If you are regular Skinnee Labs readers then you already know we do things a little differently. For those of you who found your way here for the first time, brace yourself, because we go all out on our test methodology and specification. We are not here for publishing fluff pieces; we want the hard data on every product we test even if that means more work. After all, we want to know how well each product performs and where to spend our money as well… we are enthusiasts just like you are. With the proper forewarning given, strap yourself in as we dive into the test method and specification.
When planning your loop there should be a list of things that come to mind, flow and pressure should be near the top of that list. Pressure drop is the measurement of inlet pressure minus outlet pressure, or the pressure loss of flow through the component.
I have a flowmeter and manometer permanently installed in my CPU Block Testbed. There is no way for me to vary flow in my loop other than by changing pumping power. The flowmeter is mid-loop but the manometer spans the CPU block. The bottom port on the flowmeter is the inlet and top port is the outlet. I use Bitspower cube fittings (which also house the block inlet/outlet temperature sensors). At the block, I use use Bitspower 1/2″ fittings with a Bitspower rotary at the inlet and a Koolance rotary at the outlet. For the pressure lines I use Feser 3/8-1/2″ tubing connected to Koolance 3/8-1/2″ compression fittings with Koolance G1/4 to G1/8 adapters at the manometer. Numbers shown are the pressure drop numbers of just the block, the fittings and tubing pressure drop has been subtracted.
Each block is tested in five series of tests. The first test is seven mounts with Arctic Cooling MX-2, this is the main baseline testing. The second test is seven mounts with AquaTuning Silver Grease, this helps determine the quality of contact. The third test is a mount using Indigo Xtreme and running nine different pump settings, this helps determine maximum performance (best mount), temperature vs. flow rate, pressure drop, and quality of contact. The fourth test is a second Indigo Xtreme mount testing and finalizing all the same aspects of performance. The fifth test is seven mounts with the stock/included thermal compound, this helps determine out-of-the-box performance. For MX-2, ATSG, and Stock TIM testing, the worst two of the seven mounts gets excluded from the average. This is done so the final result is more representative of what to expect when you get a good mount.
Each mount and pump setting run is a 2 hour and 35 minute test with OCCT 3.1.0 with the first minute at idle, the first thirty minutes of load discarded (as the block reaches equilibrium with the water temperature and as water temperature steadies), 2 hours of load are used in every data point, then the last four minutes are idle. I use OCCT’s Small Data Set algorithm at normal priority and I use OCCT’s internal logging (I’ve found no other application can reliably avoid stalls and log a data point every second).
Each run I log flowrate and pressure drop. Between each mount I do a basic cleaning of the IHS using Xtreme Clean and a lint free wipe. Every TIM switch I do a complete clean with Xtreme Clean and also Q-Tips around the edges of the IHS. The CPU is never removed from the socket; the socket latch is never opened.
Calipers will not be used with any mounting system, even if recommended by the manufacturer–we believe calipers are a tool that should not be required when mounting a CPU block. With all mounting systems, mounting will be done by feel; a good mounting system should separate itself from a bad mounting system. In the (hopefully) rare case of a broken mounting system, a backup will be used. If no backup is available or if the backup also fails, a mounting system will be concocted using the various spare parts available and will attempt to mimic the core functionality of the stock mounting system.
Airflow is provided over the RAM, VRMs, IOH sink, and over the GPU. Testbed failure is the biggest concern we have for testing longevity–so far each CPU Block Testbed has failed and ushered in the next set of testing (rather than completing a set of testing to the full degree we would like to do)–keeping the motherboard and secondary components cool is one way we’ll help mitigate component failure. When a single component fails, the entire testbed is considered dead and the test roundup is finished.
The watercooling loop I use is fairly abnormal, but allows me to test the way I want to test. The biggest abnormality is that the radiators and CPU block are not in the same “subloop.” The radiators are in a loop with the reservoir and their own pump; the CPU block is in a loop with the reservoir and four DDC pumps. This allows me to isolate the flowrate through the CPU block from the flowrate through the radiators. Not all radiators have the same performance response to flowrate (and thus, , so it’s better to just completely remove them from the varying flow.
24 Dallas One-Wire temperature sensors are used throughout the loop and on the radiators. Twelve are used on the radiators. Six are used in the water in the radiator subloop (two at the radiators’ inlet, four at the radiators’ outlet) and sixe are used in the water in the CPU block subloop (four at the block’s inlet, two at the block’s outlet). All but the six temperature probes in the CPU block loop play a secondary role in the results. I have found that basing results on Air Temperature instead of Water Temperature to be extremely unreliable–maintaining consistent dust levels on the radiators and maintaining temperature gradients across the radiators are basically impossible in the long term.
Water temperature is measured as the average of the block inlet and outlet temperature values (four inlet probes are averaged to a single inlet value, two outlet probes are averaged to a single outlet value). Due to the fact that flowrate is not held constant across blocks (and especially not during flowrate vs. temperature testing), this is the best way to gauge the temperature in the block. Exclusively using block inlet or block outlet temperatures artificially favors low-restriction and high-restriction blocks, respectively (as well as skewing flowrate vs. temperature testing).
All the Dallas One-Wire temperature probes are calibrated against each other so that if they’re in the same place at the same time, they’ll all read the same thing. From there, we’ve calibrated and linearized the raw readouts from the CPU thermal sensors to the Dallas One-Wire probes. This means that for every 1C increase in water temperature, the CPU now responds with a 1C increase in CPU temperature. We’ve found that Intel CPUs, depending on their voltage, have different thermal sensor behavior and this can dramatically skew results if uncorrected.
Below is list of all of the tools, gadgets, goop along with all the hardware used for testing: