Hybrid graphics

From ArchWiki
Revision as of 12:46, 23 February 2014 by Fengchao (Talk | contribs) (Remove stub tag. Auto style fix.)

Jump to: navigation, search

Hybrid-graphics is a concept involving two graphics cards on same computer, it was first designed to control power consumption in laptops and is extending to desktop computers as well

About Hybrid-graphics Technologies

The laptop manufacturers developed new technologies involving two graphic cards in an single computer, enabling both high performance and power saving usages. This technology is well supported on Windows but it's still quite experimental with Linux distributions.

We call hybrid graphics a set of two graphic cards with different abilities and power consumptions. There are a variety of technologies and each manufacturer developed its own solution to this problem. Here we try to explain a little about each approach and models and some community solutions to the lack of GNU/Linux systems support.

The "Old" Hybrid Model (Basic Switching)

This approach involves a two graphic card setup with a hardware multiplexer (MUX). It allows power save and low-end 3D rendering by using an Integrated Graphics Processor (IGP); or a major power consumption with 3D rendering performance using a Dedicated Graphics Processor (DGP). This model makes the user choose (at boot time or at login time) within the two power/graphics profiles and is almost fixed through all the user session. The switch is done by a similar workflow:

  • Turn off the display
  • Turn on the DGP
  • Switch the multiplexer
  • Turn off the IGP
  • Turn on again the display

This switch is somewhat rough and adds some blinks and black screens in laptops that could do it "on the fly". Later approaches made the transition a little more user-friendly.

The New Dynamic Switching Model

Most of the new Hybrid-graphics technologies involves two graphic cards as the basic switching but now the DGP and IGP are plugged to a framebuffer and there is no hardware multiplexer. The IGP is always on and the DGP is switched on/off when there is a need in power-save or performance-rendering. In most cases there is no way to use only the DGP and all the switching and rendering is controlled by software. At startup, the Linux kernel starts using a video mode and setting up low-level graphic drivers which will be used by the applications. Most of the Linux distributions then use X.org to create a graphical environment. Finally, a few other softwares are launched, first a login manager and then a window manager, and so on. This hierarchical system has been designed to be used in most of cases on a single graphic card.

Nvidia Optimus

Nvidia Optimus Whitepaper

Current Problems

  • Switching between cards when possible.
  • Switching on/off the discrete card.
  • Be able to use the discrete card for 3D render.
  • Be able to use both cards for 3D render (problem arised in this post).

Software Solutions So Far

  • asus_switcheroo -- a solution for Intel/Nvidia switching on ASUS and other laptops with a similar hardware mux -- by Alex Williamson
  • byo_switcheroo -- a solution to build your own handler (like acpi_call) to switch between cards with vga_switcheroo -- by Alex Williamson
  • vga_switcheroo -- the original GPU switching solution primarily for Intel/ATI notebooks -- by David Airlie
  • acpi_call -- allows you to switch off discrete graphics card to improve battery life -- by Michal Kottman
  • PRIME -- long-term Optimus solution in progress -- by David Airlie
  • Bumblebee -- allows you to run specific programs on the discrete graphic card, inside of an X session using the integrated graphic card. Works on Nvidia Optimus cards -- by Martin Juhl
  • hybrid-windump -- dump window using Nvidia onto Intel display -- by Florian Berger and Joakim Gebart

ATI Dynamic Switchable Graphics

This is a new technology similar to the one of Nvidia as it uses no hardware multiplexer.

Current Problems

The Dynamic Switch needs Xorg support for the discrete videocard assigned for rendering to work [1]. So, rendering on the discrete gpu will not work until the Xorg team adds support for it.

This means that with a muxless intel+ati design, you cannot use your discrete card by simply modprobing the radeon module.

As of now, there are 3 choices:

- Disable the discrete card and use only the intel igpu. This is not needed for kernels version >= 3.12 with radeon DPM enabled; the open source graphics driver manages the card automatically. For kernels older than 3.12, see the solution below.

- Test and improve some virtualGL based program to make the switch, like the common-amd branch of bumblebee project. Check the project repository and this useful post.

- Use the proprietary driver with powerxpress (a.k.a. pxp) support maintained by Vi0l0 (remember to check for xorg compatibility).

Solution for kernels < 3.12 or without radeon dynamic power management enabled

Warning: This method, on a mux-less system, works only to shutdown the radeon card. This will not enable rendering on the radeon gpu. See Current Problems section above for detail.

This solution is not needed on kernel version >= 3.12 because the opensource driver automatically manages the power of the radeon gpu, so there is no need to manage the cards from userspace.

This means that on kernels >= 3.12, vgaswitcheroo is not needed anymore to turn off the discrete gpu, only if you wish to verify the power state.

If you have kernel >= 3.12 with vgaswitcheroo enabled, you can verify if the driver automatically shut down the discrete gpu

# cat /sys/kernel/debug/vgaswitcheroo/switch

The output should be similar to this, where DIS is the radeon discrete gpu and IGD the intel gpu. DynOff means the radeon driver automatically powered off the discrete gpu.

0:DIS: :DynOff:0000:01:00.0
1:IGD:+:Pwr:0000:00:02.0

If you are using kernels older than 3.12 then you can use vga_switcheroo with a combination of opensource drivers to disable the radeon card from userspace at boot.

To do this, follow the instructions below.

  • Preliminaries

Make sure you have installed the drivers. Run in terminal:

$ pacman -Q | grep -E "xf86-video-ati|xf86-video-intel"

In case you get output similar to this:

xf86-video-ati 6.14.1-1
xf86-video-intel 2.15.0-2

you are good to go. In other case install drivers:

# pacman -S xf86-video-ati xf86-video-intel

In order to be able to access vgaswitcheroo add this line to your fstab:

none            /sys/kernel/debug debugfs defaults 0 0
Note: KMS must be activated for both cards, otherwise there will be no vgaswitcheroo in /sys/kernel/debug/
  • Automatic radeon shutdown

Systemd can use tmpfiles to shutdown the discrete gpu at boot.

Important: Make sure the video drivers are loaded in initramfs before systemd calls vga_switcheroo, otherwise a kernel oops/panic may occur.

First add the drivers to MODULES array in /etc/mkinitcpio.conf. Adding radeon and i915 yields

/etc/mkinitcpio.conf
MODULES="radeon i915"

Next rebuild initramfs (details at initramfs)

# mkinitcpio -p linux

Then create the systemd tmpfile at /etc/tmpfiles.d/vgaswitcheroo.conf

/etc/tmpfiles.d/vgaswitcheroo.conf
w /sys/kernel/debug/vgaswitcheroo/switch - - - - OFF

Reboot and the discrete gpu should be off by default. It can be powered back on using the manual method described below.

  • Manual method

To verify the state of the dgpu

# cat /sys/kernel/debug/vgaswitcheroo/switch

Power off the dgpu

# echo OFF > /sys/kernel/debug/vgaswitcheroo/switch

Power on

# echo ON > /sys/kernel/debug/vgaswitcheroo/switch

Fully Power Down Discrete GPU

You may want to turn off the high-performance graphics processor to save battery power, this can be done by installing the the acpi_call-gitAUR package from the AUR.

Once installed load the kernel module:

modprobe acpi_call

With the kernel module loaded run the following (requires root):

turn_off_gpu.sh

This script will go through all the known data buses and attempt to turn them off. You will get an output similar to the following:

Trying \_SB.PCI0.P0P1.VGA._OFF: failed
Trying \_SB.PCI0.P0P2.VGA._OFF: failed
Trying \_SB_.PCI0.OVGA.ATPX: failed
Trying \_SB_.PCI0.OVGA.XTPX: failed
Trying \_SB.PCI0.P0P3.PEGP._OFF: failed
Trying \_SB.PCI0.P0P2.PEGP._OFF: failed
Trying \_SB.PCI0.P0P1.PEGP._OFF: failed
Trying \_SB.PCI0.MXR0.MXM0._OFF: failed
Trying \_SB.PCI0.PEG1.GFX0._OFF: failed
Trying \_SB.PCI0.PEG0.GFX0.DOFF: failed
Trying \_SB.PCI0.PEG1.GFX0.DOFF: failed
Trying \_SB.PCI0.PEG0.PEGP._OFF: works!
Trying \_SB.PCI0.XVR0.Z01I.DGOF: failed
Trying \_SB.PCI0.PEGR.GFX0._OFF: failed
Trying \_SB.PCI0.PEG.VID._OFF: failed
Trying \_SB.PCI0.PEG0.VID._OFF: failed
Trying \_SB.PCI0.P0P2.DGPU._OFF: failed
Trying \_SB.PCI0.P0P4.DGPU.DOFF: failed
Trying \_SB.PCI0.IXVE.IGPU.DGOF: failed
Trying \_SB.PCI0.RP00.VGA._PS3: failed
Trying \_SB.PCI0.RP00.VGA.P3MO: failed
Trying \_SB.PCI0.GFX0.DSM._T_0: failed
Trying \_SB.PCI0.LPC.EC.PUBS._OFF: failed
Trying \_SB.PCI0.P0P2.NVID._OFF: failed
Trying \_SB.PCI0.P0P2.VGA.PX02: failed
Trying \_SB_.PCI0.PEGP.DGFX._OFF: failed
Trying \_SB_.PCI0.VGA.PX02: failed

See the "works"? This means the script found a bus which your GPU sits on and it has now turned off the chip. To confirm this, your battery time remaining should have increased. Currently, the chip will turn back on with the next reboot to get around this we do the following:

Note: To turn the GPU back on just reboot.

Add the kernel module to the array of modules to load at boot:

/etc/modules-load.d/acpi_call.conf
#Load 'acpi_call.ko' at boot.

acpi_call

To turn off the GPU at boot we could just run the above script but honestly that is not very elegant so instead lets make use of systemd's tmpfiles.

/etc/tmpfiles.d/acpi_call.conf

w /proc/acpi/call - - - - \_SB.PCI0.PEG0.PEGP._OFF

The above config will be loaded at boot by systemd. What it does is write the specific OFF signal to the /proc/acpi/call file. Obviously, replace the \_SB.PCI0.PEG0.PEGP._OFF with the one which works on your system.

Note: After every kernel upgrade acpi_call-git will need to be reinstalled.

See Also