NVIDIA nForce 590 SLI Technology Chipset Overclocking
DU-02451-001_v01
24
05/17/06
Balance CPU Frequency, HyperTransport, and Memory
Overclocking Results
Since CPU frequency, HyperTransport bandwidth, and memory frequency
overclocking all depend on CPU HTT reference clock, it is often difficult to obtain
a specific configuration that would generate a maximum overclocking for all.
Hence, overclocking tradeoffs must be made in order to generate the optimum
achievable performance. Higher CPU frequency generally influences performance
the most in comparison to HyperTransport and memory interfaces. 3D applications
are generally CPU-bound, so increasing CPU frequency tremendously boosts
performance. A frequency combination that maximizes CPU frequency is often the
best tradeoff.
Review what you have achieved by following the steps in the previous sections will
provide some guidance as to how a combination can be obtained. As described in
the section called “
CPU HyperTransport Overclocking
” on page 15, the CPU
HyperTransport interface is capable of running at about 1.4 GHz (280 MHz × 5).
The CPU frequency achieved is 3.07 GHz (256 MHz × 12). Memory is capable of
running at 1,108 MHz:
400 MHz for even-numbered CPU multiplier × percentage overclocking of reference clock
400 MHz × 277 MHz ÷ 200 MHz = 1,108 MHz
Since all three variables have different CPU HTT reference clock settings to achieve
independent overclocking, a value for the reference clock must be selected to
achieve maximum performance for all.
In this case, using 256 MHz as a reference clock generates:
3.07 GHz CPU frequency.
1,280 MHz HyperTransport frequency
1,024 MHz memory frequency
Under other circumstances, however, increasing CPU HTT reference clock and
reducing CPU multiplier while maintaining the same CPU frequency may produce
better results.
It is all about balance. Make it a priority to maximize CPU frequency (sometimes by
lowering the multiplier and raising the CPU HTT). Next, consider memory timings.
In the example setup discussed earlier, the memory frequency is a bit lower than the
physical modules are capable of running. To alleviate this performance reduction
impact, tune latency to improve memory throughput. A memory timing
configuration that reduces
t
CL
,
t
RCD
,
t
RP
,
t
RAS
,
t
RD
,
t
RC
,
t
WTR
, or
t
RAS
will boost
performance but might cause instability. Only trial-and-error will help optimize
memory latency. Finally, focus on HyperTransport speed. HyperTransport
overclocking provides some added performance benefit, but its impact is not as
significant as altering CPU frequency and memory performance settings.
Fortunately, NVIDIA nTune 5.0 provides a complete, easy-to-use solution to
optimize memory performance in a Windows environment.
