RTX 5090 Support and Performance

General
1

Hey AddaxAI team,

I have a question regarding RTX 5090 support. During initialization, there’s a warning regarding the PyTorch install.

C:\Users\wildlife\AddaxAI_files\envs\env-base\lib\site-packages\torch\cuda_init_.py:209: UserWarning:
NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90 compute_37.
If you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at Get Started

Contrary to the warning, it seems like the actual process is using the GPU, albeit at a lower capacity than I would anticipate. Does the “Forward-compatibility issue detected, patching” message indicate that AddaxAI is patching a more current version of PyTorch on the fly?

image
image1029×976 232 KB

I can run MegaDetector v1000 redwood in a custom script that uses Python 3.11.14 and torch-2.10.0+cu128 but my techs prefer working in the the Timelapse/AddaxAI GUI. That script sees processing rates around 70 images/s which is closer to what I’d expect than the ~10 images/s in AddaxAI. Is the performance being throttled or is it falling back on the CPU and reporting GPU?

Best,
Alex

2

Hi Alex @spengler ,

This is a structural issue with AddaxAI’s current PyTorch environment. It’s too old for your new GPU. I will be updating this, but that will only be done in the next major version, and that can still take some time (a few weeks?). I need to update the analysis environment, and that takes a bit of testing. More info here: Model Warning on running with GPU - #16 by petervanlunteren

Running MD via its package gives you full control over all the settings and package versions, which is what you’ve already succesfully shown. Can you elaborate on the reason that your techs would prefer you using AddaxAI/Timlapse. Basically, what AddaxAI is doing for timelapse is just running the MD scripts you already succesfully done, and renaming the file to timelapse_recognition_file.json, then import via Timelapse menu. Thats it.

AddaxAI does a bit more if you also want classification predictions, but it seems like you just want MD predictions. Is that right? If so, I would skip AddaxAI all together and wait for the update that fixes the RTX 50** issue.

Does that make sense?

Cheers,

Peter

Sorry! Hope you understand.