Error: CUDA driver version is insufficient for CUDA runtime version

I just installed CUDA 10.0 and NVIDIA-390.87, and the driver version is the latest for my GTX 1060.
And when I was installing CUDA, I saw a notification that said it supports nvidia-390.xx. Why this error happened while I was testing whether cudnn could work.

CUDA 10.0 requires a 410.48 or higher driver. See table 1 here:

https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#major-components

If you install CUDA 10 properly, you will get a proper driver for your GTX 1060.

Use a runfile install method or a package manager install method, and get your installers from http://www.nvidia.com/getcuda

Follow the instructions in the linux install guide carefully:

https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html

When I went to the official site trying to download a nvidia driver, it only offer nvidia-390.87 for my GPU info.
Anyway, what I need is just installing a 410.48+ driver, right?
Thank you for quick reply.

Yes, you need a 410.48 or higher driver for CUDA 10

If you install CUDA 10 properly, you will get such a driver.

If you are just looking for a driver installer on the nvidia website (e.g. http://www.nvidia.com/drivers ) you need to select BETA drivers to see the latest. It is beta because CUDA 10 and the associated drivers are still pretty new. If you do that, you will be offered this driver, for example:

https://www.nvidia.com/download/driverResults.aspx/137935/en-us

which will also work. However that is a driver runfile installer, so it will only be convenient to use it if you installed CUDA via runfile install method.

I tried installing the 410.48 and 410.57 driver, but they could not work properly when I logged into my desktop. Only when I reinstalled a 390+ driver, it became normal.
My operation system is Ubuntu 18.04 LTS. Does this system support the latest driver?

Yes.

what is the min driver version to run cuda 10 on Ubuntu 16.04?

deviceQuery gives me the below error, but I already have 410.79 installed.

fangq@taote:/local_mount/space/taote/1/users/fangq/git/Temp/deviceQuery$ ./deviceQuery 
./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 35
-> CUDA driver version is insufficient for CUDA runtime version
Result = FAIL


fangq@taote:/local_mount/space/taote/1/users/fangq/git/Temp/deviceQuery$ nvidia-smi
Thu Feb 21 15:08:32 2019       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.79       Driver Version: 410.79       CUDA Version: N/A      |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  TITAN V             Off  | 00000000:01:00.0 Off |                  N/A |
| 34%   47C    P8    N/A /  N/A |      0MiB / 12036MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   1  TITAN V             Off  | 00000000:02:00.0 Off |                  N/A |
| 33%   46C    P8    N/A /  N/A |      0MiB / 12036MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

The minimum driver version for CUDA 10.0.130 on any linux platform is 410.48. 410.79 should work fine. That doesn’t mean that 410.48 will install correctly on any future linux platform, of course. But it should be usable with Ubuntu up through 18.04

sometimes the only thing needed is a reboot

If you’ve done that already, and the problem persists, then the driver install is incorrect.

Without knowing how you installed the driver, it’s hard to say exactly what is wrong, but almost certainly there were multiple attempts to install a driver on this machine or a history of a previous driver on this machine. Some driver component is left over from that.

The solution is to properly clean out the old driver. The exact methodology will depend on how you installed the 410.79 driver as well as how you isntalled any previous drivers or install attempts.

The CUDA linux install guide gives some tips for the cleanup process in the section “handling conflicting install methods”

Another possible source of error is if you installed the 410.79 driver from some non-NVIDIA resource, such as PPA. If you installed a NVIDIA GPU driver from a 3rd party resource such as PPA, it may not work correctly with CUDA.

The problem was eventually resolved by installing 415.27 from a ppa. Somehow, I just could not get 410.79 to work, despite cuda 10 is supposed to support anything above 410.48.

the 410.79 package was actually from nvidia resources

http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/

Hi,
I am seeing same error, when I am running sample_hello_world
and here is the output of nvidia-smi

±----------------------------------------------------------------------------+
| NVIDIA-SMI 435.21 Driver Version: 435.21 CUDA Version: 10.1 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Quadro GP100 Off | 00000000:9E:00.0 On | Off |
| 26% 39C P0 25W / 235W | 439MiB / 16277MiB | 1% Default |
±------------------------------±---------------------±---------------------+

±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 2900 G /usr/lib/xorg/Xorg 177MiB |
| 0 15539 G …AAAAAAAAAAAAAAgAAAAAAAAA --shared-files 150MiB |
| 0 16861 G /usr/bin/gnome-shell 108MiB |
±----------------------------------------------------------------------------+

Hi,
After updating drive version, hello world sample runs fine but other sample doesn’t want to work and generating errors
No protocol specified
WindowGLFW: Failed initialize GLFW
terminate called after throwing an instance of ‘std::exception’
what(): std::exception
Aborted (core dumped)
Thanks,