Overclocking
Overclocking
For our overclocking tests we push the FSB as high as possible. Bear in mind that the actual CPU used may hinder the overclock of the motherboard.
To do this we lower our QX9650 CPU's multiplier to its minimum (x6), set the memory speed to the lowest available divider, and increase the appropriate voltages to 10% over reference levels, or as close as we can get with the BIOS options available without going over.
We test for
stability using four instances of Prime95, incrementally
increasing the FSB clock between runs until failure. We then back the
FSB back down until we find the maximum stable limit.
With overclocking, your results may vary from ours. Some examples of a
board are
liable to overclock to a greater extent than others, and as such the
figures given should be taken as a rough guide rather than a definitive
or guaranteed limit.
Intel DX48BT2
Using our test procedures we managed to overclock our QX9650 to 394MHz for an effective 1,576MHz FSB. This is despite our QX9770 running stable at its native 1,600FSB, which leads us to believe that with further tweaking you should be able to achieve a little better, but Intel pushing this as an enthusiasts' board for overclocking may be over-reaching.
It has neither the ease of use of an ASUS or Gigabyte board,
nor the range of options of the DFI, so even if your willing to spend a
lot of time tweaking away to get a stable overclock, you're likely to
look
elsewhere.
DFI LANPARTY LT-X48-TR2
Overclocking
using the DFI board proved interesting to say the least.
DFI provides such a wide array of options for tweaking, it can leave
even experienced users a little baffled.
Trying to overclock starting from our saved QX9650 profile and
adjusting the settings according to our overclocking procedure resulted
in dismal results, refusing to POST when the FSB was raised to even
340MHz. Clearly something was amiss, so starting from optimised
defaults
we meticulously changed the settings one by one to uncover the root of
the problem.
Eventually we traced the problem down to changing memory
speeds or timings on the profiles. It seems the
‘auto’ timings
are saved as set numbers after the first setup, yet appear simply as
auto in
the bios. So trying to make adjustments with the 'auto' settings
enabled resulted in failure as the auto values were no longer correct
for the new speeds or timings. Even setting the memory page to
optimised defaults didn't fix the problem, only starting from scratch
with optimised defaults and setting the memory to the desired settings
before you save for the first time. Hopefully DFI will fix this issue
in a future BIOS release.
With the cause of the issue finally divined, we managed to achieve a
respectable if unspectacular 444MHz (effective 1,776FSB). With array of
options available, fine-tuning may see further gains, but to
allow a degree of comparability we try to spend the same amount of time
overclocking each board.
EVGA 132 CK-NF79
EVGA's offering gave us a maximum overclock of 426MHz (effective 1,704FSB).
Following NVIDIA's recommendations to leaving the voltages - except for memory - and everything else except for CPU multiplier to auto, enabled the board to boot into Vista at 450MHz (1,800 QDR) but with substantial instability.
MSI P7N Diamond
MSI's 'diamond' product proved everything but in our overclocking tests, managing a meagre 349MHz (1,396FSB effective) FSB.
The FSB-to-memory ratio couldn't be set to sync as the BIOS doesn’t allow FSB adjustment without also specifying the memory-speed manually.
The situation wasn't improved by following NVIDIA's advice to leave voltages - except for memory - and all other settings except for CPU multiplier, FSB and Memory speed to auto, either..