How do I use from X86_64 to arm64 architecture? (LabVIEW installation using cuda)

[y.menjour1]

Hello Everyone,

I need a technical assistance.

We are currently working on an embedded system with LabVIEW (X86_64). We’re interested in the Jetson for advanced computing purposes, but our current problem is that LabVIEW Linux is designed for X86_64 architectures (Intel - AMD), and therefore can’t run natively on the Arm64 architecture your Jetson devices are equipped with.

Our interest in the Jetson stems from the fact that it’s an NVIDIA GPU for edge computing, but we need to be able to work with LabVIEW64 on this computer.

Currently, as a solution, we’re considering using a hypervisor to install Ubuntu Linux (x86_64) or standard Windows (x86_64) on the Nvidia Arm Linux. We’ve done this on a Raspberry Pi, which has an Arm processor, and it works (install Windows (x86_64) and LabVIEW).

Our problem is that Windows or Linux installed on top of Nvidia Arm Linux using the hypervisor (KVM with QEMU) doesn’t recognize the GPU device, preventing us from using it properly. This is a configuration problem for which we haven’t yet found a solution. → we need help

Obviously, we’re not the first ones to want direct access to the GPU from an X86_64 operating system.
Who can help me ?

Thank you and I look forward to hearing from you.

[linuxdev]

I probably can’t help, but I can suggest something related to this: Most third party GPU applications tend to look for NVIDIA GPUs using “nvidia-smi”. That program is for query of a PCIe bus discrete GPU (dGPU). The GPU on a Jetson is integrated directly to the memory controller (an iGPU). This means that any program which depends on the GPU being a dGPU on PCI will fail to find the GPU. Quite possibly your hypervisor is working just fine, but the lack of nvidia-smi is making it look like there is no GPU.

I happen to wonder, if you were to try to use this program through strace (which shows system calls), perhaps it looks for “nvidia-smi”? Don’t know, but if it does, then this would confirm why the GPU is invisible.

I don’t know of alternate methods to tell a program where a GPU is located. This is why Jetsons come with their own version of CUDA and video drivers.