Hi All,
I would like to know any solution to perform install DGX OS in multiple systems remotely. Solution something similar to Canonical MAAS being used to install cloud images (https://cloud-images.ubuntu.com/)
or any other alternative.
Hi @shan_8992 ,
NVIDIA Base Command Manager is what you’re looking for. See NVIDIA Base Command Manager - NVIDIA Docs for details and documentation. It ships with DGX OS images, and makes it trivial to install/configure/monitor multiple systems.
If you want a more roll-your-own solution, PXE Boot Setup — NVIDIA DGX OS 6 User Guide 1 documentation has the basics of how you’d configure things to answer the normal install questions - up to you to adapt that to MaaS or your favorite setup. I strongly prefer Base Command Manager though.
1 Like
NVIDIA BASE Command Manager we cannot choose DGX OS 5/6 iso as image, instead we need to prepare image by installing it in a node and further preparation of image right ?
Does “NVIDIA BASE Command Manager” uses iPXE to provision nodes?
If your NVIDIA Base Command Manager (BCM) license is for the DGX version, it includes DGX OS 5 and DGX OS 6 images out of the box. You can copy/customize those or use them as-is.
If you want to make your own DGX OS 6 image, start with the Ubuntu image and follow the procedure in Installing DGX Software on Ubuntu — NVIDIA DGX OS 6 User Guide 1 documentation inside of the BCM cm-chroot.
BCM provides a PXE server, and use iPXE on the client side.
Could you please share the reference manual link of “NVIDIA Base Command Manager (BCM) license is for the DGX version, it includes DGX OS 5 and DGX OS 6 images out of the box. You can copy/customize those or use them as-is.”
I couldn’t spot it out in admin-manual.pdf (brightcomputing.com)
As well as not mentioned about it in “Base Command Manager Administration” course. Rather it’s mentioned to build base image from head node or to prepare a base node.
Hi,
I have a BCM license and used that licence to download the BCM image from the Bright Computing website. While selecting the options, I have chosen an Ubuntu version, since there is no option for a DGX OS, but for the hardware I have chosen the DGX option. However, after the installation, i found out that the OS is still the Ubuntu, not the DGX OS. My questions:
- How can I download a BCM image which includes a DGX OS
- How can i verify that my licence is for the DGX Version
- How can i install a DGX software image to my Ubuntu based BCM (I do not want to install DGX OS to an existing Ubuntu image like you mentioned in your previous reply)
Thank you.
When you downloaded the ISO, did you check the box to include the DGX OS images?
When you do that, there will be DGX A100 and DGX H100 categories, corresponding DGX OS images, disklayouts, etc. so all you need to do is put the DGXs into the (e.g.,) dgxh100 category and it’ll get DGX OS installed.
Your description sounds like you missed checking that “DGX OS images” box when you downloaded the ISO.
ScottE
Thank you Scott, this was exactly what i needed. The DGX OS is not available for some Ubuntu versions, thats why I did not see it. By the way, can i install an DGX OS image to another Head node if i did not choose DGX OS while I was downloading the ISO image? For example, can i transfer an DGX OS image to another head node via USB for example?
I’d love to say “Yes, moving the image between headnodes is easy, just follow these steps…”, but it’s not something I do a lot. :-)
If the BCM versions are different between the two head nodes, you can use cm-create-image and basically tar/untar the directory to move it between head nodes. If the BCM versions are the same, you should be able to just copy the DGX softwareimage dir as-is and then set softwareimage to the new image.
I think there’s also a disklayout you’ll need to find and move over, which is burried somewhere like /cm/local/apps/cmd/etc/htdocs/disk-setup/ if memory serves. Check on the “new” head node where the disklayout for the category points to go to get the actual path and filename, then move it over and set it on the “old” head node to match.
Feel free to open a support ticket to get some help with this. Our support team does a great job (IMHO).
ScottE