Have a Gigabyte 2070 Super with some bonkers power draw readings. Card runs fine, other than some GPU throttling because of the power numbers. But GPU power always reads too high, more than the power reading for the entire card. And VRAM power always reads way too low, along with PCIe and 6-pin power readings. The 8-pin reading seems plausible.
Typical readout while running 720p Furmark, 8x AA:
GPU clock: 1590 MHz
Mem clock: 1750 MHz
GPU temp: 58.8 °C
Hot spot: 71.8 °C
Fan speed: 33% / 1000 rpm
Mem used: 454 MB
GPU load: 97%
Mem controller load: 80%
Video engine load: 0%
Bus interface load: 1%
Board power draw: 143W
GPU chip power draw: 220W
MVDDC power draw: 3.1W
MVDDQ power draw: 5.3W
PCIe slot power: 15.7W
PCIe slot voltage: 12.2V
6-pin power: 16.9W
6-pin voltage: 12.1V
8-pin power: 115.3W
8-pin voltage: 12.1V
USB-C power draw: 3.8W (there is no USB-C on this card)
Power consumption: 67.2% TDP
PerfCap reason: Pwr
GPU voltage: 0.79V
Not too worried about the USB-C power, Turing cards often have spurious USB-C readings.
Have changed the 45491 chip, no difference. Reading the shunt resistors manually, it appears that the 45491 is interpreting the data correctly.
Measured the 3x 12V input shunt resistors (for PCIe, 6-pin, and 8-pin), and all measure 0.005 ohms like they should be.
Have done some testing on the current sense lines of the GPU's uP9512P, and all 7 phases look uniform. All phases are firing properly (obviously, or the card wouldn't work under load).
So I'm running out of ideas and commonalities.
How are the power consumptions of the GPU and VRAM measured on these cards? Is this calculated by the respective uP9512P controllers, or is it measured by a separate set of shunt resistors that I'm missing? It would be highly improbable for both 9512s to go bad at the same time.
Typical readout while running 720p Furmark, 8x AA:
GPU clock: 1590 MHz
Mem clock: 1750 MHz
GPU temp: 58.8 °C
Hot spot: 71.8 °C
Fan speed: 33% / 1000 rpm
Mem used: 454 MB
GPU load: 97%
Mem controller load: 80%
Video engine load: 0%
Bus interface load: 1%
Board power draw: 143W
GPU chip power draw: 220W
MVDDC power draw: 3.1W
MVDDQ power draw: 5.3W
PCIe slot power: 15.7W
PCIe slot voltage: 12.2V
6-pin power: 16.9W
6-pin voltage: 12.1V
8-pin power: 115.3W
8-pin voltage: 12.1V
USB-C power draw: 3.8W (there is no USB-C on this card)
Power consumption: 67.2% TDP
PerfCap reason: Pwr
GPU voltage: 0.79V
Not too worried about the USB-C power, Turing cards often have spurious USB-C readings.
Have changed the 45491 chip, no difference. Reading the shunt resistors manually, it appears that the 45491 is interpreting the data correctly.
Measured the 3x 12V input shunt resistors (for PCIe, 6-pin, and 8-pin), and all measure 0.005 ohms like they should be.
Have done some testing on the current sense lines of the GPU's uP9512P, and all 7 phases look uniform. All phases are firing properly (obviously, or the card wouldn't work under load).
So I'm running out of ideas and commonalities.
How are the power consumptions of the GPU and VRAM measured on these cards? Is this calculated by the respective uP9512P controllers, or is it measured by a separate set of shunt resistors that I'm missing? It would be highly improbable for both 9512s to go bad at the same time.
Comment