So you want to take advantage of your GPU to do some local LLM stuff or some brute force cracking, but you're a man of culture. You're on Linux, and you've already installed Nvidia proprietary drivers so that you can do hardware encoding and other stuff like that.
So when you go to the website and follow their install instructions, you run a bunch of commands they provide, execute some bullshit, and in the middle of the install it crashes, dependencies are conflicting, you can't --fix-broken, you need to purge everything (even though you already did following their bad instructions).
FUCK!
FUCK YOU NVIDIA!
Purge again I guess. Maybe you need a more aggressive strategy?
dpkg -l | grep nvidia | awk '{print $2}' \
| xargs sudo dpkg --force-all -P
sudo apt --fix-broken install
sudo apt remove --autoremove --purge \
"*cuda*" "*cublas*" "*cufft*" "*cufile*" \
"*curand*" "*cusolver*" "*cusparse*" "*gds-tools*" \
"*npp*" "*nvjpeg*" "nsight*" "*nvvm*" "*nvptx*"
NOW WHATEVER YOU DO
DO NOT TYPE THIS COMMAND
sudo ubuntu-drivers autoinstall
sudo ubuntu-drivers | grep autoinstall
autoinstall Deprecated, please use "install" instead
because
if you do it is going to install a new kernel and build some kernel modules and after you reboot your display, network, and audio drivers will all be fucked and you'll have to load up GRUB and boot an older kernel and come back in and purge the Nvidia and CUDA stuff (again).
Use the Install Nvidia Drivers (Ubuntu Docs) guide instead (if you're on Ubuntu). Look at the log file as per that guide, get the correct version.
sudo ubuntu-drivers install nvidia:580
Nvidia will not be able to get it right for you. By the grace of God alone will touching anything from Nvidia spare your computer.
nvidia-smi
Thu Feb 5 23:06:43 2026
+-------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.126.09 Driver Version: 580.126.09 CUDA Version: 13.0 |
+-------------------------------------+------------------------+----------------------+
So that's awesome, however...
hashcat -I
hashcat (v6.2.6) starting in backend information mode
Successfully initialized the NVIDIA main driver CUDA runtime library.
Failed to initialize NVIDIA RTC library.
* Device #1: CUDA SDK Toolkit not installed or incorrectly installed.
CUDA SDK Toolkit required for proper device support and utilization.
Falling back to OpenCL runtime.
We are full circle back to this guide
NOW WHATEVER YOU DO
DO NOT USE THIS NVIDIA INSTALLER OR THIS GUIDE
We're going to try something provided by the Linux distribution, which you should also do on your end.
sudo apt install nvidia-cuda-toolkit
Let us pray...
hashcat -I
hashcat (v6.2.6) starting in backend information mode
CUDA Info:
==========
CUDA.Version.: 13.0
Backend Device ID #1 (Alias: #2)
Name...........: NVIDIA GeForce GTX 1060 6GB
Processor(s)...: 10
Clock..........: 1784
Memory.Total...: 6057 MB
Memory.Free....: 5624 MB
Local.Memory...: 48 KB
PCI.Addr.BDFe..: 0000:26:00.0
OpenCL Info:
============
OpenCL Platform ID #1
Vendor..: NVIDIA Corporation
Name....: NVIDIA CUDA
Version.: OpenCL 3.0 CUDA 13.0.97
Backend Device ID #2 (Alias: #1)
Type...........: GPU
Vendor.ID......: 32
Vendor.........: NVIDIA Corporation
Name...........: NVIDIA GeForce GTX 1060 6GB
Version........: OpenCL 3.0 CUDA
Processor(s)...: 10
Clock..........: 1784
Memory.Total...: 6057 MB (limited to 1514 MB allocatable in one block)
Memory.Free....: 5568 MB
Local.Memory...: 48 KB
OpenCL.Version.: OpenCL C 1.2
Driver.Version.: 580.126.09
PCI.Addr.BDF...: 26:00.0