Originally Posted by skupples
Originally Posted by eduncan911 Warning: Spoiler! (Click to show)
Update to my build... I've been sitting on my board for about 6 days and just now getting to the waterchiller build part and TECs within the self-contained unit (you can see the housing I am using on top of the chassis in the pic below). The chassis also doesn't like E-ATX boards too well. I had to cut away to get the ASUS Rampage IV Extreme to fit. This board is no exception (the SATA ports are in the same location, dammit).
Believe it or not, this is actually my 3rd most expensive "personal machine" build in my 25 years of building servers and PCs (think quad Xeon builds in the past).
This one is by far the most fun with the built-in ESA Master Control Board that controls all fans, temp, waterchiller's TECs, PWM pump and all - with its firmware. I've been slowly reverse engineering the MCB's firmware and ESA controls over the last few years (even have a Wiki built on it). This build will finally force me to complete the reverse engineering, since Windows 8.1 is killing the 6-year old ESA tools installed from Nvidia. Time to get back to work on this stuff.
What temp will you be setting the chiller to?
Interesting you ask that... Actually, none. You can read more about the H2C here on my wiki: http://730x.org/h2c
In short, the pelters are only activated at high CPU temps. I have them set at 50C to activate at 0.5V, and ramps up to 12V at 70C. The 4x 35W 16V pelters are divided into two circuits with two pelters in series, that cuts the voltage in half to 6V and a very efficient ~24W removal - at full load, though they never run at full load (see below). The entire TEC waterchiller only removes about ~90W in my tests. The 120x38mm heat exchanger can remove about 280 to 320 watts with a high static pressure fan. Therefore, the combination can remove about ~400W of heat from the CPU at max load, while idling dead silent. Though it never gets that high with a 4.7 Ghz 4930k. My 3930k at 4.8 Ghz did run hotter than the Ivy Bridge-E.
To recap about TECs: they are very inefficient at full voltage. Your target voltage should always be between 20 and 40% of their rated max. Any more than that and you run deep into the diminishing returns - on a huge curve (the higher volts you go, the less and less watts is removed).
The MCB also turns down the voltage to the TECs if the watertemp gets close to ambient - so that no moisture starts to form anywhere to damage the system.
Finally, the entire system is easily removable, intact. This is so I can service the CPU, motherboard, etc without breaking the loop.
Those benefits (easily removable/serviceable, idles silently, automagically ramp up/down only when needed) outweigh the trade-offs to me (not being dead silent at full load and not idling sub-ambient).
I have an 4930k at 4.7 Ghz that idles at 28C with the H80 waterblock/pump combo unit with a H80 120x38 heat exchanger and 5500 RPM high-static fan at 800rpms, so I'm not even close to needing the TECs at idle that are in the loop. Though that was for a client build using the RIVE. The trick is to keep all of the Intel power saving features enabled, but with enough vcore so it is stable when it ramps up from 1.2 Ghz to 4.7 Ghz. Lots of load line calibration and tweaking required for that (and every 4930k chip is different, I've ran through 3 so far and each one required slightly different LLCs). Under full IntelBurnTest load, with 1.28V vcore, the TECs are only activated to about 40% (about 2.3V to each pelter), fan at 2,100 rpms (noticeable, but not too much), which cools the entire system at about 65C. There is more room there, but it's very stable like this.
My RIV:BE build will be using an old Apogee XT Rev2 and Swiftech MCP-35X PWM pump that I have lying around for a few years, along with an 120x38 3/8" heat exchanger sourced from an H80i I tore apart (didn't even turn it on - opened box, cut the hoses, got me a new waterblock/pump combo and radiator to work with). I'm pretty sure the Apogee XT Rev2 is far more efficient than the H80/H80i waterblock, as well as the head pressure I get from the MCP-35X at idle is far greater than the minuscule "dripping?" head pressure from the H80 unit. The H80i seems to have a bit more than the H80, but not much.
To recap: the 120x38mm heat exchanger is by far enough to cool the 4930k @ 4.7 Ghz at idle with a very high static pressure fan, at idle speeds (800rpm) so you don't hear it. It's only when pushing the system under load that the 120x38 cannot handle it alone, and that's when the TECs and higher RPMs on the high-static fan come into play. They start ramping up at 50C.
The MCB controls the speeds of the Fans, Pump and TEC voltage (since the same 5500 fan does double duty of radiator and TEC heatsink fan cooling). This is all based on the temp range I set it to.
"Set it too" is the inherit problem I am having with the Asus RIVE, and now the RIV:BE though. All motherboards (except the EVGA X79 Dark) allows me to set a temp range of the PWM Duty Cycle based on the CPU temps (which is the single input I am using to the MCB). The problem with the RIVE is that it has a forced minimal of 20% duty cycle. Well, per the RIV:BE's manual, that's even worse as it has it forced at a 40% duty cycle minimal. That's really annoying since my MCB is setup to take a 0% input.
I have to revert to using SpeedFan instead of the RIVE's BIOS for that stuff, only on the Asus RIVE and RIV:BE though. The Asus RIVF was fine since it let me set 0%, and the Asrock board I had while letting me set 0%, had a seriously long delay in ramping up. I hope SpeedFan can pick up on the RIV:BE's fan controllers as well, or I may be SOL.
Why does Asus have to set a minimal duty cycle anyways? Really irks me.Edited by eduncan911 - 11/25/13 at 6:07pm