Troubles running Rootpainter on a Blackwell Nvidia GPU

[Williamebf]

Hey I recently bought a new NVIDIA GeForce RTX 5070, and am trying to run the trainer server locally on Windows.

While I can run the trainer server on it, it runs really slowly with a batch size of 3. To get it to run at full speed with a Blackwell GPU, I think I need to run CUDA 12.8 and a Pytorch that supports CUDA 12.8. Is it possible to run rootpainter on an updated python, pytorch and Cuda?

This is the following warning I got when starting the trainer in powershell

NVIDIA GeForce RTX 5070 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 compute_37.
If you want to use the NVIDIA GeForce RTX 5070 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

warnings.warn(incompatible_device_warn.format(device_name, capability, " ".join(arch_list), device_name))

William

[Abe404]

Hey William,

It should be. If you just install the updated cuda version, the latest pytorch version and then the other dependencies and then start the RootPainter server is should work. Let me know if you run into any deprecated functions or other issues.

Also let me know if you need any help with this if you get stuck with it yourself.

When you say 'really slow' , what exactly are you referring to? and could you give me some numbers i.e time in seconds or minutes?

Although a 5070 is modern, 12GB of VRAM is still not optimal for RootPainter - but it should work OK.

Kind regards,
Abraham

[Williamebf]

Back when I set it up, it took about 2000 seconds for the first segmentation, and 1100-1300 seconds for the following about 8 ones, I then put the project on pause till you wrote this.

After reading this, I opened it up to try again, and without doing anything different, it was actually down to 300-500 seconds per segmentation, which is MUCH more manageable, since I can spend the time annotating. My GPU is quite new, maybe there was also some missing drivers or it using just in time compiler.

But many thanks clearing it up for me
William