CUDA on optimus graphics with nvidia's repo

Latest response

After installing cuda the login screen does not show, how to fix it?

How can I install it in a better way?

Responses

Hi Kwong,

Uninstall all NVIDIA software you have installed and reboot the operating system - when the login screen appears press Ctrl + Alt + F3.
Add RPM Fusion repo to the software sources and install the NVIDIA drivers (https://rpmfusion.org/Howto/NVIDIA) and NVIDIA CUDA (https://rpmfusion.org/Howto/CUDA) - once finished reboot the operating system again and now everything should work as expected.
Additional information : wait a few minutes before reboot, though the installation command seems to have finished, it takes some time.

Cheers :)
Christian

That one seems to be for Fedora.

I finally used the runfile method with --no-opengl-libs flag and it works very well but I don't really know how to check for updates.

No, it's not only for fedora, you can use it on RHEL as well : https://rpmfusion.org/Configuration ... and you won't get updates when you don't add a repository - you have to reinstall it every time when a new drivers version gets released, because you installed the NVIDIA software manually by using the "run file method".

"yum install cuda" which gives me the driver form nvidia's repo is the breaking my GUI, although it is CUDA 8.0 but not 7.8.

Actually how is RPMFusion driver better than the one from the CUDA repo?

Well, I'm not talking about the drivers, they are the same. I'm talking about the installation method : it is a known fact that installing the drivers with the NVIDIA run file causes a whole lot of trouble in many cases.

"yum install cuda" gives me the opengl libraries and not working with optimus graphics.

I also tried the method provided in "Advanced Setup" but it still gives me the GL files as dependency and breaks my GUI. Will it be the same for installing the package from RPMFusion? http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#advanced-setup

I am going to stick with Intel graphics for rendering display anyway. I use NVIDIA merely for offscreen rendering.

All I can say is that on my machine which has NVIDIA graphics with NVIDIA Optimus, installing the NVIDIA drivers using the RPM Fusion repository went smoothly without any problems ... maybe just give it a try. :)

After you install, which graphics does it show in Settings -> Details (the page you see the version number of RHEL)?

If you use hashcat, does your screen lag when you run benchmark with -w 4?

I installed the drivers for testing purposes, I am using the open source nouveau drivers.
It showed the NVIDIA chip version in System Settings (Details tab) and the output from lspci -k | grep -EA2 'VGA|3D' showed : kernel driver in use: nvidia.
In the NVIDIA Settings application the used drivers version was listed correctly too.

Are you using a MUXed Optimus graphics and disabling the Intel one or an older kind of switchable graphics? My laptop always use Intel GPU for rendering the display.

I have a NVIDIA GEFORCE GTX 860M graphics adapter in a CLEVO barebone based SCHENKER XMG A704 notebook and use the nouveau drivers right because they support NVIDIA Optimus out-of-the-box (a big advantage). I don't need the proprietary NVIDIA drivers because I do not run any applications which need higher graphics performance. Learn more about all the differences on negativo17.org ... I hope it helps. :)

I am using Lenovo T470p (but not 20J6Z320US on red hat certification) with GT 940MX and it is not recognized by nouveau showing unknown chipset on boot and I need CUDA anyway.

However, I think it should be similar. https://support.lenovo.com/hk/en/solutions/pd104609 http://psref.lenovo.com/syspool/Sys/PDF/ThinkPad/ThinkPad%20T470p/ThinkPad_T470p_Platform_Specifications.pdf

I just tested on another laptop with Optimus graphics and CentOS.

I install the cuda xorg subpackage from rpmfusion, and then cuda 8.0 from nvidia repo, then run hashcat, showing CL platform not found. After that, I install the nvidia driver following the instruction on RPMFusion with --disablerepo=cuda flag passwd to yum, then reboot, not even able to boot properly. Fortunately I didn't try on my Thinkpad.

Close

Welcome! Check out the Getting Started with Red Hat page for quick tours and guides for common tasks.