Difference comes from the fact that nVidia measures(and limits) power consumption before VRM losses occur and AMD does that after VRM conversion. There's no 'lying' misinforming and so on, just different ways of doing measurements (there was a video that Buildzoid did on how Texas Instruments INA chip works that nV uses to measure power consumption)
Power consumption for GPUs
- Thread starter DavidGraham
- Start date
AMD is also unified across CPU and GPU. The same SVI protocols that are used for CPU's have been adopted for GPU's for a number of generations now.
DavidGraham
Veteran
No, it's mostly just the Youtube videos with gameplay footages with overlays showing power consumption figures.
Reviewers actually measure things. Most use "from the plug" method aka whole computer consumption (excluding monitor). Those who measure actual card power draw use specialized tools (sitting between card and pcie slot and on the power plugs). Software measurement differences are thus irrelevant for them.
YouTubers like @DavidGraham mentioned are another story though
I see. Where are the good GPU reviews at these days? Ever since Scott Wasson and Ryan Shrout moved on I've been having a tough time finding high quality reviews that I know I can trust.
Everyone has their favourites I suppose. TPU and Tom's (at least sometimes) measure just the card power draw rather than whole rig from the plug. TPU also has easy to read "overall results" per res (and per some features even), ComputerBase has similar easy to use overall performance graph, but the site & articles are in german. Obviously by far best reviews, by leaps and bounds, in different league altogether totally impartially considered are at io-tech.fi but they're in finnish so good luck with that.
DavidGraham
Veteran
That's not an actual 464w, that's the reading of the software overlay, those don't read AMD GPUs accurately. You need to add 30w to 50w on top of them to account for the full consumption of the card.
Nebuchadnezzar
Legend
This is not a blanket rule for all AMD cards. The software reading is accurate on these cards.
Do you have actual evidence for this for the RDNA3 cards, or are you just pushing your usual agenda?
DavidGraham
Veteran
That's not consistent with Igor's findings.
AMD Radeon RX 7900XTX and RX 7900XT Review - One giant step forward and one step sideways | Page 10 | igor´sLAB
The time has finally come and after a lot of theory and unboxing in the run-up, we can finally test what the new Radeon RX 7900XTX and RX 7900XT can really do and where there might even be further…
www.igorslab.de
That's not consistent with Igor's findings.
AMD Radeon RX 7900XTX and RX 7900XT Review - One giant step forward and one step sideways | Page 10 | igor´sLAB
The time has finally come and after a lot of theory and unboxing in the run-up, we can finally test what the new Radeon RX 7900XTX and RX 7900XT can really do and where there might even be further…
I don't see anything there that states it's tracking the die only? To the contrary, this would imply the opposite:
The 7900XTX is limited to max 375W (pci-e slot 75W+2*150W 8pin power extensions) or it would be a serious system hazard.
DavidGraham
Veteran
This doesn't:
Then he links again to the article discussing that software doesn't read the entire consumption of AMD cards.
In short, he is implying that status quo hasn't changed. AMD also seems to have indicated as much.
Last edited:
The original german version of the article is pretty clear about this. The new telemetry option is for whole card power consumption, that's why it's "limiting" the effect a stress test like furmark can have.
Maybe it helps for a better understanding to read the corresponding paragraph in his summary: "AMD has completely turned telemetry on its head, which has worked out well in most respects. The fact that you can finally read out a TBP that is reasonably accurate, even if it is only a good estimate, is a big step forward. NVIDIA has long relied on real monitoring of the rails via shunts, while AMD now at least uses the summation of all values from the DCR and some mathematics, which is also possible."
DavidGraham
Veteran
Oh, I see.
Hmmm, review articles already give overclocked power figures that are higher than these Youtube videos with software overlays, which means these overlays are probably still not accurate enough.
The old "TGP" value is still there and lower than TBP. You have to update your readout-software (to make use of the new "sensor") AND your overlay config (to actually display the new value instead of TGP) in order to see the actually correct results. Else, you'll still have the old TGP, which has the issues mentioned here.
DavidGraham
Veteran
Sigh, which gets us back right to where we started, these Youtube videos are not reliable for power consumption figures, because the person making the video has to actually use the latest version of the software, and has to actually "manually" caliberate it to show readings from the new semi accurate sensor. Otherwise, his readings will be false.
Nebuchadnezzar
Legend
Uh no, it's not. It goes to 417W just fine, with 55W of that on the slot. It's limited beyond that on the reference cards because of the software lock that we can't change the PL beyond 115%.
