[NEW] Getting Started with GPU #2516
Conversation
- Performing a copy edit instead of a tech edit, as our test server isn't available at the moment - This edit focused on spelling/grammar/syntax, with some expansion of a few instructions to make them more explicit - Will also add a few comments in GitHub related to some more content that I think we should include and will talk about this with rsyracuse
nmelehan
left a comment
nmelehan
left a comment
There was a problem hiding this comment.
Most of the comments in this review are just talking a bit about why I changed what I did. A few comments are about new things I'd like to see added to the guide. Let's talk about this on the phone or in person as part of our new effort to do GitHub reviews ✍️ 😸
|
|
||
| Issue a reboot to make sure you can use the missing kernel headers. | ||
| {{< note >}} | ||
| After running these commands, reboot your Linode from the [Cloud Manager](https://cloud.linode.com) to make sure you can use the missing kernel headers. |
There was a problem hiding this comment.
I think we should have more explanation around "missing kernel headers" (and more explanation in general around what the dependencies are doing here. Does that only apply to some of the distress' commands? I.e. does installing the build-essential package install kernel headers?
If we add this explanation, I think it should live either in the first paragraph of this section or under each distro (if the distros' dependencies are doing different things)
| The following guide will help you to get your dedicated GPU up and running on a number of popular distributions. In all cases where possible, this guide will use [NVIDIA's CUDA Toolkit](https://developer.nvidia.com/cuda-toolkit) to install the necessary NVIDIA driver. | ||
|
|
||
| {{< note >}} | ||
| For distributions like Debian 9 in which CUDA is not officially supported, visit the [Manual Install](#manual-install) section after completing the [Before You Begin](#before-you-begin) section to install the NVIDIA driver only. |
There was a problem hiding this comment.
I added a link to the Before You Begin section to make it a bit clearer that the user shouldn't directly skip to the Manual Install section
| --- | ||
|
|
||
| The following guide will help you to get your dedicated GPU up and running on a number of popular distributions. In all cases where possible, this guide will use [NVIDIA's CUDA Toolkit](https://developer.nvidia.com/cuda-toolkit) to install the necessary NVIDIA driver. For distributions like Debian 9 in which `CUDA` is not officially supported, the guide will show you how to manually install the NVIDIA driver only. | ||
| The following guide will help you to get your dedicated GPU up and running on a number of popular distributions. In all cases where possible, this guide will use [NVIDIA's CUDA Toolkit](https://developer.nvidia.com/cuda-toolkit) to install the necessary NVIDIA driver. |
There was a problem hiding this comment.
There should be a bit more info in the opening statements:
-
A clear statement of what this guide does (installs the drivers). This paragraph says it will get the Linode prepared for the GPU, and it says that it will use the CUDA toolkit to install the driver, but it doesn't say that installing the driver is how you prepare the Linode.
-
A short discussion around why this guide does what it does (i.e. the drivers are not pre-installed, and they are needed). Not sure how deep we want to go into why they are not preinstalled.
| Some dependencies may need to be installed prior to installing your driver. Listed below are the commands for installing these packages on many popular distributions: | ||
|
|
||
| ### Ubuntu 18 | ||
| ### Ubuntu 18.10 |
There was a problem hiding this comment.
The CUDA toolkit page lists other Ubuntu versions, but I think our draft has only been tested on Ubuntu 18.10 (Ryan said an image for 18.04 wasn't on the GPU host)
|
|
||
| ## Installing with CUDA | ||
|
|
||
| This section of the guide will highlight how to install your GPU driver using [NVIDIA's Cuda Toolkit](https://developer.nvidia.com/cuda-toolkit), currently fully supported on CentOS 7,OpenSUSE, and other popular distributions. Optionally, you can also choose to install the full toolkit in it's entirety to gain access to a set of tools that will empower you to create GPU-accelerated applications. |
There was a problem hiding this comment.
I split this paragraph into two paragraphs and split a few sentences into two to make the reading experience a bit more bite-sized
There was a problem hiding this comment.
Also, I set a rule in Vale which prefers "CUDA" to "Cuda"
There was a problem hiding this comment.
its (possessive) vs it's (it is)
|
|
||
| 1. Visit NVIDIA's [Driver Downloads Page](https://www.nvidia.com/Download/index.aspx?lang=en-us). | ||
|
|
||
| 1. Make sure that the options from the drop-down menus reflect the following values: |
There was a problem hiding this comment.
Added a text table here for the values (in addition to the screenshot). A blind reader wouldn't be able to see the values in the screenshot
| | Product Series | Quadro RTX Series | | ||
| | Product | Quadro RTX 8000 | | ||
| | Operating System | Linux 64-bit | | ||
| | Download Type | Optimal Driver for Enterprise (ODE) | |
There was a problem hiding this comment.
I actually couldn't get this second Download Type option to appear in their form. Can you show me where you saw this? I only saw one listing for the Download Type in the form, and the only option available for it was Linux Long Lived Driver
There was a problem hiding this comment.
This is a bug involving a cached response after searching for the driver. After clicking "search", go back to the downloads page and you should see the second Download Type, along with your previously chosen options pre-selected. We should use the non-cached response and consider adding a warning
|
|
||
|  | ||
|
|
||
| 1. Select "Search", "Download", and download the Linux driver that appears on the following screen to your Linode. |
There was a problem hiding this comment.
For UI elements, we generally bold the text (instead of putting them in quotes)
|  | ||
|
|
||
| 1. Select "Search", "Download", and download the Linux driver that appears on the following screen to your Linode. | ||
| 1. Click the **Search** button, and a page will appear that shows information about the driver. Click the green **Download** button on this page. The file will not download to your computer; instead, you will be taken to another download confirmation page. |
There was a problem hiding this comment.
Revised the existing instructions and modeled them on the copy-link-and-wget instructions from the Installing with CUDA section
There was a problem hiding this comment.
Also, there was one extra download confirmation page on NVIDIA's site that the previous instructions didn't mention
| 1. Once the installer has completed, use `nvidia-smi` to make sure that you're currently using your NVIDIA GPU with it's associated driver: | ||
| 1. Once the installer has completed, use `nvidia-smi` to make sure that you're currently using your NVIDIA GPU with its associated driver: | ||
|
|
||
| nvidia-smi |
There was a problem hiding this comment.
I think we should have a Next Steps section that follows the installation instructions. Maybe this could link to the other GPU guide we're writing when it's finished
|
|
||
| {{< note >}} | ||
|
|
||
| Installation on CentOS and RHEL operating systems will fail following this as the installer requires a reboot to fully remove the default Nouveau driver. Reboot, run the installer again, and your installation will be successful. |
There was a problem hiding this comment.
What is the Nouveau driver? Can we include a one-sentence explainer for what that is? Also, is it possible to remove the driver in the Before You Begin section so that it doesn't interrupt the flow here
|
|
||
| ### Ubuntu 18.10 | ||
| {{< note >}} | ||
| After running these commands, reboot your Linode from the [Cloud Manager](https://cloud.linode.com). The commands install missing kernel headers needed by NVIDIA's driver, and rebooting will ensure that they will be available. |
There was a problem hiding this comment.
@rsyracuse does this wording sound ok? it's a revision of the previous phrasing:
After running these commands, reboot your Linode from the [Cloud Manager](https://cloud.linode.com) to make sure you can use the missing kernel headers.
| {{< /note >}} | ||
|
|
||
| {{< disclosure-note "Why do NVIDIA's drivers need to be installed?" >}} | ||
| NVIDIA's proprietary drivers are closed-source, and Linode has chosen not to bundle them with the Linux images available in the Linode Cloud Manager. While some operating systems are packaged with the open source [Nouveau](https://nouveau.freedesktop.org/wiki/) driver, the proprietary driver will provide optimal performance for your GPU-accelerated applications. |
There was a problem hiding this comment.
@rsyracuse does this sound ok to you
Current edits have not retested installation on CentOS 7 and OpenSUSE
3 participants
No description provided.