Future GPU support roadmap for NVIDIA AI Aerial (Grace-Hopper vs Grace-Blackwell)

Accelerated Computing Aerial Forum
, , ,
1

Hello,

I am currently preparing a server infrastructure for NVIDIA AI Aerial, with the goal of actively using components such as cuBB, Sionna, pyAerial, AODT, etc.

According to the current documentation, AI Aerial supports GPUs up to GH200.

However, from a procurement and supply perspective, GH200 systems are becoming increasingly difficult to source, and it appears that NVIDIA’s data center production focus is shifting toward the GB-series platforms.

Given this situation, I would like to ask the following questions:

  1. Is there any plan to deprecate or discontinue GH200 support in NVIDIA AI Aerial in the foreseeable future?
  2. Are there plans to add official support for Blackwell-based GPUs such as GB200 or GB300 platforms in AI Aerial?
  3. From a long-term operational standpoint (both in terms of software compatibility and vendor support), given that both GH- and GB- series GPUs involve a substantial upfront cost, we would appreciate guidance on which platform NVIDIA would recommend for new AI Aerial deployments.

Since these GPUs represent a significant investment, ensuring long-term compatibility, driver and framework support, and alignment with the AI Aerial roadmap is a critical factor in our purchasing decision.

Any guidance on recommended GPU platforms or insight into the general direction of AI Aerial hardware support would be greatly appreciated.

Thank you in advance for your help.

2

At this time we plan to continue support of these packages on GH200 and have no immediate plan to support GBx00-based platforms. However, we will shortly introduce support on DGX Spark.

The initial release of Aerial Testbed (aka ARC-OTA) on the DGX Spark will support 1 peak cell and 5 OTA UEs.

The Spark is available easily from NVIDIA and multiple OEMs, such as Dell, HP, ASUS and others.

If you let us know your use cases we can guide you on which platform will be best for you.

3

@eobiodu We are very excited to hear that support for DGX Spark is coming soon.

4

Thanks for the clarification, DGX Spark support sounds really promising.

From a developer perspective, hardware accessibility plays a big role in how easily we can experiment with tools like Sionna and Aerial.

Curious to see how DGX Spark changes the entry barrier for running these simulations in more practical setups.