Optimus on the Gigabyte p34w v5 under Ubuntu 15.10

2016 Mar 09 -

Through much experimentation I have finally been able to get optimus to work on the Gigabyte p34w v5; however, the specific implementation is far from flawless. In particular, starting the bumblebee daemon on boot causes the system to hang indefinitely. Thus one must start and stop bumblebeed as needed.

Before getting started, please read through my previous post about this machine and ensure you are running the linux 4.5-rc4 or later kernel.

Installing software

Getting bumblebee installed is a bit interesting on 15.10. The only instructions I found that work can be found in this Stack Overflow post. Because bumblebeed causes the machine to hang on boot, I’ve had to modify the instructions somewhat.

1. Boot your machine

2. Install nvidia driver and bumblebee in the same command. I actually used the CUDA Debian package for Ubuntu 15.04 described in my previous post about the p35w v5 however you can use anything after nvidia-352 where support for the GTX 970m was added. The Simple method of doing this is as follows:

   sudo aptitude install bumblebee nvidia-352 prime-select

3. After this is done, do not reboot. Instead immediately switch to the Intel driver using prime-select:

   sudo prime-select intel

4. Now configure bumblebee. Edit /etc/bumblebee/bumblebee.conf as follows: Set Driver=nvidia and replace nvidia-common with nvidia-352 in the [driver-nvidia] section. I’ve posted my full bumblebeed.conf file below as a reference if you need it.

5. Next /etc/bumblebee/xorg.conf.nvidia, uncomment BusID "PCI:01:00:0"

6. Now disable the bumblebeed service with the following command

   sudo systemctl disable bumblebeed.service

7. Reboot the system. With any luck it will come up running on the Intel card.

Using Optimus / Bumblebee

Because we disabled the bumblebee daemon on boot, we have to activate it whenever we want to use it. This is fairly easy to do with a few scripts. To turn on the NVIDIA card and start bumblebeed, I execute this nvidia-start.sh script:

#!/bin/bash

sudo modprobe bbswitch
sudo service bumblebeed start

And to put the machine back into a power-saving mode after using the NVIDIA card, I use the following nvidia-stop.sh script:

#!/bin/bash

sudo modprobe bbswitch
sudo rmmod nvidia_uvm
sudo rmmod nvidia
sudo service bumblebeed stop
sudo tee /proc/acpi/bbswitch <<< OFF
cat /proc/acpi/bbswitch

The nvidia-stop.sh script (1) unloads all of the modules (including the NVIDIA unified virtual memory module if you use CUDA), (2) shuts down the bumblebee service, (3) tells bbswitch to shut off the card, and then (4) echos the current state of the card.

Power consumption

Using a Kill A Watt power usage meter I checked the power consumption of my laptop with and without the NVIDIA card enabled. My results, shown in the following table, show that the bumblebee and bbswitch do indeed disable the discrete GPU; however, there appears to be a small (0.4 Watt) draw when the card is off, but not disabled.

Running applications on the discrete card

To execute an application on the NVIDIA card, simply run it using either optirun or primusrun as follows:

optirun program-name

One thing you will undoubtedly notice is that your FPS is inherently limited to about 60, which is the default vsync rate of the intel graphics card. To get higher throughput, simply change the vblank mode. Here is an example of how you could run glxgears

export vblank_mode=0
optirun glxgears
export vblank_mode=1

Lastly, here are a few performance results for the classic FPS, Nexuiz:

As you can see, we get a 40% boost in performance by running Nexuiz via. optirun. I suspect the difference would be more considerable for more modern games that push the GPU’s capabilities.

After getting it working, I haven’t experimented with different bridge methods (valid options are auto, virtualgl and primus) or with transport methods for VirtualGL (the options are proxy, jpeg, rgb, xv, and yuv) although I hear the performance differences are not considerable.

Configuration files

In case you need them, here are several of my configuration files:

My /etc/bumblebee/bumblebee.conf file:

[bumblebeed]
VirtualDisplay=:8
KeepUnusedXServer=false
ServerGroup=bumblebee
TurnCardOffAtExit=true
NoEcoModeOverride=false
Driver=nvidia
XorgConfDir=/etc/bumblebee/xorg.conf.d

[optirun]
Bridge=primus
VGLTransport=proxy
PrimusLibraryPath=/usr/lib/x86_64-linux-gnu/primus:/usr/lib/i386-linux-gnu/primus
AllowFallbackToIGC=false

[driver-nvidia]
KernelDriver=nvidia-352
PMMethod=auto
LibraryPath=/usr/lib/nvidia-352:/usr/lib32/nvidia-352
XorgModulePath=/usr/lib/nvidia-352/xorg,/usr/lib/xorg/modules
XorgConfFile=/etc/bumblebee/xorg.conf.nvidia

[driver-nouveau]
KernelDriver=nouveau
PMMethod=auto
XorgConfFile=/etc/bumblebee/xorg.conf.nouveau

And the /etc/bumblebee/xorg.conf.nvidia file:

Section "ServerLayout"
    Identifier  "Layout0"
    Option      "AutoAddDevices" "false"
    Option      "AutoAddGPU" "false"
EndSection

Section "Device"
    Identifier  "DiscreteNvidia"
    Driver      "nvidia"
    VendorName  "NVIDIA Corporation"

    # This should be the only line you need to modify by default.
    # just uncomment it.
    BusID "PCI:01:00:0"

    Option "ProbeAllGpus" "false"

    Option "NoLogo" "true"
    Option "UseEDID" "false"
    Option "UseDisplayDevice" "none"
EndSection

Ancillary information

How to run nvidia-settings

After bumblebee is running, simply execute nvidia-settings using optirun on the same virtual display specified in the /etc/bumblebee/bumblebee.conf file:

optirun nvidia-settings -c :8

This opens nvidia-settings on virtual terminal 8, where the NVIDIA GPU is rendering frames.

Check or change the power state for the discrete GPU

The power state of the discrete GPU is managed by bbswitch. After the module is loaded, you can manually change the power state of the discrete GPU using the following commands:

sudo tee /proc/acpi/bbswitch <<<OFF
sudo tee /proc/acpi/bbswitch <<<ON

Please note that the discrete GPU cannot be in use when you attempt to turn it off, hence you might need to unload the nvidia and nvidia-uvm kernel modules (see the scripts above) for examples.