Performance and Test Results
Before we jump into the results, we have to put the TIM print photos on display. All of the cards made good contact to the GPU core, memory and Mosfets, and pre-spread TIM application to the GPU was a little overboard. Next time we venture down this path we will go back to the tried and true dollop method where we let pressure spread the TIM.
Those of you with keen eyes will notice that several blocks have two small sections of the thermal pads over the chokes removed; this was to ensure the sensors were not affected by thermal pad contact. Aqua Computer and Danger Den are the only two using TIM on the memory, with EVGA stepping outside the norm with TIM on the mosfets.
Thermal Results – Stock Clocks
We have finally made it to the thermal results; since I have made you endure my textual ramblings this long, I will let the charts do most of the talking. For all stock runs, we used our saved profile in MSI Afterburner, of 700/924/1400 @ 1065mV.
If we exclude Bitspower as the high-side outlier, all of the blocks fall in a 3-4C temperature range. The tight grouping surprised me a little; I expected to see a larger temperature range. The GPU focused cooling on the EVGA Hydro Copper really stands out, with Koolance and Aqua Computer trailing by over 1C, a small temperature difference in terms of stability but we look at thermals like racing lap times where lower is always better.
Putting our crafty (I think they are crafty) VRM sensors to use, we see the EVGA block move to the other side of the pecking order which is expected with their design. Excluding the TIM on the VRM’s experiment, mosfet/VRM temps stay in the 50’s, remember this is stock voltage as well so 50C gives us plenty of room to increase voltage for overclocking. The surprise here to me are the Aqua Computer and Watercool blocks which some active cooling over the VRM’s on the lower end of the chart compared to Bitspower’s clean flow path.
Thermal Results – Overclocked
Time to bump the clocks and voltages and see how our blocks perform with the added thermal load. In pretesting we settled on a GPU Core/Shader of 875/1750 and memory clock of 1065/2130 with a bump in voltage to 1138mV. With three runs each for all of our blocks, the data all rolls up to the two charts;;
As with the stock profile results, the EVGA Hydro Copper takes the low and Bitspower takes the high temperature spots. Those in middle changed order, but all within 2C. Even though our average GPU core temp did not rise a significant amount with our higher clocks, only 5-7C.
We may not have seen a big thermal jump on the GPU core, the mosfets on the other hand shift up 10C or more. The EVGA temperature might stand out, but 90C for your mosfets is well within the operating range. At the top of the chart is our experiment with TIM versus thermal pads, we peeled off the thermal pads and went with a serving of MX-2 instead and even tried a small dollop of TIM for the chokes as well. Yes, lathering up your VRM’s with TIM does produce lower average temps, but the cleanup was a nightmare between mounts.