Guide to install Ubuntu 14.04.05 on a brand new Alienware 15 R3 to dual boot with Windows 10 and be able to run CUDA code on the Nvidia GeForce 1070 GTX.
The major part of this guide may be useful for Ubuntu 16.04.
I hope it helps someone else, it took me 12 straight hours to get it working!
Just next, next, next filling up your data.
You should get a BIOS update alert from the Alienware Update widget. If not,
right click on the Down arrow icon in the bottom right extra icons ^ thing and
right click, then click Check for Updates.
Install it. It will reboot your computer, try to not touch anything up until you are back to Windows.
Go to the disk manager (right click on Windows icon > Disk management) and
Shrink the OS partition (right click on it, Shrink Volume...). It offered
me shrinking by 115XXX MB. Just shrink it 110000 MB. Shrinking it more won't work.
For Ubuntu to see the NVME disk it needs to boot on AHCI not on RAID mode (the default). As you don't want to go to the BIOS to change the SATA mode every time you want to boot in one or another OS, we need to force Windows to be able to boot in AHCI mode.
For that follow the instructions (kindly taken from here):
- Run Command Prompt as Admin.
- Invoke a Safe Mode boot with the command:
bcdedit /set {current} safeboot minimal. - Restart the PC and enter your BIOS during bootup (F2 key).
- In tab
Advancedchange optionSATA OperationfromRAID ontoAHCImode then go toExittab and useSave Changes and Reset. - Windows 10 will launch in Safe Mode.
- Right click the Window icon and select to run the Command Prompt in Admin mode from among the various options.
- Cancel Safe Mode booting with the command:
bcdedit /deletevalue {current} safeboot. - Restart your PC once more and this time it will boot up normally but with AHCI mode activated.
- Enjoy your awesomeness.
You need to disable secure boot in order to boot any other OS.
Enter your BIOS (F2 key on boot). Go to Boot tab and change Secure Boot option to Disabled.
Note that Boot List Option should be UEFI (it's the default).
I got my image from the official Ubuntu releases link scrolling down to find ubuntu-14.04.5-desktop-amd64.iso. I used Firefox + DownThemAll! addon to download it faster.
I use UNetbootin for writing my images.
I use Gparted to format my pendrive to FAT32.
The easiest way is to just clone or download this repo to get the files you'll need later on.
You can download it either doing:
git clone https://github.com/awesomebytes/alienware15r3_ubuntu14
Or clicking in Clone or download > Download ZIP.
The gparted debian came from ubuntuupdates.org gparted. Direct link is here which may or not work.
Copy the repo to the pendrive root folder into a new folder, I created one called
alienware_stuff. You'll need it later on.
Press F12 while booting and choose under UEFI OPTIONS to boot from your pendrive, for me it was
USB1 - UEFI OS( USB DISK 3.0 PMAP).
In order for the installation wizard to be able to deal with your NVME disk (the SSD) you need the newest Gparted.
To install it, open a terminal (Control+Alt+T) and:
cd /cdrom
cd alienware_stuff # Or whatever you called the folder
sudo dpkg -i gparted_0.24.0-1~getdeb1_amd64.deb
Just double click the Install Ubuntu 14.04.05 LTS desktop icon.
Configure as you like BUT DON'T ENABLE DOWNLOAD UPDATES WHILE INSTALLING NOR INSTALL THIRD PARTY SOFTWARE. It will freeze your installation. If you don't believe me, just try and enjoy your reboot.
Click on Something else.
Now you should see some partitions like /dev/nvme0n1. If you don't, you missed some step.
Now choose the free space partition that corresponds to the shrinked space we made before. For me it's
115343 MB. I'll just make a partition for / and another swap one.
In order to be able to suspend in Ubuntu you'll need at least your amount of RAM as swap. I doubt very much it will actually work, but hey, you need to try.
I have 16GB of RAM so I'll do 115343 - 17 * 1024 = 97884 MB partition. (Yeah that's a 17, I'm a bit lazy to check for how much exactly it should be).
Click on that free space to be selected and click on the + symbol. Put your amount
of MB for it (97884) in Size. Choose Logical as Type. Leave Location as Beginning of this space. Use as Ext4. Mount point as /.
Then on the left free space, repeat the process but make it of type swap.
IMPORTANT now you need to change the Device for boot loader installation to /dev/nvme0n1.
Now you can click Install Now.
In a few minutes you should be good to go!
Once rebooted and in your freshly installed Ubuntu 14.04, you'll need to copy some files from the bootable pendrive.
For the Wifi card QCA6174 you need a newer binary of the firmware (based on this askubuntu post).
Just copy the folder of this repo QCA6174 to /lib/firmware/ath10k. Note that you'll be
overwriting what is already there.
cd /media/YOURUSER/YOURPENDRIVE/alienware_stuff
sudo cp -r QCA6174 /lib/firmware/ath10kOnce done, reboot. Wifi and Ethernet will be working after.
I found no way of using the Nvidia 100% of the time, but using the Intel HD 530 card
for normal use and executing programs with optirun worked.
In all my other approachs when trying to use CUDA demos I'd get (I hope this helps someone googling):
CUDA Device Query (Runtime API) version (CUDART static linking)
cudaGetDeviceCount returned 35
-> CUDA driver version is insufficient for CUDA runtime version
Result = FAIL
Which by the way you will get when you forget to prepend optirun to
whatever that needs to run in the nvidia card for CUDA.
Instructions next:
Download from Nvidia Cuda 8.0, this is based on jayakumar blog post:
wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1404_8.0.61-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1404_8.0.61-1_amd64.deb
sudo apt-get update
sudo apt-get install cudaThis previously installed nvidia-367, now it installs nvidia-375.
You may want to add to your .bashrc:
export PATH=/usr/local/cuda-8.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
Before rebooting do (if you reboot without doing this you'll break your X):
sudo prime-select intelsudo apt-get install bumblebee bumblebee-nvidia primus linux-headers-genericNow reboot.
Put a copy of the CUDA demos in your home:
/usr/local/cuda-8.0/bin/cuda-install-samples-8.0.sh ~
Then you can compile a couple of demos:
cd ~/NVIDIA_CUDA-8.0_Samples/1_Utilities/deviceQuery
make
optirun ./deviceQueryI got as output:
./deviceQuery Starting...
CUDA Device Query (Runtime API) version (CUDART static linking)
Detected 1 CUDA Capable device(s)
Device 0: "GeForce GTX 1070"
CUDA Driver Version / Runtime Version 8.0 / 8.0
CUDA Capability Major/Minor version number: 6.1
Total amount of global memory: 8113 MBytes (8507555840 bytes)
(16) Multiprocessors, (128) CUDA Cores/MP: 2048 CUDA Cores
GPU Max Clock rate: 1645 MHz (1.64 GHz)
Memory Clock rate: 4004 Mhz
Memory Bus Width: 256-bit
L2 Cache Size: 2097152 bytes
Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048 layers
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 49152 bytes
Total number of registers available per block: 65536
Warp size: 32
Maximum number of threads per multiprocessor: 2048
Maximum number of threads per block: 1024
Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
Maximum memory pitch: 2147483647 bytes
Texture alignment: 512 bytes
Concurrent copy and kernel execution: Yes with 2 copy engine(s)
Run time limit on kernels: Yes
Integrated GPU sharing Host Memory: No
Support host page-locked memory mapping: Yes
Alignment requirement for Surfaces: Yes
Device has ECC support: Disabled
Device supports Unified Addressing (UVA): Yes
Device PCI Domain ID / Bus ID / location ID: 0 / 1 / 0
Compute Mode:
< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 8.0, CUDA Runtime Version = 8.0, NumDevs = 1, Device0 = GeForce GTX 1070
Result = PASS
Another demo:
cd ~/NVIDIA_CUDA-8.0_Samples/1_Utilities/bandwidthTest
make
optirun ./bandwidthTestWhich gave me:
optirun ./bandwidthTest
[CUDA Bandwidth Test] - Starting...
Running on...
Device 0: GeForce GTX 1070
Quick Mode
Host to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 6230.3
Device to Host Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 6346.4
Device to Device Bandwidth, 1 Device(s)
PINNED Memory Transfers
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 192311.3
Result = PASS
For other demos you may need to install:
sudo apt-get install freeglut3-dev mesa-common-dev
Furthermore, check usage of the Nvidia 1070 GTX with glxheads:
sudo apt-get install mesa-utils
sam@alien:~$ glxheads
glxheads: exercise multiple GLX connections (any key = exit)
Usage:
glxheads xdisplayname ...
Example:
glxheads :0 mars:0 venus:1
Name: :0
Display: 0xe81010
Window: 0x5000002
Context: 0xe90200
GL_VERSION: 3.0 Mesa 11.2.0
GL_VENDOR: Intel Open Source Technology Center
GL_RENDERER: Mesa DRI Intel(R) HD Graphics 530 (Skylake GT2)
And:
sam@alien:~$ optirun glxheads
glxheads: exercise multiple GLX connections (any key = exit)
Usage:
glxheads xdisplayname ...
Example:
glxheads :0 mars:0 venus:1
Name: :0
Display: 0x1f324e0
Window: 0x5200002
Context: 0x2276610
GL_VERSION: 4.5.0 NVIDIA 367.57
GL_VENDOR: NVIDIA Corporation
GL_RENDERER: GeForce GTX 1070/PCIe/SSE2
Checking dmesg it complained about not having nss-myhostname.
sudo apt-get install libnss-myhostname
Solves it.
There is a tool to change the colors of the LEDs for Alienware 17 that should be easy to make it work on this laptop: https://github.com/acyed/qtFx
I'm following abhay.harpale.net blog post.
- Download the latest CuDNN v5.1. I choose the debians, note that you need to register to acess:
cuDNN v5.1 Runtime Library for Ubuntu14.04 (Deb)
cuDNN v5.1 Developer Library for Ubuntu14.04 (Deb)
cuDNN v5 Code Samples and User Guide (Deb)
- Install it.
$ ls
libcudnn5-dev_5.1.10-1+cuda8.0_amd64.deb
libcudnn5_5.1.10-1+cuda8.0_amd64.deb
libcudnn5-doc_5.1.10-1+cuda8.0_amd64.deb
$ sudo dpkg -i libcudnn5*.deb- Add to your
LD_LIBRARY_PATHcuda, you probably want to add this to your.bashrc:
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64"
export CUDA_HOME=/usr/local/cuda
- Install Tensorflow dependencies:
sudo apt-get install swig python-dev python-numpy python-wheel
Install Bazel, for Ubuntu 14.04 we have nice instructions.
First we need java:
sudo add-apt-repository ppa:webupd8team/java
sudo apt-get update
sudo apt-get install oracle-java8-installer
Add Bazel APT source:
echo "deb [arch=amd64] http://storage.googleapis.com/bazel-apt stable jdk1.8" | sudo tee /etc/apt/sources.list.d/bazel.list
curl https://bazel.build/bazel-release.pub.gpg | sudo apt-key add -
Install it:
sudo apt-get update && sudo apt-get install bazel
- Install Tensorflow:
Get the latest version
git clone https://github.com/tensorflow/tensorflow
To avoid the error:
Auto-Configuration Error: Cannot find cudnn.h at /usr/lib/x86_64-linux-gnu/include/cudnn.h
I needed to do:
cd /usr/lib/x86_64-linux-gnu
sudo ln -s /usr/include includeAs locate cudnn.h told me it was at /usr/include/cudnn.h.
Configure and compile:
cd tensorflow
./configureI got as output (and my input, I added some flags as my CPU admits extra optimizations):
./configure
~/Downloads/tensorflow ~/Downloads/tensorflow
Please specify the location of python. [Default is /usr/bin/python]:
Please specify optimization flags to use during compilation when bazel option "--config=opt" is specified [Default is -march=native]: --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-mfpmath=both --copt=-msse4.2 --config=cuda
Do you wish to use jemalloc as the malloc implementation? [Y/n]
jemalloc enabled
Do you wish to build TensorFlow with Google Cloud Platform support? [y/N]
No Google Cloud Platform support will be enabled for TensorFlow
Do you wish to build TensorFlow with Hadoop File System support? [y/N]
No Hadoop File System support will be enabled for TensorFlow
Found possible Python library paths:
/usr/local/lib/python2.7/dist-packages
/usr/lib/python2.7/dist-packages
Please input the desired Python library path to use. Default is [/usr/local/lib/python2.7/dist-packages]
Using python library path: /usr/local/lib/python2.7/dist-packages
Do you wish to build TensorFlow with OpenCL support? [y/N]
No OpenCL support will be enabled for TensorFlow
Do you wish to build TensorFlow with CUDA support? [y/N] y
CUDA support will be enabled for TensorFlow
Please specify which gcc should be used by nvcc as the host compiler. [Default is /usr/bin/gcc]:
Please specify the CUDA SDK version you want to use, e.g. 7.0. [Leave empty to use system default]:
Please specify the location where CUDA toolkit is installed. Refer to README.md for more details. [Default is /usr/local/cuda]:
Please specify the Cudnn version you want to use. [Leave empty to use system default]:
Please specify the location where cuDNN library is installed. Refer to README.md for more details. [Default is /usr/local/cuda]:
Please specify a list of comma-separated Cuda compute capabilities you want to build with.
You can find the compute capability of your device at: https://developer.nvidia.com/cuda-gpus.
Please note that each additional compute capability significantly increases your build time and binary size.
[Default is: "3.5,5.2"]:
INFO: Starting clean (this may take a while). Consider using --expunge_async if the clean takes more than several minutes.
.......
INFO: All external dependencies fetched successfully.
Configuration finishedBuild it with GPU support and extra optimizations (from this thread):
bazel build -c opt --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-mfpmath=both --copt=-msse4.2 --config=cuda -k //tensorflow/tools/pip_package:build_pip_package
That was 25 minutes of compilation.
Create the pip package and install:
sudo apt-get install python-setuptools python-pip
bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
sudo pip install /tmp/tensorflow_pkg/tensorflow-1.0.1-cp27-none-linux_x86_64.whl
You can find my tensorflow-1.0.1-cp27-none-linux_x86_64.whl in the releases section (68MB) or direct link here.
If you want to skip compiling.
Install latest tflearn:
sudo pip install git+https://github.com/tflearn/tflearn.git
optirun python test_tensorflow_gpu.py
I tensorflow/stream_executor/dso_loader.cc:128] successfully opened CUDA library libcublas.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:128] successfully opened CUDA library libcudnn.so.5 locally
I tensorflow/stream_executor/dso_loader.cc:128] successfully opened CUDA library libcufft.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:128] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:128] successfully opened CUDA library libcurand.so.8.0 locally
I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:909] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
I tensorflow/core/common_runtime/gpu/gpu_device.cc:885] Found device 0 with properties:
name: GeForce GTX 1070
major: 6 minor: 1 memoryClockRate (GHz) 1.645
pciBusID 0000:01:00.0
Total memory: 7.92GiB
Free memory: 7.83GiB
I tensorflow/core/common_runtime/gpu/gpu_device.cc:906] DMA: 0
I tensorflow/core/common_runtime/gpu/gpu_device.cc:916] 0: Y
I tensorflow/core/common_runtime/gpu/gpu_device.cc:975] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0)
Device mapping:
/job:localhost/replica:0/task:0/gpu:0 -> device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0
I tensorflow/core/common_runtime/direct_session.cc:255] Device mapping:
/job:localhost/replica:0/task:0/gpu:0 -> device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0
MatMul: (MatMul): /job:localhost/replica:0/task:0/gpu:0
I tensorflow/core/common_runtime/simple_placer.cc:827] MatMul: (MatMul)/job:localhost/replica:0/task:0/gpu:0
b: (Const): /job:localhost/replica:0/task:0/gpu:0
I tensorflow/core/common_runtime/simple_placer.cc:827] b: (Const)/job:localhost/replica:0/task:0/gpu:0
a: (Const): /job:localhost/replica:0/task:0/gpu:0
I tensorflow/core/common_runtime/simple_placer.cc:827] a: (Const)/job:localhost/replica:0/task:0/gpu:0
[[ 22. 28.]
[ 49. 64.]]
First install:
sudo apt-get install python-h5py python-scipy
Now run the test script from the tflearn docs test_tflearn.py:
optirun python test_tflearn.py
You'll get an output like:
19:12:06
~/Downloads$ python test_tflearn.py
Downloading Titanic dataset...
Succesfully downloaded titanic_dataset.csv 82865 bytes.
2017-03-21 19:12:26.354733: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2017-03-21 19:12:26.354754: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2017-03-21 19:12:26.354759: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2017-03-21 19:12:26.354763: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2017-03-21 19:12:26.354766: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
2017-03-21 19:12:26.512376: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:901] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2017-03-21 19:12:26.512778: I tensorflow/core/common_runtime/gpu/gpu_device.cc:887] Found device 0 with properties:
name: GeForce GTX 1070
major: 6 minor: 1 memoryClockRate (GHz) 1.645
pciBusID 0000:01:00.0
Total memory: 7.92GiB
Free memory: 7.46GiB
2017-03-21 19:12:26.512793: I tensorflow/core/common_runtime/gpu/gpu_device.cc:908] DMA: 0
2017-03-21 19:12:26.512798: I tensorflow/core/common_runtime/gpu/gpu_device.cc:918] 0: Y
2017-03-21 19:12:26.512804: I tensorflow/core/common_runtime/gpu/gpu_device.cc:977] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0)
2017-03-21 19:13:45.450308: I tensorflow/core/common_runtime/gpu/gpu_device.cc:977] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce GTX 1070, pci bus id: 0000:01:00.0)
---------------------------------
Run id: 6PYEA0
Log directory: /tmp/tflearn_logs/
---------------------------------
Training samples: 1309
Validation samples: 0
--
Training Step: 82 | total loss: 0.63943 | time: 0.941s
| Adam | epoch: 001 | loss: 0.63943 - acc: 0.6746 -- iter: 1309/1309
--
Training Step: 164 | total loss: 0.62572 | time: 0.225s
| Adam | epoch: 002 | loss: 0.62572 - acc: 0.6604 -- iter: 1309/1309
--
Training Step: 246 | total loss: 0.53110 | time: 0.218s
| Adam | epoch: 003 | loss: 0.53110 - acc: 0.7426 -- iter: 1309/1309
--
Training Step: 328 | total loss: 0.54184 | time: 0.221s
| Adam | epoch: 004 | loss: 0.54184 - acc: 0.7509 -- iter: 1309/1309
--
Training Step: 410 | total loss: 0.54874 | time: 0.215s
| Adam | epoch: 005 | loss: 0.54874 - acc: 0.7721 -- iter: 1309/1309
--
Training Step: 492 | total loss: 0.51747 | time: 0.217s
| Adam | epoch: 006 | loss: 0.51747 - acc: 0.7841 -- iter: 1309/1309
--
Training Step: 574 | total loss: 0.53963 | time: 0.227s
| Adam | epoch: 007 | loss: 0.53963 - acc: 0.7770 -- iter: 1309/1309
--
Training Step: 656 | total loss: 0.48992 | time: 0.214s
| Adam | epoch: 008 | loss: 0.48992 - acc: 0.7625 -- iter: 1309/1309
--
Training Step: 738 | total loss: 0.48196 | time: 0.225s
| Adam | epoch: 009 | loss: 0.48196 - acc: 0.7795 -- iter: 1309/1309
--
Training Step: 820 | total loss: 0.49632 | time: 0.216s
| Adam | epoch: 010 | loss: 0.49632 - acc: 0.7696 -- iter: 1309/1309
--
DiCaprio Surviving Rate: 0.126031458378
Winslet Surviving Rate: 0.84097880125
19:13:49
Or how to disable Intel HD 530 card.
You may want this if you:
- Want to ONLY use the nvidia card.
- Want to use an external monitor (it will enable HDMI and display port).
- Want to use HTC Vive or Steam as it needs to go thru the nVidia card.
- You want to use these instructions to honor the long hours I spent writing them.
You need to still have Windows 10 with the alienware and Nvidia drivers installed.
You need to press Fn+F7 which is the I/D GFX key. Windows will ask you if you want
to switch to using the discrete graphics card and to reboot if so. Accept.
Be warned that this change does not appear in the BIOS or anywhere else, and it can only be undone from Windows doing the same thing again.
Reboot and log into Windows, it will figure out the hardware change and just work.
Now you can reboot again to log into Ubuntu, but on the grub screen you'll need to press
e on your Ubuntu grub line to display the editing of boot string options.
You'll need to add just before quiet splash the word nomodeset so it will look like:
nomodeset quiet splash. Then press F10 to boot into it.
The graphical system will fail and if you try to login you won't be able, go to a console
pressing Control+Alt+F2, login with your username and password.
Now we'll uninstall anything related to bumblebee.
sudo apt-get remove bumblebee
It will remove bumblebee bumblebee-nvidia primus.
Now we reinstall the nvidia-367 driver. Update: for the latest version of CUDA it forced me to upgrade to nvidia-375. It works nicely, so this command may need to upgrade to this driver.
sudo apt-get install --reinstall nvidia-367
Now we will add to grub the nomodeset option so we don't need to do it manually anymore.
Edit /etc/default/grub (sudo nano /etc/default/grub) and add to GRUB_CMDLINE_LINUX_DEFAULT to make it look like: GRUB_CMDLINE_LINUX_DEFAULT="nomodeset quiet splash" save with Control+O, then Enter, then Control+X.
Now update grub with:
sudo update-grub
And reboot.
Apparently you need to remove everything from nvidia first.
sudo apt-get purge nvidia* cuda* bumblebee*
Reboot, now you can login (in crappy graphics) and execute: Update: for the latest version of CUDA it forced me to upgrade to nvidia-375. It works nicely, so this command may need to upgrade to this driver.
sudo apt-get install nvidia-367
Go to a system shell Control+Alt+F2 login and do:
sudo service lightdm restart
Or reboot.
Now you are only using the nvidia card!