemed relatively flat already, but the CPU definitely had big gaps of light coming through. Regardless, I decided to lap the heatsink so I could practise my lapping technique before doing the CPU.
I conducted the lapping process using a large piece of tempered
glass (compliments of Scoobth) which I used to tape the wet-and-dry sandpaper on for a flat surface. I used a spray bottle of water to keep the sandpaper moist at all times. The sandpaper grit sizes I used were 200, 400, 600, 800, 1000, and 1200.
The most time consuming stage was using the very coarse 200 grit size, just to get the surfaces flat before I could progressively smooth them down. I only had 1 piece of each grit size for the CPU, and 1 piece of each grit size for the heatsink, and all of the sandpaper became too worn for them to be worthwhile using anymore. If I were to do it again I would have a lot more sandpaper handy because once it wears down, the efficiency of the sanding action decreases dramatically and you really just waste time.
I couldn’t work out how to remove the thermal plate surface from the heat-pipes of the Scythe Ninja heatsink, which made for a very high centre of gravity as the complete unit slid over the sandpaper. Consequently I found it difficult to stop the heatsink from shuddering.
The CPU was much easier to sand because as a flat plate it has a very low centre of gravity. I held it with the black piece of protective plastic that it came with, so that way I didn’t touch any of the sensitive side of the unit. You can see after starting to sand, the edges of the unit were scratching away (nice shiny silver instead of a matte finish). It took a large amount of sanding to actually notice this. After a while, you could notice a circle in the middle also start to sand away
I kept using the 200 grit sized sandpaper up until about this stage where I switched to the 400 grit sized because the 200 grit had completely worn out! Probably a good idea anyway, because there wasn’t much sanding now at this stage to get it completely flat.
By the time I got to about here, the 400 grit paper was also fairly worn out, but I kept going until all the plate was uniformly shiny and all of the matte surface had gone. I read on a few forums that the silver surface is a nickel plating. The only reason I can think of why nickel plating would be used on a CPU cover would be for corrosion resistance, because I would imagine corrosion of (what appears to be) copper underneath the nickel plating would have a very low thermal conductivity.
Somewhere at about this stage, I noticed a red liquid appearing on the sandpaper. It was blood from my finger, which I didn’t feel was being sanded to the flesh as my hand was numb from holding the CPU! Lesson to be learned: keep lots of fresh sandpaper handy so you don’t have to waste time sanding with worn out sandpaper!
Eventually, after going through all the grades of sandpaper, the end result!
I then cleaned the surfaces using a few drops of the Arctic Silver ArctiClean product (compliments of Scoobth) and applied a line of Arctic Silver ArcticSilver5.
I applied the line of AS5 as the instructions suggested, although I think the instructions would be written with the non-flat surface of the Core 2 Quad Q6600 in mind, and I should probably have used a much thinner line.
Next time I ever do it I would definitely still use it in a line for the purpose of ensuring all cores have good contact in the immediate vicinity, but just used a lot less. It would perhaps have been easier to apply if I heated the AS5 liquid up somehow. Not in the microwave as it contains metal particles, but perhaps out in the hot Australian sun or in the oven on a maximum of 30 degrees celcius. The AS5 compound probably doesn’t reduce in viscosity when it heats up, but it would be worth a try.
The core temperatures of the CPU instantly ran about 10 degrees celcius lower, and after a few thermal cycles and a few days about 12 degrees celcius lower. I tested this temperature drop using ambient temperature that was plus or minus 2 degrees celcius, and taking measurements using SpeedFan & CoreTemp after about 1 hour of Prime95 v25.6 build #2 using the small FFT torture test.
I was then able to overclock my CPU to 2.7GHz, but the cores were starting to run in the red zone (above the maximum rated temperature of 71 degrees Celcius) so I also used another overclocking hardware modification called the pencil Vdroop mod. This mod involves using a graphite pencil to shade over a (I think the component was a) resistor to effectively lower the resistance by adding another resistor (the conductive graphite) in parallel. Now although I’m a qualified Mechatronic engineer, I have absolutely no idea what any of the circuitry on the mainboard does, so I was taking a giant leap of faith from fellow overclockers on the internet! The idea is you keep lowering the resistance until the voltage across the CPU core(s) is exactly the same voltage regardless of whether or not the CPU is experiencing a full load or not. That is, there is not voltage drop (aka droop) when the CPU is under computational load. The bad side of this is temperatures immediately sky-rocket when you start to next apply computational load because the voltage doesn’t decrease at the same time. So, the idea is to lower the overall voltage to compensate. The end effect is increased voltage stability of the CPU due to it not fluctuating, which means the CPU is less likely to produce errors, and there is less heat generated at idle capacity. It definitely helped to decrease the temperature of my cores under load, which is particularly important for this computer as it’s used to do 24/7/ number crunching for Team_Bammann.
So I now have a 2.4GHz Core 2 Quad running at 2.7GHz, at a lower temperature than it was before! I’ve also upgraded all of the case fans to medium speed Scythe S-Flex fans, and the fan for the CPU heatsink to the high speed Scythe S-Flex fan. So there is also more airflow but with less noise, and this also further cools the CPU cores by another couple degrees.
This definitely helped the performance of the Colin McRae: Dirt game, but what REALLY helped the performance of the game was adding an extra 2Gb of RAM to a total of 4Gb. Unfortunately I can only use 3.0Gb due to the Vista 32bit limitation with my mainboard and graphics card, so an effective total of 3Gb. This SERIOUSLY increased the gaming performance, completely eliminating the stuttering that was occuring at times when lots of cars were on the track. What I realise now is that Windows Vista Ultimate uses so much RAM that 2Gb is just not enough for computer games. I expect that if you want to get serious about a beast of a gaming machine with a Vista installation, you need to go for 64bit and get a minimum of 4Gb.
I then also added (as I mentioned in a previous post) the Logitech Momo steering wheel so that I can really enjoy the new gaming beast I have. 
