Effects_demo.exe getSupportedDevices failed with NVAFX_STATUS_FAILED on RTX 3090

AI & Data Science Deep Learning (Training & Inference) AI for Media
1

Hello NVIDIA team and community,

I’m currently evaluating the NVIDIA Maxine AFX SDK (version 2.1.0.9) on a Windows 10 machine. I’ve successfully compiled the effects_demo sample from the SDK, but when I run it, it fails during GPU device enumeration with the following error:

text

[E] Denoiser Version 1 getSupportedDevices failed .
[INFO] Could not get number of supported devices with error: NVAFX_STATUS_FAILED

This happens regardless of the effect I try (e.g., denoiser or dereverb_denoiser).

System details:

  • GPU: NVIDIA GeForce RTX 3090 (Ampere architecture)

  • NVIDIA Driver: 610.62 (WDDM mode)

  • CUDA Runtime: 12.8 (installed as standalone runtime only; no development tools)

  • TensorRT: 10.7.0.23 (Windows ZIP package)

  • Maxine AFX SDK: 2.1.0.9

  • Visual Studio: 2019 Community (MSVC 14.29)

  • OS: Windows 10 Pro 64‑bit (build 19045)

What I’ve already tried:

  1. Set environment variables:

    • AFX_SDK_ROOT (pointing to the SDK root)

    • CUDA_PATH (pointing to CUDA 12.8 installation)

    • CUDA_VISIBLE_DEVICES=0

    • TENSORRT_ROOT (pointing to TensorRT 10.7.0 directory)

    • Added %CUDA_PATH%\bin and %TENSORRT_ROOT%\lib to PATH

  2. Copied all required DLLs into the same folder as effects_demo.exe:

    • All CUDA 12.8 *64*.dll files from C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\bin

    • All TensorRT 10.7.0 DLLs from TensorRT-10.7.0.23\lib

    • Also copied NVAudioEffects.dll and nvafxdenoiser.dll (and its dependencies)

  3. Verified CUDA installation with deviceQuery.exe – it successfully detects the RTX 3090 and reports Result = PASS.

  4. Checked that the model files are present:

    • nvafxdenoiser/models/ampere/ contains the expected model files.

  5. Tried switching the configuration to dereverb_denoiser – same error occurs.

Additional diagnostics:

  • Output of nvidia-smi shows the GPU is recognised and driver version is correct.

  • dumpbin /dependents nvafxdenoiser.dll lists all dependencies, and I’ve confirmed they are present.

  • The config.txt used is:

    text

    effect denoiser
model ./features/nvafxdenoiser
input_wav D:/input.wav
output_wav D:/output.wav
real_time 0
intensity_ratio 1.0
input_sample_rate 48000
output_sample_rate 48000
enable_vad 0

Question:

What could be causing NvAFX_GetSupportedDevices to return NVAFX_STATUS_FAILED even though deviceQuery passes successfully?

  • Is there a specific version of TensorRT that is strictly required for Maxine SDK 2.1.0.9?

  • Does the SDK require any additional CUDA components (e.g., cuDNN, NVML) that I might have missed?

  • Could the issue be related to the WDDM driver mode vs. TCC mode? Or any other environment settings I should check?

Any help would be greatly appreciated. I can provide full logs or any other diagnostic output if needed.

Thank you!