Page 1:Because We Can...
Page 2:CPU And Cooler
Page 3:Motherboard And Memory
Page 4:Graphics And Storage
Page 5:Audio And Optical
Page 6:Case And Power
Page 7:Accessories And Installation
Page 9:Test Settings
Page 10:Benchmark Results: First-Person Shooters
Page 11:Benchmark Results: Real-Time Strategy
Page 12:Benchmark Results: Audio And Video Encoding
Page 13:Benchmark Results: Productivity
Page 14:Benchmark Results: Synthetic
Page 15:Power Consumption
One might think that buying Intel’s best desktop processor would lead to a superior overclocking experience. After all, the Core i7 Extreme Edition 965 is supposed to be specifically selected for its tolerance to higher speeds and comes with an unlocked multiplier to further assist overclocking. We had high hopes, since our very first Core i7 retail sample, a lowly 920, was able to reach 4.0 GHz using air cooling and could maintain speeds of around 4.4 GHz on a water-cooled test bench.
Unfortunately, the i7 965 we received required more voltage to achieve similar overclocking-stability levels compared to the previously-tested i7 920. The higher voltage requirement made it produce more heat, which is, of course, an overclocker’s worst enemy. Even with our monster cooling system, the processor we used for today’s system couldn’t even reach 4,266 MHz at 1.45 V, and adding more voltage simply made more heat. Our i7 965 was actually worse than the i7 920 we’d tested before.
With 1.45 core V as the limit we’d chosen due to heat-and-longevity issues, we could easily reach 4.20 GHz but not much more. Additional testing proved that 1.3125 V was enough to keep us stable at this speed, and the lower voltage helped us to reduce the core temperature from approximately 83 degrees peak (Celsius at full CPU load) to around 79 degrees Celsius peak.
Our full-load CPU stability test uses four instances of the Prime95 Small FFTs torture test with affinity set to one instance per, and the temperature readings were taken using Realtemp at an ambient temperature of 26 degrees Celsius. These relatively harsh test conditions assured that the system would remain stable no matter what benchmarks we used.
A speed of 4,200 MHz is not a whole-number multiple of a standard 133.3 MHz base clock, so we increased this clock to 150 MHz. The resulting speed of each memory ratio changed and we selected DDR3-1800.
The higher DRAM clock required looser timings, so we switched from its DDR3-1600 rated 7-8-7-20 latencies to 8-9-8-20.
Nothing is more disappointing than overclocking a graphics card to a super-high speed, only to have it become unstable a few hours later. After trying dozens of settings, we finally settled on a GPU clock of 690 MHz, a shader engine clock of 1,520 MHz, and a memory data rate of GDDR3-2160. This probably isn’t the ultimate speed of these cards, but we couldn’t afford to waste any more time seeking diminutive gains.
One question that plagued our minds was whether the Aquagate Max pump, with its 3/8” coolant lines, was really capable of providing adequate flow to our high-capacity water block and radiator. After we had completed all tests, we pulled the platform and put it on our test bench, which uses a super-sized Swiftech MCP-655b water pump and ½” lines in addition to a three-fan radiator and Apogee GTZ water block identical to those of our $5,000 PC. Peak temperatures dropped by 3 degrees Celsius, allowing us to reach a stable 4,266 MHz clock speed at 1.450 V. These improvements are barely worth noting and were not used for today’s benchmarks.
- Because We Can...
- CPU And Cooler
- Motherboard And Memory
- Graphics And Storage
- Audio And Optical
- Case And Power
- Accessories And Installation
- Test Settings
- Benchmark Results: First-Person Shooters
- Benchmark Results: Real-Time Strategy
- Benchmark Results: Audio And Video Encoding
- Benchmark Results: Productivity
- Benchmark Results: Synthetic
- Power Consumption