Finally we get to the big system, for which only the smallest compromises were made in the name of value.

The sky was almost the limit for our high-end system's $4,000 hardware budget, but the need to justify each component prevented us from spending the entire amount. A rather large portion of the budget was allotted to CPU cooling to assure top overclocking results, and now we will see how that investment paid off.

In spite of rumors about the 780i chipset not reaching high bus speeds when equipped with a Yorkfield-based processor, our Core 2 Quad was perfectly stable at speeds over 450 MHz FSB. Yet bus speed is not the total solution to performance, and we had to take memory speed into account when selecting an FSB clock and CPU multiplier

We chose a core voltage limit of 1.45V to protect the CPU from damage, and found a highest stable speed for our Core 2 Duo QX9650 processor of 4.35 GHz. This incredible 45% increase from the processor's stock 3.00 GHz was tested inside a quiet closed system, rather than a more overclock-friendly open air environment.

Typical NVidia BIOS has few manually-selectable fixed memory ratios and the Striker II Formula was no exception. Settings included FSB clock to DDR data rate ratios of 1:1, 5:4 3:2, and "Synchronous", which translate into real memory clock multipliers of 2.00, 1.60, 1.33, and 1.00. Another "Linked" option for FSB to DRAM clocks is the "Automatic" setting, which allows BIOS to choose from additional ratios that are not manually selectable. While the "Unlinked" option allows manual adjustment of memory speeds in smaller increments, many of the strange ratios it enables can hurt memory controller performance.

Our Ballistix DDR2-800 samples pleased us by reaching 1146 MHz at 2.30 volts and CAS 5-5-5-15, using the 3:2 BIOS setting with an FSB of 430 MHz. Moving up to 435 MHz FSB took the memory to an unstable 1160 MHz, and adding voltage simply didn't help. As expected, the higher-capacity 2x 2 GB kit doesn't overclock as well as previously-tested 2x 1 GB kits. We compromised by using the "Automatic" option, BIOS chose a 1.20x multiplier, and our resulting memory speed was only 1044 MHz.

Not happy with such ordinary data rates in so highly an overclocked system, we were instead forced to look for lower latency values in an effort to reach maximum performance. The resulting CAS 5-4-4-12 stable timings far exceed the memory's EPP timings of 5-5-5-15 at 1000 MHz data rate.

Graphic Overclocking Versus Hot 3-Way Action

A third GeForce 8800GTX wasn't included in the original configuration because we thought we could get better overall performance from overclocking, but we did leave room for a triple-graphics option by forgoing any additional sound card. As a bonus feature, our high-end overclocking charts include 3-Way SLI tests.

In 3-way SLI mode our 8800GTX samples got so hot that the labels started blistering during the harshest gaming conditions. There simply isn't enough room between cards to allow adequate airflow, and we'd suggest anyone attempting 3-way SLI choose liquid-cooled graphics cards instead.

Though the cards were heat-limited to stock speeds in 3-way SLI, we were able to get a mild overclock out of two cards. At 622 MHz GPU and 2040 MHz graphics memory data rate, the performance should barely exceed the stock 575 MHz GPU clock and 1800 MHz memory data rate.