Amazon Linux AMI with NVIDIA TESLA GPU Driver
NVIDIA | 2018.03.0.20201028.0Linux/Unix, Amazon Linux 2018.03 - 64-bit Amazon Machine Image (AMI)
External reviews
External reviews are not included in the AWS star rating for the product.
Amazon Linux AMI with NVIDIA TESLA GPU Driver
- Leave a Comment |
- Mark review as helpful
AWS is the best
Outdated version of the driver
The image features version 352.99 of the driver which is incompatible with CUDA-8.0/cuDNN-6. In my case that kind of diminishes the purpose of having such an AMI in the first place.
Couldn't get Blender running
Although I'm not a linux guru I do know how to handle the console, compile stuff etc.
I spent about 6 (payed) hours trying to get Blender running in this machine, without success.
- There's no Blender yum package
- After enabling Epel (and confirming that it's enabled) strangely there still is no Blender package
- Yum cannot resolve the dependencies of a recent Blender rpm (libGLEW etc)
- Compiling Blender there's version collisions with libgcc
- Tediously installing dependencies by hand resulted in more conflicts
Etc. pp.
I have no idea what kind of system that is and honestly I have no experience with RHEL. So some of you might laugh about me now, I'm ok with that.
But to my defense nobody on stackoverflow could answer that simple question "How to install Blender into this AMI" either.
I'm trying now my way with a generic Ubuntu image and getting CUDA and Nvidia drivers up and running there. Blender's working there, also a custom build, without issues.
why not up to date?
Uses old version of CUDA which is incompatible.
The whole point of using an AMI is that it works out of the box without having to install drivers. This is useless as not up to date and incompatible with other software.
Up to date and fully functional!
TLDR Ignore the old reviews complaining of old versions of stuff. This is running the same CUDA 7.5 that the Nvidia website says is the latest stable. Presumably, this marketplace AMI will stay up-to-date too! ----------------- I finally, *finally* managed to install Torch 7 without errors. > Single tear of joy :') I have tried on a bare metal Ubuntu 14 LTS box. And a bare metal Fedora 20 box. (Same box in my living room!) And a g2 instance running Ubuntu 14 LTS, with the drivers installed from Nvidia's .run file and another with the .deb packages. And two different community AMIs claiming to already have CUDA and/or Torch running. Every single one of those was a totally different, very time consuming, and fairly opaque failure. Out of date graphics drivers. Everything halts because nouveau is running. Your kernel headers are incompatible. CUDA 7.0 thinks it's 6.5. deviceQuery can't find your GPU after you restarted. Ugh. My unix-fu is not strong enough for this. I just fired this up, ran the torch install commands, and everything worked! Finally I can just push some data through an LSTM! Thank you Nvidia for the support :)
slicker than snot
Needed to do some CUDA development using a modern CUDA environment. Works great. Easily installed all the other tools I needed (emacs, git, X11 dev, etc...). Able to ssh and tunnel X11 -- wonderful -- saved me from having to buy a laptop. Code builds fine.
nVidia drivers updated as of June 2016
There are lots of complaints about the old nVidia drivers. These have been updated:
2016.03.2 Point Release (Released on June 9th, 2016)
The Amazon Linux AMI with NVIDIA GRID GPU Driver now includes NVIDIA driver version 352.79 and CUDA 7.5.18.
Drivers are outdated and lacks ecs-agent
It's a trade-off between installing drivers or other services.
1-star lost for not having updated drivers even though drivers are the entire point of this AMI
2-star lost for not having the ecs-agent installed and ready
Does the job
We've been using this image to develop and test our genetic analysis software on GPUs. Works very well. Moving to production.