What does it take to hit a front-side bus rate of 250MHz? I was assigned the task of putting together a new development rig for myself and my team, so I thought I would use this opportunity to experiment with different components and find out what goes into a strong, stable, and fast machine. This was my first real venture into the realm of overclocking, and I’ve learned a lot of interesting things.
First, a word on equipment. I started out with an Antec SX835II enclosure, which includes an Antec TruePower 350W power supply. I decided to base the rig on the new ASUS P4C800-E Deluxe because of a variety of factors, including dual channel memory and chipset-level RAID. I picked up a P4 3.0GHz with stock heatsink/fan. That’s the basic story. For those mathematicians out there, yes I am trying to overclock to 3.75GHz. Hooray for lofty goals!
For the memory, I used 1GB Corsair TwinX 4000 Pro (with the cool little tachometer lights – very awesome!). I tried several different power supplies: Antec TruePower 350, Antec TruePower 480, and Antec TrueControl 550. Please don’t complain about my choice in parts. I picked everything up at Fry’s so I could return it with no restocking fee.
As for the benchmark tools, I used Prime95 to load-test the CPU, with Motherboard Monitor 5 to monitor temperature and voltage. I used AIDA32 and SiSoft Sandra to get benchmark scores on the memory.
OK, so now it comes down to experimentation. Using 1GB Corsair TwinX 4000 Pro, and trying Antec 350W, 480W, and 550W power supplies, I was unable to drive the P4C800-E Deluxe at 250FSB. Usually, the system would not POST or Windows wouldn’t boot all the way. In a few tests I was able to bring up the O/S, but about half the time, the system was slow, semi-unresponsive, and hot, as if it were under load (what I call “phantom-load”).
Using the 350W PSU was a joke: the system would never boot. I went to the 480W and I was successful at booting the system, but it exhibited the symptoms of phantom-load which I described above. I took an MBM5 log of a Prime95 torture test and noticed that the core voltage was running low (0.065V under BIOS) and fluctuating (±0.01V). Low, unstable voltage? More power!!! On to the 550W.
After installing the 550W PSU and dialing the core voltage to 1.6250 or 1.6500, I was able to boot the system at 250FSB and load Windows into a seemingly clean, cool state. The system seemed stable, but the Prime95 torture test still failed, and failed instantly. Although the upgraded PSU did make the idle voltage very stable and accurate (within 0.005V of BIOS and not varying more than 0.005V), there was a sharp voltage drop at load (0.07V) which was apparently enough to choke Prime95.
I spent more time adjusting voltages, but I met with only more of the same aberrant behavior. I considered upgrading the cooling on the CPU, despite my suspicions that temperature wasn’t the problem. CPU temp never got above 55°C, and the Prime95 failures always seemed to be correlated with voltage drops/fluctuations. Therefore, I reasoned, the source of the problem must be that the power supply to the core is not stable enough. However, just to rule out the possibility that temperature has an effect on voltage tolerance, I installed a Zalman 7000 AlCu and cooked it in with some Arctic Silver 3 thermal compound for a few days. Now I was able to load Windows with the 480W PSU, but Prime95 was still failing on both it and the 550W.
Since I’ve already worked my way up to a 550W Antec PSU and still couldn’t drive 250, I figured I should admit defeat and set my sights on a stable 233FSB.
<b>Runner Up: 233</b>
So now I would have to decide between the expensive Corsair PC4000 Pro (with the cool lights!), and the less expensive 1GB Corsair CMX512-PC3700. If I could get a stable system out of the 3700, I would go with that and save about $50.
What I discovered is that I could run a stable system at 233FSB, 1.6000 core voltage, auto RAM voltage, RAM timings of 3 4 3 7, all on the Antec 350W with the Intel stock HSF on the CPU! AIDA32 gave me 5648 MB/s read and 1985 MB/s write (taken as averages of three benchmarks). If you do the math, you know that the CPU is riding at 3.5GHz under this set up.
I’m pretty happy with this, so I’m going to stop here for now and do some extra tests to ensure rock-hard stability. I went back to the Zalman HSF just for better cooling, and the Antec 480W, just to be extra safe.
Among the lessons I learned during this trial, one is that higher wattage is not just about power capacity. It is also about power fidelity or “clean” signal. Going from 350 to 480 to 550 showed noticeable improvements in voltage logs, 550 being the best of course, bringing core fluctuation below the resolution of the monitoring tool (0.01V). Unfortunately, it wasn’t enough of an improvement to meet my goals.
Also, I discovered that cooking the thermal compound makes it more effective. Throttling the system overnight heats the compound over a prolonged interval, allowing it to mold into the microscopic crevices in the heatsink and in the CPU heat spreader. It also purges out any little bubbles of air that are trapped, and lets it sort of “cure”. This actually has a noticeable impact on system stability (when you’re pushing the system to 10/10ths, anyway).
1.) I wasn't limited to Antec power supplies, I actually used good ones. Both my hand modified cheapo 400W unit, and my Fortron FSP 350-60PFN provided PERFECT line voltages, yet I still experienced voltage fluctuation from the motherboard's VRM which caused stability issues
2.) I didn't have a heat problem.
<font color=blue>Only a place as big as the internet could be home to a hero as big as Crashman!</font color=blue>
<font color=red>Only a place as big as the internet could be home to an ego as large as Crashman's!</font color=red>