The Obsidian Series 750D "Yamamura" Custom Liquid Cooling Build: Optimization

This is the fourth in a series of blogs about the Yamamura Obsidian Series 750D build. The first details component selection and can be found here; the second details the assembly and can be found here; and the third details overclocking and system performance and can be found here.

While one big reason to assemble a custom liquid cooling loop is the ability to massively increase your system’s capacity for dissipating component heat, and thus improve overclocking headroom (or at least clock stability with NVIDIA GPUs), another less talked about key benefit is the potential to run your system far more quietly.

This is perhaps one of the biggest reasons to “overdo it” on cooling capacity. Yamamura would have more than adequate cooling capacity with just the single 360mm radiator and three fans. Instead, the massive cooling capacity of the Yamamura’s loop and use of push-pull fans on two of the three radiators lets us substantially reduce the amount of noise the system produces. The D5 Vario pump, potentially one of the loudest components in the system, is largely decoupled from the chassis and produces virtually no noise at its midrange setting.

To keep noise levels under control, the twelve SP120 LED radiator fans are connected to a single Commander Mini (note that this does exceed spec on the fan headers and comes dangerously close to exceeding the overall unit spec) using fan header splitters.

The loop is also arranged with heat generating components essentially isolated between radiators. The heat generated by the Intel Core i7-4790K immediately feeds into the 360mm radiator in the top of the enclosure where it’s dissipated and the coolant flows into the two GeForce GTX 980s. After that, the coolant flows into both 240mm radiators in sequence before heading back into the reservoir.

The radiator-component arrangement lets me isolate fan groups to compensate for the heat generated. Fans are controlled in pairs, and I used Corsair Link to program very permissive fan curves: these fans all run at their minimum speed (~750RPM) until the CPU exceeds 75C or the GPUs exceed 70C. Meanwhile, the almost ornamental SP140 in the rear exhaust of the Yamamura has had its voltage reduced to just 5V: enough to spin and move some air, but not enough to generate any real noise.

How does it work out in practice? Incredibly well, actually. Yamamura’s massive cooling capacity and push-pull on two of the three radiators means the fans frankly never have to spin up. It’s only when doing extreme stress testing on the CPU that any of the fans spin up; otherwise, everything runs at its lowest speed and component temperatures remain very reasonable. I have to be seriously stressing the GPUs to get them to break 55C; the CPU jumps around a little more due to the high overclock, but spends most of its time under 60C.

This is why I maintain that the Commander Mini’s true mission in life is to handle fan control for a custom loop. A PWM-controlled pump like Swiftech’s MCP35X could be operated by the Commander Mini and kept at its lowest speed until heat in the loop reaches a certain point, at which time it can kick into a higher gear.

In the next and final chapter of the Yamamura build log, we’ll do a postmortem and look at what worked – and what didn’t.