This is the first of six (1/6) posts serializing the FAQ report from the most recent NVIDIA 6G Developer Day (sessions playback here). All questions and answers are directly from attendees on the day and the answers provided to them.
The Sections are:
1/6: Introducing NVIDIA 6G Research & Development Portfolio
2/6: Prototyping 6G with the NVIDIA Sionna Research Kit
3/6: Designing a Modular 6G System Using NVIDIA Aerial Framework
4/6: Aerial Framework – Performant GPU-Accelerated Code From Python
5/6: Optimizing 6G Performance With CUDA and NVIDIA DOCA
6/6: Simulating 6G with NVIDIA Aerial Omniverse Digital Twin
The full FAQ report is here:
2025_6GDD_NVIDIA AI Aerial_FAQ.pdf (374.9 KB)
Introduction to NVIDIA 6G Research & Development Portfolio
Q 1.1: NVIDIA states that “6G will be AI-Native”. What is the single biggest technical challenge NVIDIA foresees in making a complete shift from the “Evolutionary Approach” of prior generations (2G-5G) to this “Revolutionary Approach” (AI-Native 5G, 6G)?
A: The challenge is less technical and more about companies’ attitudes towards making AI foundational to 6G. 3GPP has a mix of attitudes and are working through the technical aspects of going AI-native, such as LCM (Lifecycle Management). Besides working to make 6G AI-native in 3GPP, NVIDIA is working with companies in the AI-RAN Alliance to build AI on top of what 3GPP defines.
Q 1.2: What is NVIDIA 6G Developer Program?
A: The NVIDIA 6G Developer Program equips developers and researchers with the learning resources and support to accelerate 6G development using the NVIDIA AI Aerial platform. Members receive updates on new software releases, NVIDIA 6G events and technical training for NVIDIA AI Aerial.
Q 1.3: In the AERIAL documentation, we read a statement: “Aerial support of AX800, A100X, A100, CX6-DX has reached End of Life.” Does this mean we cannot use version 25.3 of AERIAL to achieve physical layer acceleration for the Openairinterface project on A100 GPUs.
A: EOL means NVIDIA is unable to provide support for user problems when using these components. All converged card and devkit (e.g. A100+CX6_DX) are no longer available from CICD and developer cluster, so it is not possible to confirm that they would work on the latest Aerial releases.
Q 1.4: With your simulation solutions, how do you ensure ease of fine tuning to specific/custom environments like a factory floor?
A: Sionna RT and AODT are two ways in which NVIDIA is enabling site-specific optimization. By modeling the environment with high fidelity and accuracy, data collected from specific sites can be used to train and tune implementations.
Q 1.5: What is NVIDIA Aerial RAN Computer?
A: NVIDIA Aerial RAN Computer (often referred to as ARC) is a high-performance, GPU-accelerated server platform purpose-built to run NVIDIA’s CUDA-Accelerated RAN software for vRAN/AI-RAN. It integrates CPU, GPU, fronthaul networking, and timing to host 5G/6G baseband (L1/L2) and AI workloads on the same system, enabling O-RAN–compliant, software-defined, AI-native RAN deployments.
Q 1.6: A revolutionary approach to 6G feels good, but what about operators with the investments they have done on the evolutionary approach. Is anything planned for them?
A: As with every wireless generation, the introduction of 6G will be alongside existing 5G and earlier deployments. 3GPP is defining technologies that account for such mixed spectrum usage, enabling operators to continue to use existing spectrum with legacy G’s while deploying 6G on new spectrum. Over time, operators can migrate spectrum from legacy G’s to 6G. While 6G aims to be AI-native, the performance of existing G’s can be improved by applying AI to vendor-proprietary, standards-compliant implementations. This will extract further value from operators’ investments.
Q 1.7: When will NVIDIA deploy AI RAN? Will you only be going with the SW or will NVIDIA hardware replace the legacy RUs?
A: NVIDIA’s Aerial RAN offering is focused on O-DUs and includes HW and SW components. RAN vendors who adopt Aerial will use NVIDIA HW and base SW components like CUDA and DOCA to build their own HW/SW platforms. Aerial SW implements a complete L1 and a part of L2 from the 3GPP stack and allows RAN vendors to use as little or as much of NVIDIA’s code for their commercial offerings.
Q 1.8: What about HW power consumption vs SoC based approach on the Baseband part?
A: NVIDIA has done extensive benchmarking with Aerial HW and SW and they are highly competitive in power and cost to ASICs/SoCs.
Q 1.9: What about footprint of NVIDIA’s solution inside baseband cabinet?
A: NVIDIA’s solutions are designed for scalability across D-RAN and C-RAN topologies and are being built to meet stringent environmental, size, weight, power and cost requirements even for D-RAN deployments. They can even scale down to compact, private network requirements.
Q 1.10: Does NVIDIA have any products for 5G core part?
A: NVIDIA is focused on RAN O-DU products for now.
Q 1.11: What is NVIDIA’s current and future plan for the Sionna project, the 6G Developer Program, and the deployment of AI-native technologies, including the technical challenges and hardware considerations involved?
A: Sionna is a very active project with continued enhancements tailored for 6G research and quick prototyping. The 6GDP continues as a key program to announce Sionna/Aerial product launches, events, training, etc. 6GDP forums are now hosted in GitHub too, since many Aerial components are open source. AI-native 6G is being pursued in 3GPP and in the AI-RAN Alliance. Stay tuned for exciting product announcements, technical papers and blogs, demos and such by NVIDIA and partners.
Q 1.12: For running Aerial stack which one is compatible, jetson or DGX spark?
A: DGX Spark is a great choice for price, availability and capabilities for running Aerial. Aerial is not supported on Jetson.
Q 1.13: Is NVIDIA working on telecom RAN software or doing testing with the vendors, to verify RAN software? Or is the testing/verification totally based on simulations?
A: NVIDIA offers solutions that cover building/training, simulating and deploying RAN HW and SW. Aerial CUDA-Accelerated RAN SW and Aerial Testbed (previously known as ARC-OTA) are commercial-grade implementations of O-DUs that have been extensively tested by NVIDIA, RAN vendors, Mobile Network Operators and researchers. An example is testing done by Softbank and 1Finity.
Q 1.14: What are the hardware and software requirements for running the Aerial stack, including compatibility with different devices, chipsets, and access options?
A: NVIDIA AI Aerial is a portfolio of accelerated hardware, software libraries, and tools for designing, simulating, training, and deploying AI-native wireless networks, especially AI-RAN for 5G and 6G. AI Aerial software include open-source software libraries and tools for building, training, simulating, and deploying AI-native wireless networks. NVIDIA AI Aerial runs on a family of GPU-accelerated telecom platforms plus standard O-RAN radios and fronthaul. Supported systems include Aerial Testbed and Aerial RAN Computer. They connect to O-RAN 7.2x-compliant O-RUs from multiple vendors via qualified fronthaul switches and PTP GrandMaster timing.