This document describes the procedure for installing the official Nvidia DGX software stack in a Bright RHEL8 software image. The instructions in this document target the DGX A100, but the same procedure can be used for other DGX systems such as DGX-1, DGX-2 and DGX Station.
Step 1: Prepare a copy of the software image
This example uses the default software image as the base image. Any other software image can be used for this purpose.
Clone the base software image to an image called dgxa100-image.
# Create a clone of the default Bright software image, add raid modules to the new image if necessary
$ cmsh
softwareimage
clone default-image dgxa100-image
kernelmodules
list | grep raid
add raid0
..
add raid1
..
..
commit
# Important: Remove the Bright cuda-dcgm/cuda-driver packages(if installed)
dnf --installroot /cm/images/dgxa100-image remove cuda-dcgm cuda-driverStep 2: Assign new image to the node(s) or category
Define a new node. say dgx-a100 and assign the prepared dgx-image to it. After the initial ramdisk has been generated, provision the DGX node.
In the example below we will set the software image for an individual node, but if you have many DGX nodes it makes more sense to create a node category, make your DGX nodes part of this category, and set the software image for the category (which will then be inherited by all nodes in the category).
$ cmsh
device use dgx-a100;
set softwareimage dgxa100-image;
commit
# Wait for ramdisk to be generated
Step 3: Provision DGX node
Provision the DGX node dgx-a100 (see Bright documentation for details).
Step 4: Install DGX software stack
The steps in this section must be performed on the DGX node dgx-a100 provisioned in Step 3. The URLs, names of the repositories and driver versions in this section are subject to change. Please refer to the official DGX RHEL8 Install Guide for latest information on DGX repositories and additional RHEL8 repositories that must be enabled.
Enable repositories
# Enable additional RHEL8 repositories
subscription-manager repos --enable=rhel-8-for-x86_64-appstream-rpms
subscription-manager repos --enable=rhel-8-for-x86_64-baseos-rpms
subscription-manager repos --enable=codeready-builder-for-rhel-8-x86_64-rpmsPerform OS update
dnf update --nobestInstall DGX tools and Nvidia drivers
# Enable DGX repositories
dnf install -y https://repo.download.nvidia.com/baseos/el/el-files/8/nvidia-repo-setup-21.06-1.el8.x86_64.rpm
# Install DGX tools
dnf groupinstall -y 'DGX A100 Configurations'
# Install Nvidia drivers
dnf module install -y nvidia-driver:450/{fm,src}
dnf install -y nv-persistence-mode nvidia-fm-enableOptional package installation
Installation of CUDA toolkit, NVIDIA Collectives Communication Library (NCCL) Runtime, CUDA Deep Neural Networks (cuDNN) Library Runtime, TensorRT or NVIDIA GPUDirect Storage (GDS) is optional, those can be installed by following the relevant section at the official DGX Software installation guide for RHEL 8.
Step 5: Grab image from DGX node
Sync the disk image from the running DGX node onto the software image dgxa100-image on the head node.
$ cmsh
device use dgx-a100
grabimage -wAs a result of the OS update performed in Step 4, a newer version of the kernel could have been installed, and hence it is required to point the software image to use the new kernel.
$ cmsh
softwareimage use dgxa100-image
set kernelversion <TAB> # select the latest kernel version
commit
# Wait for ramdisk to be generatedStep 6: Provision node(s) with new image
The image is now ready to boot any number of DGX nodes. If you have created a Bright node category, you can configure any node to be part of that category by setting its category property. This will make the nodes inherit the softwareimage setting that you have defined for the category. Alternatively, you can configure the softwareimage property for individual nodes.
When nodes are powered on, they will be imaged using the Bright software image to which we have added the DGX software stack. The following command can be used to verify that the Nvidia drivers and services are working as expected:
[root@dgx-a100 ~]# dcgmi discovery -l
8 GPUs found.
+--------+----------------------------------------------------------------------+
| GPU ID | Device Information |
+--------+----------------------------------------------------------------------+
| 0 | Name: A100-SXM4-40GB |
| | PCI Bus ID: 00000000:07:00.0 |
| | Device UUID: GPU-8cda93c7-c7b3-5bb1-c5ae-d18f14ec21b5 |
+--------+----------------------------------------------------------------------+
| 1 | Name: A100-SXM4-40GB |
| | PCI Bus ID: 00000000:0F:00.0 |
| | Device UUID: GPU-4b1d7230-8739-451f-4143-19e35cc34e3b |
+--------+----------------------------------------------------------------------+
| 2 | Name: A100-SXM4-40GB |
| | PCI Bus ID: 00000000:47:00.0 |
| | Device UUID: GPU-7d4ef5ea-b2f2-d509-a507-a20f2e656dc7 |
+--------+----------------------------------------------------------------------+
| 3 | Name: A100-SXM4-40GB |
| | PCI Bus ID: 00000000:4E:00.0 |
| | Device UUID: GPU-a41e5efb-3df9-9846-8ab7-4adb39f0467f |
+--------+----------------------------------------------------------------------+
| 4 | Name: A100-SXM4-40GB |
| | PCI Bus ID: 00000000:87:00.0 |
| | Device UUID: GPU-900945ac-42c9-b6df-e263-9f1477fab578 |
+--------+----------------------------------------------------------------------+
| 5 | Name: A100-SXM4-40GB |
| | PCI Bus ID: 00000000:90:00.0 |
| | Device UUID: GPU-5f5041fd-791b-de04-d7fd-510e4c78de3f |
+--------+----------------------------------------------------------------------+
| 6 | Name: A100-SXM4-40GB |
| | PCI Bus ID: 00000000:B7:00.0 |
| | Device UUID: GPU-a558bb3a-8f95-0487-2b73-46ec419410bf |
+--------+----------------------------------------------------------------------+
| 7 | Name: A100-SXM4-40GB |
| | PCI Bus ID: 00000000:BD:00.0 |
| | Device UUID: GPU-d2839297-b553-cea3-e2bf-c76eff99279b |
+--------+----------------------------------------------------------------------+
6 NvSwitches found.
+-----------+
| Switch ID |
+-----------+
| 9 |
| 13 |
| 11 |
| 10 |
| 8 |
| 12 |
+-----------+Step 7: Install Mellanox OFED (Optional)
This can be performed on any one of the DGX nodes that was provisioned in Step 6. Follow instructions in the DGX RHEL8 Install Guide for installing the Mellanox OFED software stack on a DGX server.
Step 8: Install nv_peer_mem kernel module (Optional)
In order to use GPUDirect over Infiniband, it is necessary to install the nv_peer_mem kernel module. After installing the Mellanox OFED stack (see previous section), the nv_peer_mem can be built on the DGX node.
dnf install dkms nvidia-peer-memory-dkmsThe post-install scriptlet of the nvidia-peer-memory-dkms package will build the nv_peer_mem module, and also attempt to load it.
Note: It is expected that the module will fail to load if the OFED drivers from RHEL are still loaded. The error message can be ignored at this stage. The module will load correctly after Step 9 has been completed.
Step 9: Finalize setup and grab image from DGX node
If Step 7 or Step 8 was performed, the changes need to be synchronized back to the software image and the initial ramdisk must be recreated. This can be done with cmsh using:
$ cmsh
device use dgx-a100;
grabimage -wIf the nv_peer_mem module was installed (Step 8), then schedule the nv_peer_mem kernel module to be loaded automatically when the system boots. Doing this by adding it to the kernel modules list for the software image is not recommended, because the nv_peer_mem kernel module relies on other kernel modules that may not be loadable from the initrd. Write the following file in the software image on the head node:
echo "nv_peer_mem" > /cm/images/dgxa100-image/etc/modules-load.d/nv_peer_mem.confRe-create initial ramdisk for the software image dgxa100-image
$ cmsh
softwareimage use dgxa100-image
createramdisk
# Wait for ramdisk to be generatedWhen the DGX nodes that are set to use the sofware image are rebooted, they should come up with the correct Mellanox OFED kernel modules and nv_peer_mem kernel module loaded. This can be verified as follows:
[root@dgx-a100 ~]# lsmod | grep nv_peer_mem
nv_peer_mem 16384 0
nvidia 19378176 449 nvidia_uvm,nv_peer_mem,nvidia_modeset
ib_core 425984 9 rdma_cm,ib_ipoib,nv_peer_mem,iw_cm,ib_umad,rdma_ucm,ib_uverbs,mlx5_ib,ib_cm