Aerial SDK 23-4

This section provides a description of supported delta features and capabilities in the current minor release.

Current Version: 23-4

NVIDIA is the global leader in GPU accelerated computing and is enabling a fully cloud-native virtual 5G RAN solution to support a wide range of next-generation edge AI and RAN services using commercial off-the-shelf (COTS) servers.

NVIDIA Aerial SDK provides 5G wireless radio access network solution, with inline L1 GPU acceleration for 5G NR PHY processing. It supports a full stack framework for a gNB integration L2/L3 (MAC, RLC, PDCP), along with manageability and orchestration. Aerial SDK also supports non 5G signal processing use cases. The NVIDIA Aerial SDK package simplifies building programmable and scalable software defined 5G RAN PHY with the following components:

  • CUDA Baseband (cuBB): The NVIDIA cuBB SDK provides GPU accelerated 5G signal processing pipeline including cuPHY for Layer 1 5G PHY, delivering unprecedented throughput and efficiency by keeping all physical layer processing within the high-performance GPU memory

  • DOCA GPUNetIO: A new library to enable GPU-initiated communications so a CUDA kernel can invoke the CUDA device functions in the DOCA GPUNetIO library to instruct the GPUDirect-capable network cards (Mellanox BlueField-3, ConnectX-6 DX or AX800/A100X converged card) to send or receive packets.


  • Support Grace Hopper platform

  • 20 light load cells with BFP9 (pattern 61)

  • Support FOSS and upstreamed drivers

  • Cutover to DMA-buf, Open-Source GPU Kernel Modules (OpenRM) and inbox driver

  • Support Multi-L2 with single cuphycontroller per GPU

  • Support M-plane Hybrid Mode

  • Using ORAN YANG data models to pass RU capabilities, C/U–plane transport, U-plane config to L1

  • 32T32R Massive MIMO (Direct Access / DA release)

  • 1 peak 100MHz cell with 8DL and 4UL layers

  • Cat-B O-RU with beamforming method 2 (BFm2)

  • SRS / reciprocity-based dynamic beamforming weights, both UL and DL

  • 4T4R new feature support:

  • SRS in S-slot

  • NB 10MHz cells and multi symbol CORESET

  • 4T4R multicell improvements:

  • Allow multiple cell operation without issuing dummy config.req

  • Deliver multicell capacity with UCI OCC disabled

  • HARQ UCI.indication latency improvement


  • Support for 32T32R Massive MIMO (Early Access):

    • 1 peak 100MHz cell with 8DL and 4UL layers

    • SRS / reciprocity-based dynamic beamforming weights, both UL and DL

  • New features supported for 4T4R:

    • Removed PDU ordering limitations

    • Reduced max startup time to less than 30 sec

    • Improved robustness and error handling

  • O-RAN fronthaul features:

    • 32T32R / Massive MIMO Cat-B O-RU with beamforming method 2 (BFm2)

  • AX800 converged accelerator card features:

    • Simultaneous use of both Ethernet ports for fronthaul

  • Multicell capacity improvements:

    • 23-3 Traffic model (23-3TM): 4 UL streams from RU, 7 beam-ids, BFP14 FH compression

    • AX800 with Dell R750 supports 15 peak and 16 average cells of 4T4R @ 40MHz w/ 23-3TM

Previous Aerial Product Brief
Next Aerial SDK Layer 1
© Copyright 2022-2023, NVIDIA.. Last updated on Apr 20, 2024.