For our overclocking comparison, we chose 1.60 V CPU core, 1.50 V CPU FSB, 1.55 V northbridge core and configured each motherboard with the closest available settings. We also increased our DDR3 SDRAM from 1.80 V (0.30 V over stock) to 1.90 V, and our DDR2 SDRAM from 2.10 V (0.30 V over stock) to 2.20 V. We used Prime95 to check CPU stability, and Memtest86 v3.4 to check DRAM stability.
The MSI X48 Platinum leads slightly in maximum achievable FSB clock, reaching 542 MHz (FSB-2168) on air cooling. As impressive as these results appear, none of today's X48 Express motherboards noticeably outpaced previously-tested models that used Intel's P35 and X38 Express chipsets. This provides further evidence that a "higher validated" FSB speed is no guarantee of better overclocking capability.
The DDR2-equiped GA-X48-DQ6 supported the highest stable CPU speed, but the difference between best and worst overclockers was only two-tenths of one percent (0.2%).
One of the more stunning complaints we've heard from readers is that they've purchased a highly-rated board from one of our previous reviews, only to find it unstable when configured with four memory modules. We wanted to be sure that each motherboard was at least capable of booting with four modules at default voltage, and operating stably with four modules at rated voltage and timings.
Our first test was to install four modules and verify that all systems were bootable. All motherboards passed the first part of the test, but one of the "bootable" systems wasn't anywhere close to stable.
We then set the voltage to 1.90 V at CAS 7-7-7-20 (DDR3) or 2.20 V at CAS 5-5-5-15 (DDR2) to find the highest stable speed at relatively good latency values.
The DDR2-equiped X48-DQ6 dropped out of the race after an undefined electronic failure. These types of failures happen regardless of manufacturer - hence the warranty - but we were unable to get another sample in time for testing.
The MSI X48 Platinum booted with four modules installed, but was never stable in that configuration. With two modules installed, it only reached 1554 MHz. We could have relaxed timings a bit further, but our modules are actually XMP rated for operation at CAS 7-7-7-20 and DDR3-1600 speed. Owners of the X48 Platinum should be aware that because the board doesn't have automatic settings for individual timings, tRFC must be manually increased whenever using manual latency values and high data rates. We selected a tRFC of 48 cycles, and loosening this value further did not allow higher DRAM clocks.
The GA-X48T-DQ6 did well with two modules installed, reaching 1712 MHz. The test wasn't so smooth with four modules, however, as the motherboard simply would not work using the highest memory multiplier (DDR3-1333 with our FSB-1333 processor at stock speed). Dropping the memory multiplier to 1.60x FSB clock at 333 MHz bootstrap by using the "3.20b" BIOS setting provided better stability, and overclocking the FSB squeezed 1382 MHz from our modules.
What was true of the X48T-DQ6 is also true for our X38 reference motherboard, but with different resulting clock speeds. The P5E3-Deluxe reached 1744 MHz using only two modules, and getting to 1329 MHz with four modules required dropping the FSB strap to 333 MHz and the memory ratio to 1.60x FSB clock, followed by overclocking the FSB to the resulting DRAM speed.
The Asus P5E3 Premium was the only motherboard in this comparison to do exactly what it was supposed to do regarding memory. At a memory ratio of 2.0x FSB clock and 400 MHz FSB bootstrap, we were able to start at DDR3-1600 and overclock to DDR3-1752 with two modules (DDR3-1712 with four modules installed). The P5E3 Premium suffered a large number of miniscule losses in our benchmark suite, but made up for it in stability. In fact, stable chipset timings are probably responsible for both the tiny performance decrease and the enormous stability increase.