GTX 1080 Ti Hybrid Thermal Results – Liquid Cooling vs. Founders Edition
This first chart has a lot going on, so let’s break it down one line at a time.
The lines in the top half of the legend represent the Ti Founders Edition card, with the green line representing GPU diode temperature for the Ti FE. That line, conveniently, sits right at the GTX 1080 Ti’s thermal limit point: 84C. The card tries hard to disallow temperature to increase beyond this point, and does so by limiting the clock and increasing the fan (along a curve/lookup table). The fan speed curve seems to peak around 50% unless under duress, and so clock limiting kicks in to keep thermals under control. This means that we’ll see more sporadic frequency fluctuation when plotting the 1080 Ti FE over time, given its against-the-limits modus operandi.
Under stock conditions, the card will not allow the frequency to boost to a point of exceeding 84C. Temperature rubber bands between 84-85C, dragging frequency in tow. Looking at the GN Hybrid variant, we’re hovering around 45C. This gives us a ~40C reduction in hard temperature vs. FE, from which stems additional thermal headroom for boosting. The big question here is whether removing the vapor chamber and heatsink has negatively impacted the FET cooling in any significant way, as we haven’t previously been able to perform this testing. The card still has a thick aluminum baseplate mounted to the PCB, and we’re still using the VRM blower fan, but modifying the cooling design could impact those temperatures. Airflow is significantly changed with this mod (can’t exactly funnel air through a pump block).
In short, the answer to “does FET temperature increase?” is “no.” The big cluster of lines in the 65-70C range represents the FET2 and FET7 temperatures for both revisions of the Ti card. There is no significant difference. For the FETs we monitored, it is fair and accurate to say that the temperature is effectively equal to the Founders Edition card. The fact that they’re illegibly clustered in one spot mostly establishes this fact. Even if there were an increase, it’d have to be substantial to cause any concern. VRM components can take a serious beating (125-150C in most cases) and, although we want to avoid derating and efficiency loss, adding a few degrees won’t hurt anything. This doesn’t even do that, really; the VRM fan (at 22%) and baseplate appear to be enough to handle the FETs. We’ve reduced fan speed overall (~30% for the blower) and removed the vapor chamber cooler, but have added a CLC. This allows GPU diode temperature to plummet, doesn’t significantly impact FET temperature, and reduces blower fan noise. Granted, we do gain noise in the form of a 120mm radiator fan.
GTX 1080 Ti Hybrid Clock vs. Temperature & Time
Here’s the 1080 Ti FE clock vs. temperature & time chart, as a refresher:
As for the impact on thermals versus clock, the above data allowed us to hypothesize that the clock-rate should be flatter and less sporadic in its fluctuations when under liquid. This conclusion can be formed just by looking at the frequency resultant of our 80% fan speed, where greater performance was siphoned at the cost of increased noise.
Looking at the new frequency vs. time and temperature chart, we see the pink line representing the GN Hybrid 1080 Ti, which sustains a higher clock overall and a flatter line. There’s a lot less fluctuation here. Keep in mind that, as stated previously, this is with a power virus test – we’re mostly burning the FETs here, and so the cores don’t boost as high as in video games. Everything is burned in (and often hotter than a game would do) even though the clock speed is lower.
For thermals, we see the GN Hybrid operating below 50C, generally in the 45C range, with the FE version of the same GPU and PCB at around 65C when at 80% fan speed, or 84C when at auto speeds. Clock fluctuation is no more than 30MHz in the worst cases on the Hybrid card.
3DMark FireStrike Tests
We’re just dumping these charts here, and will follow-up with game benchmarks and analysis of those results:
3DMark FireStrike - Normal
3DMark FireStrike - Extreme
3DMark FireStrike - Ultra
3DMark TimeSpy
