Note that the article says they measured it with a hardware meter. The article describes it as a "Nvidia PCAT v2 device". That could account for some differences, perhaps.
@JarredWaltonGPU , do you have a sense of how the software self-reported power compares with your measurements, on your specific GPUs?
Succinctly: The software lies. LOL. But it really depends on what you're looking at, and it can "lie" in the wrong way as well. I do think that the monitor can impact power use still. 3440x1440 @ 144Hz is fewer pixels and thus less data than 3840x2160 @ 144Hz. To be precise, 4K has 67% more pixels than ultrawide 1440p. I don't know precisely how power use scales, but as noted, 4K @ 60Hz significantly dropped power use (by around 35%). That's less than half the data transmitted, though.
Anyway, I just checked using the reference 7900 XTX, and there are a few different options for power that I have readily available. MSI Afterburner clearly isn't getting the correct data, as at one point it showed a steady 35W for a minute or so. The PCAT readings were bouncing between ~30W and ~45W, literally changing every sampling period (I'd guess it updates the display at least 10X per second, possible 20X). Using AMD's drivers, the performance tab showed around 42~45W for Total Board Power.
So of those options, AMD's drivers and TBP were slightly higher than what the PCAT was reading, while Afterburner appeared to only show the GPU power. Older (pre-RDNA3) AMD GPUs I'm pretty sure only report GPU power and don't show TBP in the drivers (I'd need to swap GPUs to verify for certain, though). Note also that all of these required having windows open and visible that were updating portions of the screen, so all of these resulted in higher than "idle" power draw on my system.
Incidentally, bit_user, I have also collected Intel Arc GPU power draw, again at 4K 144Hz with the same monitor. And yeah, it's still not great:
pcat-idle Arc A380,18.21 <-- Gunnir
pcat-idle Arc A580,41.90 <-- Sparkle
pcat-idle Arc A750,35.62 <-- Intel
pcat-idle Arc A770 8GB,40.60 <-- ASRock
pcat-idle Arc A770 16GB,44.66<-- Intel
Sort of funny that the A580 used more power than the A750 and A770 8GB. Not sure if that's Sparkle's fault or something else (probably Sparkle is to blame). I have to go through a bunch more GPUs (AMD 6000-series first, haven't even started checking all the Nvidia GPUs for idle power), but I do plan to add this to reviews in the future.
Speaking of which: What do you (any of you) think is best for showing idle power use? "Full idle" (at the desktop, no active windows visible, which is what I have right now), or would you prefer a light workload like having PCAT's real-time data visible? Or I could launch a web browser and open Tom's Hardware. I've also thought about doing a fullscreen YouTube playback as another data point, though of course that doubles the amount of time per GPU. 🤷♂️
Facebook
Twitter
Instagram