Test design and methodology
You know the chiller, with it I guarantee the constant water temperature. In addition, there is a 20-liter expansion tank next to the chiller, so that we could also better compensate for fluctuations. The graphics card runs with Furmark (burn-in mode, post-processing on) at full load. Each run starts with cleaning the board, coating each with thermal paste, heating it up until it is completely warm (about 10 minutes) and then logging the temperature for 5 minutes. The average value is then calculated for these time windows, which together with the other two runs per method results in a final average value.
The room temperature of 22 °C is well within the limits, so I don’t have to cool anything down any further. This all sounds quite elaborate and unfortunately it is. But it’s all about evidence, and I’m pretty picky about that. Curious to see what came out of it? So was I.
The measurement results
Since I have already done the whole thing in a similar way internally for an AMD board partner card with an RX 6950 XT and the results looked very similar in percentage, I will save the second run, you can definitely transfer this to the big AMD cards. There, however, the difference between the hot-spot temperatures was much greater. But AMD also determines this at other points. However, the core message is almost identical in the comparison. Let’s see how the methods ultimately differ:
There is a difference of about 3 Kelvin between the sausage and the blob at the edge temperature (CPU diode). That doesn’t look like much, but at 20 °C water temperature it’s worlds apart when you consider the percentage difference of the delta values to the water. Between the blob and the solid surface, there is then another ample 2 Kelvin. Ergo, the sausages create a full 5 Kelvin less compared to the fully filled surface. By the way, with air cooling with the original cooler and 69 °C GPU diode, it is about 10 Kelvin difference, because then the card with full surface already goes towards 80 °C. But since that is very inaccurate, I prefer to measure with water cooling.
Summary and conclusion
There has already been a lot of positive feedback. This comes both from some manufacturers who have adapted everything so similarly, at least when screwing, or have implemented the application of the paste with gaps, and from readers who sometimes reported very euphorically (although you never know exactly what else went wrong beforehand). It is certainly not a miracle method or reinvention of physics, but simple logic after a thorough examination of the details.
In any case, I will never mount the current cards differently again, because then I would slow myself down. I can only recommend everyone to say goodbye to these thin pastes. In the long run, it’s really nothing.
163 Antworten
Kommentar
Lade neue Kommentare
Mitglied
Mitglied
Veteran
Urgestein
1
Urgestein
Veteran
1
Urgestein
Mitglied
Urgestein
Mitglied
Urgestein
Veteran
Veteran
Veteran
Mitglied
Alle Kommentare lesen unter igor´sLAB Community →