0

I just want to get started with dual booting Ubuntu and Windows. I have a Dell XPS 9570 (need NVIDIA drivers too and CUDA and TensorFlow installed) and there are bits of information I can find online but since I am new for my first time I would require an entire step by step guide (specific to dell xps 9570) that explains all steps, commands, drivers, issues that may be faced and how to fix them. That guide would also need to include how to install CUDA and TensorFlow along with NVIDIA GPU drivers and Intel GPU drivers.
Thank you in advance guys.

BeastSolos
  • 1

1 Answers1

0

The details are here in many answered questions or the help section. To summarize, since many answers are old:

  1. Pick a compatible set of software e.g. CUDA 10.1, DNN 7.6.5, Tensorflow 2.1.
  2. Install Ubuntu, using the "nomodeset" word edited onto the kernel boot line until you have the latest offered proprietary Nvidia drivers installed.
  3. Install the proprietary Nvidia drivers. No need for any PPAs for graphics drivers, the ones in the standard repositories will do nicely. Probably the latest one is 435 or 440, depending upon the kernel.
  4. Pick a CUDA installation method WITHOUT any old Nvidia drivers. Avoid the .deb packages with the Nvidia driver number in their name. There is a list of debs, pick one of the top level cuda ones, the other are brought along from that one. Normally, I'd recommend avoiding the .run file installation method, but I think that actually allows you to skip the unwanted Nvidia drivers. The real problem with the old drivers, all the CUDA files depend upon them, they (partially) delete the current drivers, and may not be fully configured, causing problems every kernel update, and maybe even subsequent logins. If you update drivers again, you will delete all the old ones, and leave all the CUDA files subject to autoremove.
  5. Install any old gcc and g++ the CUDA versions requires. Simply add links to these (gcc, g++ ar, nm, ranlib) to the ...cuda/bin. This should be before all the system bin locations in your PATH, as per standard instructions, so they will be used.
  6. Make the samples, and add any missing libraries necessary.
  7. Install the Intel Deep Neural Networking DNN packages. I just put them into the existing cuda locations (cuda/bin cuda/lib64 cuda/include). Again, make the samples, and add any missing libraries.
  8. Tensorflow has it's own web site with installation instructions. At least with the python/pip installs, when a requested (old) version of a file/package is not available, later ones will be suggested.

All documented, but old instructions typically have unneeded complications -- it's gotten simpler with time.

Those links seem reasonable, I have no direct experience with MOK utils. Definitely check that your BIOS/UEFI settings are the latest. The inst. I used were https://medium.com/@Oysiyl/install-tensorflow-2-with-gpu-support-on-ubuntu-19-10-f502ae85593c Only had to update the python3 -m pip install ...tensorflow==2.1.0rc0 to tensorflow==2.2.0rc2 (latest choice given when the original failed on an Ubutnu 20.04B install).

ubfan1
  • 19,049