Wrapping up our performance testing, the power and thermal results are actually more interesting than the raw performance. At least there's more variation between the cards in these areas, though all three custom cards stay reasonably cool and should run fine in any PC large enough to accommodate the hardware.
For these tests, we're using Powenetics software to gather the actual power draw of the graphics cards. GPU-Z is used to collect thermal, clock speed, and fan speed data. We also have an SPL meter that we set up 15cm from the side of the cards to capture noise. However, we don't have an anechoic chamber or anything particularly fancy. External noise (e.g., from traffic) means we have to simply eyeball the meter rather than logging data and generating a chart. We test all of these metrics using Metro Exodus, set to loop five times, and FurMark running at 1600x900 in stress test mode.
Let's quickly talk about the noise levels first. Recent changes to my office mean the new setup isn't identical to the old one, but the noise floor (when traffic isn't driving by) is now 34 dB. The SPL meter is kept close to the GPUs in order to isolate the noise from the card and not pick up as much fan noise from the CPU cooler, though differences in card design can be a factor. All of the latest GPUs support 0 dB fan technology, which means idle noise levels are all the same: 34 dB. Literally any other noise, like typing or someone walking around, registers at much higher levels than that, so it's pretty quiet.
Under load, the GPUs start to show their differences. For example, the Asus has two large 120mm fans on the radiator that move quite a bit of air, while the third fan on the card itself is basically silent. Asus ended up being the loudest of the three GPUs in the gaming test, measuring 45.6 dB peak and generally hovering in the 43-45 dB range, but the actual fan noise tends to be lower and, to my ears at least, not as noticeable as smaller fans. ASRock peaked at 44.5 dB in gaming, again with a range of around 41-44 dB. Sapphire showed the most variation, running as quiet as 37 dB but peaking at 42.1 dB.
Results under FurMark are more consistent, but the cards deal with the extreme power draw in different ways. As a result, ASRock was the loudest card in FurMark, peaking at 45.6 dB and basically staying in the 45+ dB range. Asus came in second at 44.1 dB, while the Sapphire card was the quietest of the three and only ran at 39.5 dB. Again, we didn't make charts because it's just the three cards, but none of the cards are particularly loud. Pay attention to the actual GPU clocks and thermals below when looking at noise levels, however, as all of these metrics are interrelated.
As mentioned earlier, the three custom AIB cards all have higher power use than the reference 6800 XT. In the Metro test, the ASRock uses the least power at around 332W, 30W more than the reference model. The Asus card uses 341W, though it has a pump and two larger fans on the radiator that undoubtedly use more power than the three smaller fans used on the other two cards. The Sapphire card ends up with the highest power use of 344W, which is as much as some of the other cards use even when overclocked. Overclocking isn't even that bad when you consider that performance improves by 4-7 percent (in Metro), while power use increases 1-6 percent. (Except for the reference card, which used 13 percent more power.)
FurMark takes power use to even higher values, ranging from 352W for the ASRock to 358W on the Asus and Sapphire cards. Overclocking pushes the three cards into the 400W and higher range, again with Sapphire using the most power of the 6800 XT models.
At factory stock settings, all of the RX 6800 XT cards meet or exceed the official boost clock, though in-game clocks will vary by game. In Metro, the custom cards reach 2386-2394 MHz average clocks, with overclocking pushing all three cards into the 2500 MHz and higher range. FurMark is a different story, with the GPUs reigning in clocks in order to keep thermals and power use in check, but 2.1GHz while running FurMark is still quite impressive, and overclocking takes that up to around 2.25GHz.
Temperatures have a direct correlation with fan speed and noise, so they can't be considered on their own. As noted earlier, the Asus card had the highest noise levels, and while the fan speeds are quite low, two large 120mm fans still move a lot of air and can create more noise and turbulence than three 90mm fans. Still, the Asus runs over 15C cooler than the other custom cards in Metro, and over 17C cooler in FurMark, all while hitting similar clocks. Somewhat surprisingly, the Sapphire card has lower fan speeds than the ASRock but tends to be just as loud while running a bit hotter. Still, 72-75C isn't particularly high for a high-end GPU playing a demanding game.
You can also see how the significant ramp in fan speed to help with overclocking affects things. Temperatures are actually lower than stock, but that's because overclocking puts enough of a strain on the GPU that we didn't want to risk instability in pursuit of lower noise levels. If you want low thermals with a decent overclock, the Asus liquid cooler ends up being far superior to traditional air cooling.
MORE: Best Graphics Cards
MORE: All Graphics Content