Please update Dependency & Tools in the DGX Spark Porting Guide

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10
1

The table listed on Dependencies & Tools — NVIDIA DGX Spark Porting Guide has some out of date or inaccurate information. A couple of examples… cuPyNumeric is listed as not supported, yet it has already been demonstrated on Nvidia Developer’s YouTube Channel (https://www.youtube.com/watch?v=ghysPxEUZwU). Warp is also listed as not supported, while Warp 1.10 apparently is supported.

Also, the list seems to have significant gaps. Omniverse and Isaac do not appear on the list. Isaac Sim runs on the DGX Spark, as I’ve already set that up following Nvidia’s documentation. Isaac Sim is built on top of Omniverse, but the base Ominiverse plaform (Kit App Streaming template) may be near to being ready for DGX Spark. I don’t know the status about Isaac Lab.

cuOpt lis listed as not supported, but based on the specs for cuOpt, it looks like it would be supported on the DGX Spark.

This situation creates confusion as to which tools are supported on the DGX Spark. Grant it, this is a moving target as more packages are making their way to the Spark.

Please update this page. If and when there are updates, either make an announcement or just make a post on this Forum.

Thanks

4 Likes
2

I was able to install cupynumeric. Are you facing the same issue reported here?

3

No. Not facing the same issue here (getting cupynumeric to run on the Spark was not the focus of the post). The primary issue is that the info presented in Dependencies & Tools — NVIDIA DGX Spark Porting Guide is significantly out of date (last updated on Oct 16) having incorrect information, and also has gaps (says nothing about the status of Omniverse, Isaac Sim/Isaac Lab running on the Spark).

The post is a request for the Nvidia Team to update this table with reliable information.