Unable to communicate with GPU because it is insufficiently powered.

Hello, I use two Titan Xps on my ubuntu (16.04.5)

I am trying to use tensorflow-gpu on my machine for further experiment.
So, I was using my machine and trying to figure out if my graphics card is working.

However, it seems that when I check the command

nvidia-smi

Unable to determine the device handle for GPU 0000:B3:00.0: Unable to communicate with GPU because it is insufficiently powered.
This may be because not all required external power cables are
attached, or the attached cables are not seated properly.

I get those as result.

I have googled a bit, and found a similar case
https://devtalk.nvidia.com/default/topic/946729/cuda-setup-and-installation/tesla-k40-nvidia-smi-error-can-acute-t-communicate-with-gpu/

I will post the results from those commands

lspci | grep -i nv

lsmod | grep -i nouveau

dmesg | grep NVRM

lspci | grep -I nv
65:00.0 VGA compatible controller: NVIDIA Corporation Device 1b02 (rev a1)
65:00.1 Audio device: NVIDIA Corporation Device 10ef (rev a1)
b3:00.0 VGA compatible controller: NVIDIA Corporation Device 1b02 (rev a1)
b3:00.1 Audio device: NVIDIA Corporation Device 10ef (rev a1)

lsmod | grep -i nouveau
(no result)

[ 1.456313] NVRM: loading NVIDIA UNIX x86_64 Kernel Module 418.40.04 Fri Mar 15 00:59:12 CDT 2019
[ 5.307707] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 5.307737] NVRM: rm_init_adapter failed for device bearing minor number 1
[ 5.566479] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 5.566533] NVRM: rm_init_adapter failed for device bearing minor number 1
[ 5.900174] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 5.900327] NVRM: rm_init_adapter failed for device bearing minor number 1
[ 33.642157] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 33.642203] NVRM: rm_init_adapter failed for device bearing minor number 1
[ 33.886988] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 33.887021] NVRM: rm_init_adapter failed for device bearing minor number 1
[ 91.710171] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 91.710219] NVRM: rm_init_adapter failed for device bearing minor number 1
[ 91.967265] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 91.967296] NVRM: rm_init_adapter failed for device bearing minor number 1
[ 487.274869] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 487.274908] NVRM: rm_init_adapter failed for device bearing minor number 1
[ 487.522474] NVRM: RmInitAdapter failed! (0x25:0x1c:1084)
[ 487.522518] NVRM: rm_init_adapter failed for device bearing minor number 1

thanks for reading.

This usually indicates that the GPU needs extra power cables plugged into it, and that the GPU is missing this. When it was installed, not all the cables needed were plugged into the GPU.

It’s not a software issue. You have to open up the PC case and look at how the GPUs were installed.

I’m having the same problem. I started with a GeForce GTX 1080 Ti and then just installed a second card - the RTX 2080 Ti.

I connected the power to the two 6-pin power cables that were available and unused on my machine.

I’m running Ubuntu 18.04 – some sample command outputs:

$ nvidia-smi
Unable to determine the device handle for GPU 0000:02:00.0: Unable to communicate with GPU because it is insufficiently powered.
This may be because not all required external power cables are
attached, or the attached cables are not seated properly.

$ lspci | grep -i nv
01:00.0 VGA compatible controller: NVIDIA Corporation GP102 [GeForce GTX 1080 Ti] (rev a1)
01:00.1 Audio device: NVIDIA Corporation GP102 HDMI Audio Controller (rev a1)
02:00.0 VGA compatible controller: NVIDIA Corporation Device 1e04 (rev a1)
02:00.1 Audio device: NVIDIA Corporation Device 10f7 (rev a1)
02:00.2 USB controller: NVIDIA Corporation Device 1ad6 (rev a1)
02:00.3 Serial bus controller [0c80]: NVIDIA Corporation Device 1ad7 (rev a1)
03:00.0 Non-Volatile memory controller: Samsung Electronics Co Ltd NVMe SSD Controller SM961/PM961

Any help would be appreciated.

The RTX 2080 Ti requires two 8-pin aux power connectors:

https://www.nvidia.com/content/geforce-gtx/GEFORCE_RTX_2080Ti_User_Guide.pdf

Directly plugging in 6-pin cables won’t work, as you’ve already discovered.

Yup - my mistake. Who would’ve thought that I could plug a 6 pin into an 8 pin without any problems. d’oh!
But the card did come with two 6-to-8 pin adaptors. I used those and still get the error.

I spoke with tech-support at PNY (that card’s manufacturer) – they asked what power supply i have. (850 Watt Multi-GPU Approved Power Supply with High Perf – from Dell/Alienware) – and was told that that that power supply cannot power two eight pin power connectors.

This is probably out of scope for this forum - but does that sound right? Dell assured me that it would easily support two graphic cards…

david,

The RTX install guide I linked suggests minimum 650W PSU for a single RTX. There would be a higher requirement for RTX2080Ti + 1080 Ti. I don’t know what it would be but a first order approximation would be to take 650W and add the max power draw of 1080 Ti to it. That suggests to me that 850 W PSU might be “marginal”.

https://www.geforce.com/hardware/desktop-gpus/geforce-gtx-1080-ti/specifications

However this isnt really how the PSU detection scheme works - the thing that is triggering this warning. It basically looks for appropriate supply voltages on all the appropriate pins for each connector. If you meet that requirement, this detection scheme doesn’t really know whether it is a 650W PSU, or an 850W PSU, or a 1500W PSU.

I’ve generally had “good luck” with the 6-pin to 8-pin dongles, but it’s really not the best way to configure a system. For best stability you want a PSU that natively has the required power connectors/cables.

It’s difficult to say exactly what is happening in your case. Plugging in the 6-pin cables into 8pin connectors on your GPU I would not expect to work - you would get exactly the error you are reporting.

But I can’t explain why the use of the 6-pin to 8-pin dongles isn’t getting you past this. If it were me, I would try that experiment again, carefully. There exist different 6-pin/8-pin connectorization schemes (so called ATX and PCIE) but I don’t believe that is the issue here, because the older ATX style connectors won’t physically mate with the newer PCIE aux connectors.

Whether or not your PSU can support 2 GPUs and of what type I really can’t say. For best stability, you should use a power supply that has sufficient rating (and 850W seems marginal to me) and it should natively have the cables needed to directly cover the connectors on your GPU(s) without the use of any conversion dongles.

Robert,

Thank you for that very helpful reply.

One thought – would it be possible for me to bypass the question of my computer’s power supply and dongle conversion – and instead purchase a separate, external power supply that’s dedicated to the RTX?

david,

I wouldn’t recommend it. The slot will still be powered from the main PSU, and so you have the possibility of cross-connected PSUs. In order for that to work correctly, there would be specific design requirements placed on the RTX card, and I know of no statement by NVIDIA that that is supported. It would void the warranty on the card.

How much more powerful is the 2080 from my 1080? If my machine isn’t easily expandable (without upgrading the power supply), I could try removing the 1080 and just using the 2080…

david,