I have several graphics card and use a 3080ti for render offload. In the past year or so, the power consumption reported by nvidia-smi at idle was 7w for the card, and the temperature was usually below 40 degrees.
Drivers were updated on the cuda repo today or yesterday. Since installing the new version (570.86.10) the idle consumption has increased from 7 to 18w and the temperature is now around 50 degrees, which is I think the minimum temperature for the fans to switch on.
There’s absolutely nothing running on the card, at least until I launch a steam game, and I don’t think anything else has changed in my setup.
Fri Jan 24 19:23:23 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.86.10 Driver Version: 570.86.10 CUDA Version: 12.8 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 3080 Ti On | 00000000:01:00.0 Off | N/A |
| 0% 44C P8 18W / 400W | 2MiB / 12288MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| No running processes found |
+-----------------------------------------------------------------------------------------+
it’s not easy to return to the 565 driver to make a test; apt install cuda-drivers-565 actually drags the 570 version of packages anyway.
Is gsync / vrr / fresync enabled by any chance? idk if these features are even available on laptops, but having vrr enabled on my desktop causes much higher power usage on idle
I am on wayland. I have excluded the nvidia card from KWin, there’s no monitor connected to it, and the card is truly unused until I boot up something that uses render offload (e.g. using a Steam game set up to use the nvidia card using env variables).
these are the temperatures on the machine after a couple of minutes, using driver 565
Obviously if I start to run something, the temperatures start to climb up (60/70+ degrees) and then go down to 35 degrees. When using driver 570, the temperature climbs to 49 degrees and does not go down.
I was searching about why there was another driver release today and found this thread… sure enough, I also noticed higher temperatures since I installed last Friday’s (UTC+8) driver and my power consumption was 18W (on a laptop 3070 that’s totally idle because I only use it for compute).
During last update I noticed they changed the modprobe.d/nvidia.conf by removing the line: options nvidia NVreg_DynamicPowerManagement=0x02 but I had put it back while trying to understand why my system was crashing at boot (turned out to be a failure to update the initramdisk, or what it’s called now since I use dracut to manage TPM2 encrypted disk unlock).
I just tried removing that line and rebooting and I see lower power consumption at idle and my temp. at idle is now almost 10°C lower.
Idle speculation on why this, (Increased idle consumption with driver 570), may have occurred.
Driver 570 does not seem to be available through the Nvidia.com drivers page and was released with Cuda Toolkit 12.8, to support Blackwell.
Perhaps the newer, more aggressive power management features of Blackwell, have had the undesired effects you’re seeing on older cards and hopefully will be worked around soon. Details starting on page 31 here.
the temperature increase is not really indicative here, I booted my computer with driver 570 first and checked multiple parameters which lead to the temperature slowly increasing.
I then switched to driver 565 and while the card consumed less power, it was already hotter at that point (that was literally not a cold boot).
If I keep my computer on, idle for several hours, then driver 565 temperature is 37 degrees.
Same with driver 570 gives me a temperature of 49 degrees.
The most reliable difference is just the extra 10 watts.
I did notice the 570 driver is not on the website yet, I am a bit disappointed that people using the debian repository got a new driver pushed that is neither beta nor new version (and not stable either, obviously) the default driver should have stayed at 565 unless people opt-in for beta/new feature drivers
Similar issue here - just upgraded to driver 570, and idle power consumption is higher than normal. Usually around 16W, now tending to sit at 20-21W. This is on a laptop:
Device 0 [NVIDIA GeForce RTX 3080 Ti Laptop GPU] PCIe GEN 1@ 8x RX: 400.0 KiB/s TX: 400.0 KiB/s
GPU 210MHz MEM 405MHz TEMP 47°C FAN N/A% POW 21 / 130 W
Note: The desktop uses the integrated graphics for display. The RTX is used purely for compute. Have not rolled back to 565 yet to confirm.
I have not rolled back… but if I suspend / resume (as per earlier posts), the power consumption jumps up briefly (along with the speed) before settling back down to the “normal” idle 15W.
Guess I’m an odd one out. With a 3080 and 565, my idle usage was ~30 watts. With 570, it’s still ~30 watts. This is from driving 3 monitors(1440p, 2x 1080p, all 120hz). Maybe the difference is only really seen when draw was lower in the first place.