NVIDIA/Intel GPU VRAM is not detected

[jeevithakannan2]

What module does this affect?
GPU

What hardware/software isn't being detected right?
GPU's VRAM is not detected and displayed as 0KB. I have a hybrid laptop Intel iGPU (Alder Lake-P GT2 (Iris Xe Graphics)) and NVIDIA dGPU (NVIDIA GeForce RTX 3050 Ti Laptop GPU)

Does detection work a little, or is it completely broken?
GPU is detected and other GPU module format placeholders are working. Only the VRAM is not detected. Didn't also work with default config generated by crabfetch -g

[LivacoNew]

This is an issue I've been aware of for quite a while, however still don't have an NVIDIA GPU to play about with this. If anyone's willing to make a PR or give some guidance as to where to find this info it would be appreciated

[jeevithakannan2]

Intel GPU VRAM is also not fetched

[LivacoNew]

Also expected, same issue.
What's happening is that the mem_info_vram_total file I use to grab vram in the module is only set on AMD GPU's. I didn't know this at the time, and don't have an Intel nor NVIDIA GPU to find where to replace this information with, so again we're kinda stuck unless someone else wants to take a look around their filesystems/syscalls to find a source.

[jeevithakannan2]

Also expected, same issue.
What's happening is that the mem_info_vram_total file I use to grab vram in the module is only set on AMD GPU's. I didn't know this at the time, and don't have an Intel nor NVIDIA GPU to find where to replace this information with, so again we're kinda stuck unless someone else wants to take a look around their filesystems/syscalls to find a source.

Ok I will take a look when I have time

[jeevithakannan2]

Ok I surfed online and found out that NVIDIA proprietary driver does not expose the VRAM in the sysfs files. We can use the nvml-wrapper crate but it will create an overhead. The test results after adding nvm_wrapper:

hyperfine ./crabfetch crabfetch -N -w 50
Benchmark 1: ./crabfetch
  Time (mean ± σ):      25.1 ms ±   6.6 ms    [User: 5.7 ms, System: 13.1 ms]
  Range (min … max):    18.7 ms …  45.4 ms    100 runs
 
Benchmark 2: crabfetch
  Time (mean ± σ):       7.2 ms ±   0.9 ms    [User: 4.7 ms, System: 2.1 ms]
  Range (min … max):     6.0 ms …   9.6 ms    444 runs
 
Summary
  crabfetch ran
    3.49 ± 1.02 times faster than ./crabfetch

If the user have nouveau driver there would be no problem in fetching the VRAM info

[jeevithakannan2]

As for intel GPU it's a integrated graphics card so I cannot check from my part.

[LivacoNew]

If there's no sysfs stuff then I guess only one way. Only other thing I can think of is querying a Graphics API like Vulkan or OpenGL but I'm not sure which ones would be faster if any.

If anyone's willing to make a PR with the NVIDIA wrapper I can merge it, just remember to validate the vendor ID n shit. As for Intel, it'll have to be someone else.

[jeevithakannan2]

I can create a PR with nvidia wrapper but it will slow down crabfetch by a large margin as you can see #46 (comment) let me know your thoughts

[LivacoNew]

Again if there's no other way it'll just have to be. It's a shame I don't have an NVIDIA GPU myself to mess about with for this.