<> = Noise levels at different speeds = == Pump == This is an EK D5 pump in a FLT120 reservoir. || PWM || RPM || Noise || || 40% || 1950 || Silent || || 50% || 2500 || Just audible || || 60% || 3060 || Quiet || || 68% || 3520 || Mild || || 75% || 3950 || Max acceptable level || || 100% || 4775 || Pretty loud || == Pump speed vs flow == Same EK D5 pump in a FLT120 reservoir. || Pump PWM || Pump RPM || Flow meter volume || Flow meter tacho || Notes || || 50% || 2495 rpm || 0.6 L/min || 13.75 rps || - || || 47% || 2330 rpm || 0.45 L/min || 11 rps || - || || 35% || 1650 rpm || 0.1 L/min || 3 rps || Sometimes triggers low-flow alarm || || 34% || 1620 rpm || no reading || no reading || Always triggers low-flow alarm || == Fan == Arctic P12 PWM PST I think this was with two fans. || PWM || RPM || Noise || || 20% || 530 || Silent || || 35% || 870 || Effectively silent || || 40% || 965 || Just audible || || 50% || 1150 || Pleasant || || 75% || 1575 || Max acceptable level || || 100% || 1925 || Pretty loud || = Heat loads = || Components || Estimated load (Watts approx) || || Idle || 25 || || Full CPU || 120 || || Full GPU || 250 || = Results = This is mostly about finding a sweet spot for being quiet at idle, and sufficiently cool under load, without being too noisy. This is the new loop layout, after Maru helped out with reworking everything. 280mm radiator with 2x 140mm Corsair maglev fans. == Low speeds == * 34% fan - 655 rpm * 45% pump - 2200 rpm * 0.5L/min flow rate - 12rps on the meter || Load || CPU temp || GPU temp || Loop temp || Notes || || CPU only || 84 C || 29 C || 28 C || Cinebench R23 || || GPU only || 63 C || 54 C || 33 C || Furmark || || GPU and half-CPU || 90 C || 59 C || 37 C || Furmark and Cinebench limited to half usual threads || || GPU and full-CPU || n/a || n/a || n/a || Not tested because the CPU throttles at 90C || Loop speeds are simply too low to get heat out of the chip fast enough, the CPU will lower the boost ceiling to try and stay at 90 degrees. With some GPU undervolting we get a very healthy improvement in temps: || Load || CPU temp || GPU temp || Loop temp || Notes || || GPU only - stock || 63 C || 54 C || 33 C || - || || GPU only - undervolted || 57 C || 48 C || 31 C || - || == Medium speeds == * 50% fan - 1000 rpm * 60% pump - 3060 rpm * 1.0L/min flow rate - 24rps on the meter || Load || CPU temp || GPU temp || Loop temp || Notes || || CPU only || 82 C || 27 C || 27 C || Cinebench R23 || || GPU only || 54 C || 48 C || 31 C || Furmark || || GPU and half-CPU || 82 C || 51 C || 33 C || Furmark and Cinebench limited to half usual threads || || GPU and full-CPU || 90 C || 52 C || 35 C || Furmark and Cinebench with unlimited threads || This is a fair bump in flow, which suggests that the CPU isn't being cooled effectively when running at fill-tilt, I think. I wish I'd taken note of the CPU speeds, because then the CPU-only and CPU+GPU test cases could be compared. == High speeds == * 60% fan - 1200 rpm * 100% pump - 4800 rpm * 2.0L/min flow rate - 47rps on the meter || Load || CPU temp || GPU temp || Loop temp || Notes || || CPU only || 83 C || - || 29 C || Cinebench R23 - 140W || || GPU only || 55 C || 48 C || 33 C || Furmark - 280W total power (230W GPU + 50W CPU) || || GPU and half-CPU || 82 C || 50 C || 34 C || Furmark and Cinebench limited to half usual threads - 340W power (230W + 110W) || || GPU and full-CPU || 90 C || - || 34 C || Furmark and Cinebench with unlimited threads || Not sure why I don't have GPU temps for all test cases. Very similar to medium flow case, suggesting no benefit from the higher speeds. CPU still hitting it's 90 degree ceiling. = GPU undervolting = Poking at this while running Furmark. * 34% fan * 45% pump Stock voltage settings * 54 degrees C (loop or GPU core?) * 230W GPU power draw * 83fps * 1068 mV * 1950 MHz (GPU core?) Undervolt applied * 47 degrees C (loop or GPU core?) * 185W GPU power draw * 83fps * 950 mV * 1950 MHz (GPU core?) = Links = * Todor's crazy setup: https://goo.gl/photos/EsgB777cs9pke3VMA