Thermal Testing Methodology/Specification
First off, if you came here looking for the same old boring test specification found at other sites then you are in the wrong place. We at Skinnee Labs are not here for publishing fluff pieces, we aim to provide an in-depth look at every product we test and GPU Blocks are no exception. So what you will see in our specification is every little piece detailed, clarified and as many variables eliminated as humanly possible (minus an environmental test chamber). 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 my little soap box paragraph out of the way lets cover the methodology.
Each block is tested in “stock” form, per manufacturer’s instructions as you saw in the installation videos. We use the same inlet and outlet unless specifically called out in the instructions (EVGA was the only exception), with the standard having the outlet closest to the PCI bracket. With the card in the PCIe slot, the QDC’s are connected, VRM sensors plugged in, and power connected I go through a 10-30 minute loop bleed before powering up the board and starting the test. Once the loop is bled, and Windows is running we use MSI Afterburner (1.60.b6) to load the stock profile. GPU-Z 0.4.3 is used to monitor all available data points from the PCB and Wintest is configured for all of the sensors and radiator fan RPM. FurMark 1.8.x is our GPU loading tool of choice, set for 1280×1024, 16X MSAA, Xtreme Burn, Displacement and Post FX. FurMark is run for 60 minutes, with the first 30 minutes being a warm-up period and the final 30 minutes being the data we use for presentation. After the 60 minute run under the stock profile, our Overclocked profile is loaded, but the test is delayed 30 minutes to let the system reach equilibrium again. After the hold, GPU-Z and Wintest start logging and Furmark is kicked off again for 60 minutes. Once the Overclocked run is complete, the machine idled for 20 minutes before I would shut down to start the mount process again.
Sensors are deployed on the intake and exhaust of the radiator, measuring Air In and Air Out data points. In the loop we have two delrin T’s each with two sensors measuring GPU Block In and GPU Block Out temperatures. The sensors are just before and after the Koolance QDC’s connecting the block to the loop and remained static during all tests. We did go a step further and installed two sensors on the GTX480 PCB as well. The GTX480 does VRM temperature sensors available, so we had to fashion our own. You can see photos at the bottom of the page. The VRM sensors were added to provide more data on the VRM cooling that blocks are being designed with. You can approach VRM cooling in two ways, make the cooling adequate to maintain stability or remove as much heat as you can and let your cooling loop dissipate the heat thus decrease operating temperature, head room and heat from case (VRM’s kick off a lot of heat). Either way, the sensors are there to understand our operating limits as well as the blocks ability to pull heat from the VRM’s.
We kicked off a secondary round of testing with some modifications, one involved a fan and the others involved a lot of TIM. We have included the TIM testing here in the comparison, while the fan use will be covered in the individual review of the EVGA block.
There is quite a bit of crossover/overlap from the Methodology and Specification sections, but the specification covers all the nitty-gritty details from the overall methodology.
Three (3) mounts are performed for each block according to manufacturer instructions. Each of the tests is conducted in identical manner to all others. The GPU IHS, memory and mosfets are cleaned using GooGone on a paper towel, followed Arctic Silver cleaning agent #1 applied to the chip and cooler then wiped with a coffee filter, finally cleaned by surface preparation with Arctic Silver #2 and coffee filter wipe. Thermal Interface Material (TIM) is then applied to the IHS using the blob method, where I then take a hard plastic square to spread and even out the TIM. TIM or Thermal Pads are applied to the memory and mosfets as dictated by manufacturer instructions. For block that called for thermal pads over the chokes, two slots were cutout over the installed VRM probes. With the card and cooler prepped, we flip the card onto the block and proceed with the mounting procedure as you witnessed in the videos. If additional experimental tests are performed, the same mounting and cleaning methods are followed. Any and all variations from the factory instructions are detailed in the tests notes. Once the block is mounted, the procedure called out in the methodology kicks in… Connect the QDC’s and commence loop bleeding. Once bled, cables plugged back in, machine is booted and in windows, WinTest is configured, GPU-Z ready and MSI Afterburner profile loaded then FurMark is kicked off. We do use Koolance QDC’s in the loop to cut down on the time required between mounts.
Each test run is 60 minutes long with a 30-minute warm-up and 30 minutes of data used for calculations. A 30-minute warm-up gives adequate time for the GPU to reach maximum temps under load. After each run, files are copied to a network share and the machine is shutdown for the cleaning and mounting process to begin.
You can see the loop order in one of the photos on the bottom of the page, however one photo does not have the King Flow Meter in the loop. Loop order for all tests consisted of Reservoir -> Pump -> Sensor T -> QDC -> Block -> QDC -> Sensor T -> Radiator -> Flow Meter -> Res. Two Bitspower rotary 90º fittings were used on the block to help with tube routing; the Bitspower block required an extra set of rotary 90’s in order to loop up on the bench.
Below is list of all of the tools, gadgets, goop along with all the hardware used for testing.
The initial plan was to circle back around and test the stock cooler for a comparison to each of the blocks and show the OC gain of moving from air to water, but we ran into a bit of a problem shortly after the stock cooler was mounted back on. Initially, the card ran just fine, completing two run under the stock profile and the overclocked profile did not. Therefore, we tweaked the clocks a bit but left voltage the same as our overclocked profile used on the blocks. That alone shows there are gains in clocks and not just silence going to water. Unfortunately, we only were able to complete one overclocked run with the reference cooler, during idle and cool down following the 60 minute run a mosfet blew. With over 30 mounts and a torturous life of nothing but boot screens and FurMark, the card had enough. A moment of silence for our sacrificial GPU…
There you have it, the full test methodology and specification. Without further delay, let us move on to the performance results.