GPU Information through Python [Linux]

explaination.md

Idea-Board

This Document is meant to explain every possible approach on how to obtain GPU Information on Linux. Each approach should include the Points: "file", "Step-by-Step", "strengths and weaknesses"

lspci

We will use lspci to deliver hardware information about all PCI Devices. Then we will use lspci again, to deliver detailed output for every "GPU" device.

file

lspci_GPUinfo.py

Step-by-Step

1. run lshw
2. iterate over result lines and determine GPUs
3. run lshw -v -s {pci_id (for each GPU)
4. Build the dict() gpu_element and fill with the information
5. append gpu_element to a list of GPUs

strengths and weaknesses

strengths	weaknesses
No Installation	python external
	output parsing difficult
	unreliable

lspci_GPUinfo.py

import subprocess
from json import dumps

all_devices = subprocess.check_output(["lspci"], text=True).strip().split("\n")
print("-" * 20)

gpu_count = 0
gpu_list = list()
for device in all_devices:
    if "VGA compatible controller" in device or "3D controller" in device:
        try:
            pci_id = device.split()[0]

            # Get detailed information for the GPU device
            detail_output = (
                subprocess.check_output(["lspci", "-v", "-s", pci_id], text=True)
                .strip()
                .split("\n")
            )
        except subprocess.CalledProcessError:
            continue

        gpu_element = dict()

        vendor = "Unknown"
        product_name = "Unknown"

        # Get vendor and product name from the initial lspci line
        if "NVIDIA" in device:
            vendor = "NVIDIA"
            product_name = " ".join(device.split()[2:])
        elif "AMD" in device or "Advanced Micro Devices" in device:
            vendor = "AMD"
            product_name = " ".join(device.split()[2:])
        elif "Intel" in device:
            vendor = "Intel"
            product_name = device.split(":")[-1].strip()
        else:
            product_name = " ".join(device.split()[1:])

        gpu_element["product"] = product_name
        gpu_element["vendor"] = vendor

        businfo = f"pci@{pci_id}"

        detail_info = dict()
        for detail_line in detail_output:
            if detail_line.strip() == "":
                continue
            try:
                key, value = detail_line.split(":", 1)
                detail_info[key.strip()] = value.strip()
            except ValueError:
                # Handle lines that do not contain key-value pairs
                detail_info["Info"] = (
                    gpu_element.get("Info", "") + detail_line.strip() + " "
                )
        
        # Adjusting for iGPU / dedicated GPU
        if "DeviceName" in detail_info:
            device_class = detail_info["DeviceName"]
        else:
            device_class = "dedicated"

        configuration = {
            "driver": detail_info["Kernel driver in use"],
        }
        gpu_element = {
            "id": f"GPU{gpu_count+1}",
            "class": device_class,
            "description": detail_info["Subsystem"],
            "product": product_name,
            "vendor": vendor,
            "physid": pci_id,
            "businfo": businfo,
            "configuration": configuration,
        }

        gpu_list.append(gpu_element)
        gpu_count += 1


print(f"Total GPUs found: {gpu_count}")
print(dumps(gpu_list, indent=4))

this looked helpful.
Sorry if you already had it......

Next time we chat we should talk about lshw -C display. I think it may be a quicker approach to what you're looking for.

Widely used distros that should be aimed at making sure are compatible.

1. Debian/Ubuntu
2. Fedora Linux
3. opeSUSE
4. Arch Linux
5. Gentoo

These seem to be the base os's that most branches go off of per Wikipedia 👀

A better source should probably be acquired but I felt this was a good start.

I have that Dev server I was planning on installing. I can do testing of any variation from whichever hardware info tool we plan to use and see what variations in reporting there are.