How to get the GPU info?

Question

I'm looking for a command that would give me the same info as:

cat /proc/cpuinfo 

Except for the GPU (type of the chip and memory, frequency).

Solution 1

That type of information is non-standard, and the tools you will use to gather it vary widely.

The command glxinfo will give you all available OpenGL information for the graphics processor, including its vendor name, if the drivers are correctly installed.

To get clock speed information, there is no standard tool.

• For ATI/AMD GPUs running the old Catalyst driver, aticonfig --odgc should fetch the clock rates, and aticonfig --odgt should fetch the temperature data. I'm not familiar with AMDGPU-Pro, but a similar tool should exist.
• For NVIDIA GPUs, the nvidia-smi tool will show all of the information you could want, including clock speeds and usage statistics.

I am not aware of an equivalent tool for the open source drivers or for Intel or other GPUs, but other information on the hardware can be fetched from the lspci and lshw tools.

Solution 2

I do not know of a direct equivalent, but lshw should give you the info you want, try:

sudo lshw -C display

(it also works without sudo but the info may be less complete/accurate)

You can also install the package lshw-gtk to get a GUI.

Solution 3

A blog post focusing on work done on the command-line is here:

http://www.cyberciti.biz/faq/howto-find-linux-vga-video-card-ram/

Find out the device ID:

 lspci | grep ' VGA ' | cut -d" " -f 1
03:00.0

You can then use this output with lspci again, forming two nested commands

lspci  -v -s  $(lspci | grep ' VGA ' | cut -d" " -f 1)

If you have more than 1 GPU card, try this equivalent command instead:

lspci | grep ' VGA ' | cut -d" " -f 1 | xargs -i lspci -v -s {}

Output from my system:

03:00.0 VGA compatible controller: NVIDIA Corporation G98 [Quadro NVS 295] (rev a1) (prog-if 00 [VGA controller])
    Subsystem: NVIDIA Corporation Device 062e
    Flags: bus master, fast devsel, latency 0, IRQ 24
    Memory at f6000000 (32-bit, non-prefetchable) [size=16M]
    Memory at ec000000 (64-bit, prefetchable) [size=64M]
    Memory at f4000000 (64-bit, non-prefetchable) [size=32M]
    I/O ports at dc80 [size=128]
    [virtual] Expansion ROM at f7e00000 [disabled] [size=128K]
    Capabilities: <access denied>
    Kernel driver in use: nvidia

EDIT: You can avoid the <access denied> by launching with sudo

So, (prefetchable) [size=64M) indicates that I have a 64-MB NVIDIA card. However, I don't, it's rather 256 MB. Why? See below.

To see how to get the most info and performance out of it, read an extremely comprehensive article on the Arch-Linux Wiki

https://wiki.archlinux.org/index.php/NVIDIA

For nvidia users, start with

nvidia-smi

(This works with the Nvidia drivers installed,but not with systems running the open-source 'nouveau' driver).

Output

Thu Dec 19 10:54:18 2013       
+------------------------------------------------------+                       
| NVIDIA-SMI 5.319.60   Driver Version: 319.60         |                       
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Quadro NVS 295      Off  | 0000:03:00.0     N/A |                  N/A |
| N/A   73C  N/A     N/A /  N/A |      252MB /   255MB |     N/A      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Compute processes:                                               GPU Memory |
|  GPU       PID  Process name                                     Usage      |
|=============================================================================|
|    0            Not Supported                                               |
+-----------------------------------------------------------------------------+

This indicates that I have a 256 MB GDDR3 Graphics card.

At this time, I don't know how to get this for Intel and AMD/ATI GPUs.

Solution 4

Run google-chrome and navigate to the URL about:gpu. If chrome has figured out how to use OpenGL, you will get extremely detailing information about your GPU.

Solution 5

Because you specified a command like cat for CPU's this is therefore the equivalent for GPU's. Specifically for Nvidia cards. It requires no software except the Nvidia device driver to be loaded.

The path here works for the cards I have. But yours may differ as others have pointed out in the comments.

1st GPU

> cat /proc/driver/nvidia/gpus/0/information
Model:       GeForce GTX 680
IRQ:         24
GPU UUID:    GPU-71541068-cded-8a1b-1d7e-a093a09e9842
Video BIOS:      80.04.09.00.01
Bus Type:    PCIe
DMA Size:    40 bits
DMA Mask:    0xffffffffff
Bus Location:    0000:01.00.0

2nd GPU

> cat /proc/driver/nvidia/gpus/1/information
Model:       GeForce GTX 580
IRQ:         33
GPU UUID:    GPU-64b1235c-51fc-d6f1-0f0e-fa70320f7a47
Video BIOS:      70.10.20.00.01
Bus Type:    PCIe
DMA Size:    40 bits
DMA Mask:    0xffffffffff
Bus Location:    0000:08.00.0

View more solutions

Comments

[Charlie Parker]

what does: cat /proc/cpuinfo do? what info are you looking for?

[Emobe]

@CharlieParker it outputs information of the cpu...

[jmardones]

Had to put gksu before the command in the menu to get lshw-gtk to work.

[Martin Thoma]

+1 for nvidia-smi (that should be highlighted a bit in my opinion)

[stiv]

How to install glxinfo?

[Brett Rigby]

@stiv: It's part of the Mesa library, and comes with the package mesa-utils on Ubuntu.

[matt wilkie]

Thanks! (thoughcat /proc/driver/nvidia/gpus/0000\:01\:00.0/information for me)

[Eliah Kagan]

This also works in Chromium (chromium-browser).

[Elder Geek]

I'm not seeing the video card frequency and memory in this answer. There are far simpler methods to obtain the model of GPU which appears to be all you are giving us. I'm not sure what this adds to the existing answers.

[Elder Geek]

aticonfig doesn't appear to be available since the retirement of fglrx. nvclock also appears to have been abandoned since the last version was for trusty. Do you have any updated solutions? Here's what I have so far..

[Elder Geek]

Any updates? I'm a fan of the command but the only clock rate (frequency) it seems to provide for me is the base bus clock 33MHz. I'm attempting to bring this Q&A up to date. Thank you!

[Braden Best]

The screenfetch program does the same thing and doesn't require a PPA to install.

[HaoZeke]

It's a shell script. Plus I linked to its github as well so you can just use it as a script.

[Camille Goudeseune]

Here's screenfetch: github.com/KittyKatt/screenFetch

[INElutTabile]

This should really be the accepted answer!

[Nathan]

Apologies, new to Deep Learning. What should it say if I have a GPU? It says product: 2nd Generation Core Processor Family Integrated Graphics Controller

[John Hamilton]

glxinfo | grep "Device" worked well enough for me on an Intel GPU

[Jacksonkr]

Clever. Along these lines I additionally went to chromeexperiments.com to see the performance there. Smooth as butter - I'm definitely on gpu

[user5280911]

This is the only correct answer in on-demand cloud/HPC cluster environment on which glxinfo or lspci both fail (the former because there's no OpenGL and display, the latter because the nVidia graphics card is abstracted by a graphics controller like Matrox G200eW3). The folder name under gpus is 0000:3b:00.0 or 0000:d8:00.0 for me, so we should type: cat /proc/driver/nvidia/gpus/0000:3b:00.0/information. The lovely Tesla V100-PCIE-16GB model shows that the qsub job limit is satisfied as desired.

[mrgloom]

nvidia-smi not showing me full GPU name.

[Chris1804505]

I use: glxinfo | egrep -i 'device|memory'

[Torsten Bronger]

All these greps yield “llvmpipe” on my system. Very odd model …

[Nayantara Jeyaraj]

@stiv glxinfo can be installed using the command : sudo apt install mesa-utils

[johny why]

I'm getting 04. My GPU is 00:02.0 VGA compatible controller: Intel Corporation Mobile GM965/GL960 Integrated Graphics Controller (primary) (rev 03) (prog-if 00 [VGA controller])

[Nickolay]

nvidia-smi -q, as suggested by @Quanlong uses more sensible output format.

[IT_User]

nvidia-smi will show stats as well to include GB size.

[boomkin]

This was also an issue solver for me with my NVIDIA card. Found this in the official CUDA installation guide: docs.nvidia.com/cuda/cuda-installation-guide-linux/…

[Jack G]

@Nathan That means that you have a GPU....probably a very weak GPU. Your GPU says integrated graphics, which means that it's integrated into the CPU. Your CPU has it own component which functions as a graphics card and probably (to save on costs) uses the ordinary RAM to store its buffers. You do not have a separate independent removable graphics card.

[Shayan Amani]

to be even shorter: $nvidia-smi -L

[Protect children of Donbas2014]

Only NVIDIA supported.

[Eric Wiener]

You can also use $ nvidia-smi --query-gpu=gpu_name --format=csv to just get the GPU name

[Mina Michael]

lspci did the trick for me.

[Admin]

nvidia-smi is only an option if the driver is installed from what I know. In case of not wanting to use the driver (e.g. libvirt and PCI passthrough of the GPU) this is not possible since the GPU is detatched from the host.