Overclocking the $1,300 Micro-ATX system was a lesson in thermal management. Given the less-than-ideal case airflow that we described above, we’re starting with high graphics card temperatures out of the gate. In any case, we began by overclocking the CPU.
The good news is that our Core i7-920 CPU and our DFI X58-T3H6 motherboard were more than happy to go ever higher. With a small voltage bump and a base clock increase to 200 MHz from the stock 133 MHz, the system booted fine at 4 GHz.
A quick Prime95 run shut the system down quickly. I say shut the system down because it didn’t lock up or crash. It simply shut down softly. The screen even pleasantly faded out, leading us to conclude it was shutting itself down due to thermal protection.
Playing out the same exercise while running Real Temp to monitor temperatures verified our hypothesis. Once Prime95 was launched, the CPU core temperature went to 100 degrees Celsius almost immediately. While idle temperatures were an acceptable 45 degrees Celsius, putting the machine under load would send them skyrocketing.
From here on in, it was a battle to see if we could keep the temperatures and voltages down low enough while seeing how high we could push the clock speed.
Everything went pretty smoothly, although the DFI LANParty Jr. X58-T3H6 demonstrated an interesting quirk: the CPU voltage as reported by CPU-Z was higher under load than the voltages set within the motherboard’s BIOS. For instance, when voltage was set to 1.21 V in the BIOS, CPU-Z reported a voltage of 1.26 V. Under load, it jumped to almost 1.3 V. One tenth of a volt might not sound like much, but it makes a huge difference to the CPU and was enough to create a lot more heat than we wanted. Fellow SBM writer Thomas Soderstrom also experienced the issue with the X58-T3H6, so it’s probably not a problem limited to this particular sample.
Adjusting for the BIOS anomaly, we found the system was completely stable at 3.5 GHz at a low 1.264 V on the core as reported by CPU-Z, running Prime95 overnight with no hiccups. Unfortunately, the CPU temperatures were in the high 90 degree Celsius range. While real-world apps won’t stress the CPU as much as a Prime95 run will, we wanted a bit more margin of error.
In the end, we settled for 3.44 GHz with a 173 MHz base clock and the CPU voltage set to 1.2125 V (1.264 V on the CPU under load as reported by CPU-Z), which bought long-term Prime95 run CPU temperatures down to a more acceptable low 90 degree Celsius range.
Here are the Genie BIOS settings we used for the DFI Lanparty Jr. X58-T3H6:
- PPM Disabled;
- Turbo mode disabled;
- QPI frequency set to 4.8 GT/s (BCLK*18*2);
- CPU base clock set to 172 MHz;
- DRAM frequency set to BCLK*08 (1,376 MHz);
- Uncore frequency set to BCLK*18 (3,096 MHz).
In the CPU-settings sub-menu, CPU thermal management, EIST, CxE, and virtualization were disabled:
In the voltage-settings sub-menu, CPU VID control was switched from AUTO to 1.2125 V and the OCP setting was increased to 180 A.
With the CPU clock speed maxed out, we turned to the graphics cards. Of course, with the GPU fans at 100% at stock speeds, we didn’t have a lot of headroom with which to play. Our stable GPU overclock was 600 MHz GPU (a 10 MHz over-clock), 1,030 MHz memory (a 31 MHz overclock), and 1,300 MHZ shaders (a 4 MHz overclock).
While these results weren't impressive, we expected the CPU frequency increase of almost 800 MHz over the stock speed to lead to a real jump in performance, as long as the application depended mostly on CPU power.