Blackwell 6000 Installation

[adamsokacz]

I've tried some initial troubleshooting but get the same fusedsim error on wsl and docker.

• tried building torch before running the script
• tried building fusedssim manually before running the script (script forces rebuild i think so i tried removing that but still no luck :( )

[blackwell-6000-install.txt](https://github.com/user-attachments/files/23303971/blackwell-6000-install.txt)

[adamsokacz]

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.95.02 Driver Version: 581.42 CUDA Version: 13.0 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA RTX PRO 6000 Blac... On | 00000000:01:00.0 Off | Off |
| 30% 35C P1 61W / 600W | 4271MiB / 97887MiB | 8% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| No running processes found |
+-----------------------------------------------------------------------------------------+

[chenyuting-netizen]

Same problem on Blackwell RTX5090 GPU

[VladSarm]

Also problem with build on RTX 5090

I am doing with
docker build --build-arg CUDA_VERSION=12.8.1 -t 3dgrut:cuda128 .

Here is an error

Collecting git+https://github.com/rahul-goel/fused-ssim@1272e21a282342e89537159e4bad508b19b34157 (from -r requirements.txt (line 23))
Cloning https://github.com/rahul-goel/fused-ssim (to revision 1272e21a282342e89537159e4bad508b19b34157) to /tmp/pip-req-build-k6g4pcus
...
Getting requirements to build wheel: started
Getting requirements to build wheel: finished with status 'error'
error: subprocess-exited-with-error

× Getting requirements to build wheel did not run successfully.
│ exit code: 1
╰─> [20 lines of output]
Traceback (most recent call last):
File "/opt/conda/envs/3dgrut/lib/python3.11/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 389, in
main()
File "/opt/conda/envs/3dgrut/lib/python3.11/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 373, in main
json_out["return_val"] = hook(**hook_input["kwargs"])
File "/opt/conda/envs/3dgrut/lib/python3.11/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 143, in get_requires_for_build_wheel
return hook(config_settings)
File "/tmp/pip-build-env-udr43q56/overlay/lib/python3.11/site-packages/setuptools/build_meta.py", line 331, in get_requires_for_build_wheel
return self._get_build_requires(config_settings, requirements=[])
File "/tmp/pip-build-env-udr43q56/overlay/lib/python3.11/site-packages/setuptools/build_meta.py", line 301, in _get_build_requires
self.run_setup()
File "/tmp/pip-build-env-udr43q56/overlay/lib/python3.11/site-packages/setuptools/build_meta.py", line 512, in run_setup
super().run_setup(setup_script=setup_script)
File "/tmp/pip-build-env-udr43q56/overlay/lib/python3.11/site-packages/setuptools/build_meta.py", line 317, in run_setup
exec(code, locals())
File "", line 2, in
ModuleNotFoundError: No module named 'torch'
[end of output]

note: This error originates from a subprocess, and is likely not a problem with pip.
ERROR: Failed to build 'git+https://github.com/rahul-goel/fused-ssim@1272e21a282342e89537159e4bad508b19b34157' when getting requirements to build wheel

[chenyuting-netizen]

@VladSarm I think I found the solution and successfully run the code.

modify the install_env.sh:

pip install -r requirements.txt
pip install -e .

to

pip install -r requirements.txt --no-build-isolation
pip install -e . --no-build-isolation

---

At the training, there are also the bug:

File "~/3dgrut/threedgrt_tracer/tracer.py", line 44, in load_3dgrt_plugin
import lib3dgrt_cc as tdgrt  # type: ignore
^^^^^^^^^^^^^^^^^^^^^^^^^^^
ModuleNotFoundError: No module named 'lib3dgrt_cc'

It cause the execute file store at wrong place.
the execute file cloud be found at /.cache/torch_extensions/py311_cu128/

So I manually copy the missing file. e.g:

cp ~/.cache/torch_extensions/py311_cu128/lib3dgut_cc/lib3dgut_cc.so ~/3dgrut/threedgut_tracer/lib3dgut_cc.so

Same as lib3dgrt_cc.so, libplayground_cc.so, lib3dgrut_gui_cc.so should put at correct folder.

Hope this work for you.

[wakuwu]

Same problem on Blackwell RTX4090 GPU

738.4 Running command git checkout -q 1272e21a282342e89537159e4bad508b19b34157
740.0 Resolved https://github.com/rahul-goel/fused-ssim to commit 1272e21a282342e89537159e4bad508b19b34157
740.0 Installing build dependencies: started
742.5 Installing build dependencies: finished with status 'done'
742.5 Getting requirements to build wheel: started
742.6 Getting requirements to build wheel: finished with status 'error'
742.6 error: subprocess-exited-with-error
742.6
742.6 × Getting requirements to build wheel did not run successfully.
742.6 │ exit code: 1
742.6 ╰─> [20 lines of output]
742.6 Traceback (most recent call last):
742.6 File "/opt/conda/envs/3dgrut/lib/python3.11/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 389, in
742.6 main()
742.6 File "/opt/conda/envs/3dgrut/lib/python3.11/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 373, in main
742.6 json_out["return_val"] = hook(**hook_input["kwargs"])
742.6 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
742.6 File "/opt/conda/envs/3dgrut/lib/python3.11/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 143, in get_requires_for_build_wheel
742.6 return hook(config_settings)
742.6 ^^^^^^^^^^^^^^^^^^^^^
742.6 File "/tmp/pip-build-env-ciw9wev7/overlay/lib/python3.11/site-packages/setuptools/build_meta.py", line 331, in get_requires_for_build_wheel
742.6 return self._get_build_requires(config_settings, requirements=[])
742.6 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
742.6 File "/tmp/pip-build-env-ciw9wev7/overlay/lib/python3.11/site-packages/setuptools/build_meta.py", line 301, in _get_build_requires
742.6 self.run_setup()
742.6 File "/tmp/pip-build-env-ciw9wev7/overlay/lib/python3.11/site-packages/setuptools/build_meta.py", line 512, in run_setup
742.6 super().run_setup(setup_script=setup_script)
742.6 File "/tmp/pip-build-env-ciw9wev7/overlay/lib/python3.11/site-packages/setuptools/build_meta.py", line 317, in run_setup
742.6 exec(code, locals())
742.6 File "", line 2, in
742.6 ModuleNotFoundError: No module named 'torch'
742.6 [end of output]
742.6
742.6 note: This error originates from a subprocess, and is likely not a problem with pip.
742.6 ERROR: Failed to build 'git+https://github.com/rahul-goel/fused-ssim@1272e21a282342e89537159e4bad508b19b34157' when getting requirements to build wheel
Dockerfile:47
45 | COPY . .
46 |
47 | >>> RUN CUDA_VERSION=$CUDA_VERSION bash ./install_env.sh 3dgrut WITH_GCC11
48 | RUN echo "conda activate 3dgrut" >> ~/.bashrc
49 |
ERROR: failed to build: failed to solve: process "/bin/sh -c CUDA_VERSION=$CUDA_VERSION bash ./install_env.sh 3dgrut WITH_GCC11" did not complete successfully: exit code: 1

[wakuwu]

@VladSarm I think I found the solution and successfully run the code.

modify the install_env.sh:

pip install -r requirements.txt
pip install -e .

to

pip install -r requirements.txt --no-build-isolation
pip install -e . --no-build-isolation

At the training, there are also the bug:

File "~/3dgrut/threedgrt_tracer/tracer.py", line 44, in load_3dgrt_plugin
import lib3dgrt_cc as tdgrt  # type: ignore
^^^^^^^^^^^^^^^^^^^^^^^^^^^
ModuleNotFoundError: No module named 'lib3dgrt_cc'

It cause the execute file store at wrong place. the execute file cloud be found at /.cache/torch_extensions/py311_cu128/

So I manually copy the missing file. e.g:

cp ~/.cache/torch_extensions/py311_cu128/lib3dgut_cc/lib3dgut_cc.so ~/3dgrut/threedgut_tracer/lib3dgut_cc.so

Same as lib3dgrt_cc.so, libplayground_cc.so, lib3dgrut_gui_cc.so should put at correct folder.

Hope this work for you.

This command works for me!

cp ~/.cache/torch_extensions/py311_cu128/lib3dgut_cc/lib3dgut_cc.so /workspace/threedgut_tracer/lib3dgut_cc.so

[Zenif-NIght]

I had to make a few modifications to the install script to get it working on a Blackwell RTX 6000 Pro card:

1. I ran into a missing torch error.
2. To fix it, I updated the file install_env.sh. Specifically, after the line:

elif [ "$CUDA_VERSION" = "12.8.1" ]; then

I added:

pip3 install --pre torch torchvision torchaudio \
    --index-url https://download.pytorch.org/whl/nightly/cu128

3. Following a suggestion from [this comment](Blackwell 6000 Installation #167 (comment)), I also updated the installation commands to:

pip install -r requirements.txt --no-build-isolation         
pip install -e . --no-build-isolation