Power, Heat, And Efficiency
We loaded up each system first with eight threads of Prime95, then with the FurMark stress test, and finally with FurMark and seven Prime95 threads simultaneously. We then waited around 30-45 minutes for the power consumption to settle to a consistent level before taking our global wattage readings from the AC power source.
Our $2,000 build uses far more power than its predecessor, even at stock settings. Overclocked power consumption caused even greater concern, as it appears to exceed the power supply's rated capacity. A quick check of the DA750 specification sheet put us at some ease however, since a shift from 50% to 100% load causes the power supply's efficiency to drop from 80% to 77%. That means it was only outputting between 660 and 680W at the above reading, and our remaining discomfort is caused by the idea that we're wasting nearly 200W of power in the conversion process.
This month’s full-load GPU temperature remained relatively constant, regardless of ambient conditions or overclock levels, simply because we left its automatic fan control enabled. Not seen above is the CPU temperature reached with the $3,000 PC during the combined load test, which is an important exclusion, since the CPU and GPU of that system were tied into the same liquid-cooling system. Looking back at our test notes, the CPU temperature under combined load was approximately the same for both systems, while the $3,000 machine’s reduced cooling performance was due mostly to a flow reduction at the GPU water block.
We also tried the powerful 38 mm and 3,900 RPM CPU fan photographed and discussed in our previous pages and found that CPU temperatures dropped by five degrees Celsius, while noise more than doubled. That five-degree reduction didn’t even allow the smallest increase in CPU clock speed. It’s a fine fan when speeds are reduced, but probably not worth the effort for most builders.
What our $2,000 machine lost in CPU overclocking it more than made up for in graphics prowess. A combined performance number relative to our comparison machine can now be used to calculate efficiency.
High-power consumption turns the new system’s performance victory into an efficiency defeat. We could almost take solace in the increased efficiency that overclocking brought the $2,000 machine, if not for the fact that even its most efficient settings are a full 10% lower than our previous build’s least-efficient settings.
