Further Upgrade Adventures
All the kit for the upgrade I discussed in my last post arrived in good order and time. I used scan.co.uk and overclockers.co.uk and I’ve never had any issues with either of them, although I hesitate to suggest that any online retailer is going to be perfect for everyone. The components were all built into my existing Antec Sonata case and the build went without a hitch, including a run of MemTest86+ and the Vista install (apart from my Audigy 2 ZS card, which still doesn’t seem to have authorised Vista drivers, so I stuck with the onboard Realtek audio for the initial tests).
Once I was happy that it was all running at stock, I set about overclocking and this is where the problems began. Now, an E6300 runs at 1.86GHz stock, but with the Gigabyte DS3P and a Tunic Tower all the reports suggested that a 3+GHz was well within reach. Colour me unhappy then when the damned thing wouldn’t clock up at all. Not even a 5MHz FSB overclock. Considering that the memory had passed MemTest86 and the system wouldn’t boot and run stable with a CPU overclock and with the memory underclocked, I could only look at the CPU. Temps were 60+ degrees C on the cores after a minute or so of an Intel TAT load test, which I wasn’t at all happy with seeing as others had reported temps in the mid 40s even on stock cooling. It seemed unlikely that the temps would stop any overclock at all, but it was the only suspect at the time. A less honest overclocker might try to RMA a chip running that hot, but I accept the luck of the draw in these things and so I had to look to other measures.
The first thing I tried should probably have been the last, but I had a day off work to play with and fancied a bit of a project and so I decided to lap the CPU and heatsink. For anyone who doesn’t know what this means, it means that I took the processor and the cooler and rubbed them on some sandpaper until they were flat and shiny. It sounds like the behaviour of a mental defective, but it’s really not that crazy. The objective of the operation is to make sure that the surfaces of the CPU and heatsink make as much contact as possible in order to transfer as much heat energy as possible. The Core 2 Duo chips are covered with an Integrated Heat Spreader, so the actual chip isn’t being sanded, just a bit of nickel-coated copper, and the process is actually pretty risk-free.
For the lapping, I used a set of 240, 600, 800 and 1200 grit wet and dry sandpaper from Halfords, which cost a couple of quid and would last two or three lappings, judging by the wear from one. There are a number of guides to lapping online but, basically, I taped the paper to a flat surface (you need tape with powerful adhesive as there’s a fair amount of friction) and worked through the grades, starting the heatsink on 240 and the CPU on the 600 grit. I worked in one axis at a time, no circular movement, and concentrated on trying to keep the surface flat and pressure even, rotating the heatsink or CPU 180 degrees every minute or so in order to make sure I wasn’t biasing one edge. During the initial flattening on the lower grit paper, I turned the CPU or heatsink by 90 degrees when I thought I had it flat, which then allowed me to see how evenly the sanding marks were removed and replaced by the ones in the new axis. This 90 degree flip was repeated when I changed grades and also on the last pass at 1200 grit. Both the heatsink and CPU were clearly uneven at the start and the process took about an hour overall, with most of the time spent on the Tunic, as there was obviously more metal to remove from it.
So, with that done, I refitted the CPU and heatsink and was rewarded with a temperature drop of about 5 degrees. Well worth the effort, except that it still wouldn’t overclock. So onto the thing I should have addressed first — the case airflow. The Sonata is a nice looking case and it’s good value too, with a decent built-in PSU, but the airflow isn’t great. Plus, seeing as the case is sold on its quietness, it really isn’t all that quiet, which means that I, like many others, had taken to running it without a system fan, trusting to the PSU to pull air out. This works, but is hardly ideal, so I grabbed an old Vantec Stealth fan I had lying around and fitted it to the rear exhaust. Temps dropped to 50-52 degrees under load. Not too bad, but still no overclock.
At this point, I had spent about 20 hours trying to get any kind of overclock and I pretty much gave up and set about just installing software and playing some games. The system was a huge step up from my old one even without overclocking anyway and it was nice to experience games like Oblivion at full 1680×1050 resolution and with the quality settings cranked up and still getting decent frame-rates. The overclocking thing still niggled at me though, and worse, I started to see a few BSODs while using the machine. After about a week of this, I remembered some issues I had with DDR in another system where the dual channel was the problem. So, as a last hope, I did some experimenting with moving the RAM sticks around. The first thing I tried was simply yanking the second stick and running with 1GB single channel with a relatively small 300MHz FSB overclock. Bingo! The system happily booted and ran for 20 minutes under Orthos. I then swapped the sticks and ran with just the second stick and the system wouldn’t even boot. I’d found my overclocking block but, unfortunately, if the stick ran at stock then I was stuck with it. However, even after I switched everything back to stock, the machine wasn’t stable. It booted Vista about one in three attempts, and wouldn’t run Orthos Blend for more than a few minutes even if it booted. Eventually, a BSOD completely corrupted my Vista install. After Vista died, I discovered that the Ubuntu 6.06 Live CD would kernel panic or seg-fault every time with the faulty stick and work without a hitch with the other stick. However, and here’s the lesson/warning, MemTest86 happily passed the faulty stick. I ran MemTest for three straight runs through all the tests and five times through test five, which always used to be a good benchmark for stability and the memory passed without a hitch. I guess MemTest86 is not infallible.
So, I’m now waiting on an RMA for the memory with overclockers. I’m a little worried that they won’t accept that it is faulty, but they’ve always handled RMAs well before now. This leaves me without a desktop again. In the meantime, I’ve decided to upgrade the case to an Akasa Eclipse 62 and fitting some Scythe S-FLEX fans, just because I feel that the Sonata’s airflow will become more of a liability with summer coming. I’ll move my server kit into the Sonata, as it isn’t overclocked at all, and the disk mounting system in the Sonata may just help remove some annoying resonance noise from that machine, which would be good as it runs 24/7.
