Overclocking is often described as an art. I for one believe this whole hearted, due to the subtleties involved in attaining just the right balance of speed and stability. Overclocking can involve a tedious trial and error methodology. The most commonly held misconception among budding Enthusiasts is that the highest frequency produces the best results. This is not necessarily true. The highest attainable frequency is usually the most unstable, producing numerous errors and poor results. When the consecution of micro-electronics such as the processor, North Bridge, and plethora of other devices are pushed beyond their normal operation problems incur exponentially.

As the internal temperatures of these devices rapidly build, their often pushed to the brink of failure. Heat is the modern PC's nemesis, and even when applying extreme cooling such as phase-change directly to CPU, other circuits are left to fend on their own.

This is one reason something as simple as leaving the side panels off my case, relieves some stress (heat). When Corsair designed a "housing" for the LED's and other atypical components in their Pro Series, they paid close attention to details. Thermal dissipation is increased on these modules, by virtue of additional surface area of the heat spreaders, and the use of thermal epoxy. As your about to see these details make a difference.


Entering the P4C800E-Deluxe BIOS I raised the external frequency to 250FSB at 1:1 aspect ratio. In the Advanced Chipset menu I left the timings on SPD, with PAT Enabled. Running Sandra in Buffered mode resulted in some excellent bandwidth. On the immediate left CPU-Z reveals Plug and Frag adjusting for frequency. All voltages remained at default. As I claimed earlier, I've had much better results leaving Prescott's Vcore in "Auto" rather then attempting to find an ideal voltage manually.

At this time I know of no other PC4000 memory on the market able to maintain CL2 at 500Mhz.




Next Sandra memory benchmark in UnBuffered mode. BIOS settings remain unchanged.




Next unto Aida32 memory READ.




Aida32 WRITE.




Science Mark 2.0.




Hexus PiFast Challange




3DMark2001SE




Aquamark 3 (1024x768)




[H]ardOCP UTK3 Low Benchmark




[H]ardOCP UTK3 CPU Benchmark




The results at 250FSB are excellent; unfortunately I was unable to run 3DMark03 at these speeds. I tried just about every fix, possible, but I believe there may be some problems with the Asus P4C800E-Deluxe, and the Sapphire X800Pro graphics card. During several Futuremark benchmarks I'd experienced complete screen disruption, resembling a Direct Draw type anomaly. In so far as Corsair's Plug and Frag technology, and CAS Latency, these are the tightest timings I've seen any memory maintain. SPD results in 2-4-4-8 performance. Given the fact this is a PC3200 part, it's SPD performance is simply outstanding. In fact among the current PC4000 to PC4500 parts which are designed to run at 250FSB (500Mhz) and above, CAS Latency usually takes the form of 2.5-4-4-8 or 3.0-4-4-8. The memory was stable in every environment at these speeds, with the exception of 3DMark03 benchmark.


Earlier I mentioned overclocking requires finding just the right balance among BIOS settings. I'd also mentioned attaining the highest speed, does not necessarily equal the best performance. To exemplify this point, I've included some benchmarks run at 240FSB (1:1) or DDR480. I was able to run PAT in Turbo mode, which, by Intel's figures, results in a 2-clock latency decrease in the MCH (Memory Controller Hub) at the North Bridge level. I manually adjusted the timings for CL2-3-3-6, and ran Sandra memory benchmark, in UnBuffered mode. This resulted in the largest bandwidth performance among all tests, as seen below.





In Buffered mode the memory didn't quite make the 6k bandwidth mark, however; it did come very close. I attempted to run CL2-2-2-6, and CL2-3-2-6 to no avail. Running SPD fared no better. The screenshot below indicates 240FSB (1:1) at 2-3-3-6.



Running the memory at 480Mhz at CL2-3-3-6 also resulted in my highest 3DMark03 score ever.




Onto the conclusion ->