PerceptronAI

/

Isaac-0.5

Like 56

Follow

Perceptron AI 106

Hi, I want to run Isaac-0.5 on an NVIDIA GB10 with 128 GB unified memory and a Linux ARM64 host.

My use case is image/video-to-text inference. I noticed that require_qualified_runtime() in modeling_isaac05.py explicitly requires CUDA 12.8, compute capability 9.0, and a GPU name starting with NVIDIA H100:

Could you clarify:

1. Is the H100 restriction a limit of the tested configuration, or do required GPU operations depend on H100 hardware?
2. Can image/video-to-text inference run on GB10? If so, what code changes and dependency versions are needed?
3. Is there a minimal image or video inference example for this checkpoint?