Table of Contents
The NVIDIA aarch64 Driver requires a supported NVIDIA GPU to be present in the system. See Appendix A, Supported NVIDIA GPU Products for a list of supported GPUs.
The NVIDIA aarch64 driver supports any supported discrete NVIDIA GPU connected to one of the following processors:
NVIDIA Grace and later
NVIDIA Tegra K1 (64-bit version)
NVIDIA Tegra X1 or later
AppliedMicro X-Gene
Cavium ThunderX
* Required for hardware-accelerated video playback. See Appendix G, VDPAU Support for more information.
** Required for applications which use the Vulkan API.
If you need to build the NVIDIA kernel module:
| Software Element | Min Requirement | Check With... |
|---|---|---|
| binutils | 2.9.5 | size
--version |
| GNU make | 3.77 | make
--version |
| gcc | 2.91.66 | gcc
--version |
All official stable kernel releases from 4.15 and up are supported; pre-release versions, such as 4.19-rc1, are not supported. The Linux kernel can be downloaded from http://www.kernel.org or one of its mirrors.
binutils and gcc can be retrieved from http://www.gnu.org or one of its mirrors.
Sometimes very recent X server versions are not supported immediately following release, but we aim to support all new versions as soon as possible. Support is not added for new X server versions until after the video driver ABI is frozen, which usually happens at the release candidate stage. Prerelease versions that are not release candidates, such as "1.10.99.1", are not supported.
If you are setting up the X Window System for the first time, it is often easier to begin with one of the open source drivers that ships with X.Org (either "vga", "vesa", or "fbdev"). Once your system is operating properly with the open source driver, you may then switch to the NVIDIA driver.
These software packages may also be available through your Linux distributor.
A CPU+GPU coherent memory platform is a platform where the CPU and GPU are connected by a memory-coherent link. Grace Hopper is one such platform where the CPU and GPU are connected by a memory-coherent NVIDIA NVLink Chip-2-Chip (C2C) link.
These platforms require the Open Linux Kernel Modules. The installer by default selects the Open Linux Kernel Modules if it detects the presence of these platforms during installation.
These platforms require the Linux kernel "online_movable" mode to be enabled in order for GPU memory to be brought online when the driver adds GPU memory to the kernel. User-mode driver components, such as libcuda.so.1, will enable the Linux kernel auto online mode online_movable, by writing "online_movable" to the file /sys/devices/system/memory/auto_online_blocks, when it has sufficient privileges. When it doesn't have sufficient privileges, it invokes the nvidia-modprobe setuid tool to apply the setting. The system administrator is required to configure this manually if the driver is installed without nvidia-modprobe. The system administrator should also ensure auto online mode is not overridden by other components in the system.