CV-CUDA可以在 arm架构下运行么？例如 Orin或者更早的NX？

[cmffire]

如题。有相关例子么？

[kgnandanwar]

CV-CUDA on ARM Architecture (Jetson Orin/NX)

Yes, CV-CUDA runs on ARM, including Jetson devices.

Supported ARM Platforms

Jetson Orin (AGX Orin, IGX Orin):

• ✅ Fully supported
• Use aarch64_cu12 build
• Requires JetPack 6.0 DP with driver r535
• Python 3.10

Older Jetson (Xavier NX, Nano, etc.):

• ❌ Not officially supported
• Requires compute capability ≥ SM7.5 (Orin is SM8.7)
• Xavier NX is SM7.2 - below minimum requirement

Installation on Jetson Orin

Option 1: Python wheel (easiest)

# On Jetson Orin with JetPack 6.0
pip install cvcuda-cu12

Option 2: Build from source

git clone https://github.com/CVCUDA/CV-CUDA.git
cd CV-CUDA
./ci/build.sh

Platform Support Summary (from README.md)

Platform	CUDA	Hardware	Driver	Status
Jetson AGX Orin	12.2	SM8.7 (Ampere)	JetPack 6.0, r535	✅ Supported
Jetson Xavier NX	-	SM7.2 (Volta)	-	❌ Below SM7.5 requirement
Jetson Nano	-	SM5.3 (Maxwell)	-	❌ Below SM7.5 requirement

Quick Test on Jetson

import cvcuda
import torch

# Create a test image
img = cvcuda.Image((640, 480), cvcuda.Format.RGB8)

# Or wrap existing GPU tensor
gpu_tensor = torch.zeros((480, 640, 3), device="cuda", dtype=torch.uint8)
img = cvcuda.as_image(gpu_tensor, format=cvcuda.Format.RGB8)

print(f"Image created: {img.size}")

References

• README.md - Full platform compatibility matrix (line 75)
• Pre-built wheels: cvcuda-cu12 on PyPI
• GitHub releases have explicit "Jetson" archives for aarch64 builds

Bottom Line

Orin: Yes, works out of the box.
NX/Nano: No, compute capability too old (need SM7.5+, they have SM7.2/5.3).