4.2GHz with P4-3.73, Corsair HydroCool, XMS2-5400UL in an hour flat

OFTEN, there are comprehensive overclocking reviews showing incredible overclocking results after several days of trial-and-error runs, then a few rounds of fine tuning, to get to where they reach. What about stable overclocked platform, one that runs non-stop for few days running tasks, yet can be set up from scratch in, say, just an hour from scratch?

This of course excludes the usual slow Windows XP setup, where no overclocking can help huh.

So, I gathered few spare things in my lab - a P4 EE 'Prescott' 3.73 GHz with 1066 MHz FSB, Corsair XMS 5400UL low-latency DDR2-667 memory that can run much faster in many overclocking setups, and borrowed a MSI P4N Diamond board, the first batch of Nforce SLI Intel boards from this vendor, together with two MSI 6800GT SLI 256 MB boards as well. On top of it, Corsair Hydrocool 200EX water cooler was used instead of the usual heat sink & fan combo to cool down the CPU (and, by extension, the surrounding, since less heat would circulate around the system).

It took about half an hour to put the stuff together physically to the point seen in the pic, with all in place and the cooler cables connected with water inside filled up. The setup ran fine at the default 3.73 GHz CPU, FSB and RAM speeds. Unfortunately, the MSI board can't unlock the FSB ratio to enable lower CPU:FSB dividers, so as we up the FSB, we had to up the CPU clock proportionally.

The aggressive start at 1333 MHz FSB / 666 MHz memory didnt work, even with relaxed memory settings - after a few tries up & down (including 1200 MHz CPU / 800 MHz memory which also surprisingly failed even with high voltages and relaxed 4-4-4-12 latencies), I settled with 1200 MHz FSB / 600 MHz memory at 3-2-2-8 latency for the Corsair 5400UL, i.e. the CPU running at 4,200 MHz exactly. Now, an extra 20 mins was spent on these five runs - was it about to run stable in Windows?

I actually installed the Windows XP Pro at this speed setting, without a hitch. In the finished system with two SLI cards and 1 GB RAM, all benchmark ran fine, including running multiple rounds of PCMark and Sandra overnight - not a single hangup, and interestingly, only a small temperature increase on the CPU, from 29 C at 3.73 GHz to 33 C at 4.2 GHz as shown by the Corsair Hydrocool (the CPU monitor showed 32 and 38 C respectively, though). The GeForce 6800GT cards ran at the default 350 MHz GPU/1GHz RAM clocks in SLI mode.

So, here are the results in Sandra, plus 3DMark05 (default, as well as SXGA normal and antialiased) and PCMark04 results. The first is a big file which will open in a new window.

The other two files are worksheets which you can find here and here.

As you can see, the Sandra memory bandwidth gains are pretty proportional to the 3.73GHz/FSB 1066 setup (or the 3.46 GHz old EE with the same FSB and same memory) as the CPU clock, FSB and memory clock were upped while keeping the same latency settings. Overall, an interesting experiment, and a safe, stable 4.2GHz PC - at least for now! µ