Question:WARNING: no GPU detected, will be using CPU #210
Comments
|
Unfortunately, this problem did not occur in my environment. Please check the following points.
$ /path/to/your/localcolabfold/colabfold-conda/bin/python3.10
Python 3.10.13 | packaged by conda-forge | (main, Dec 23 2023, 15:36:39) [GCC 12.3.0] on linux
>>> import jax
>>> print(jax.local_devices()[0].platform)
gpuI expect "gpu" returns here. |
|
I'm running into the same problem. I've performed a new installation as per instructions. gcc --version nvidia-smi but still getting the warning about no GPU found: seems jax is the problem I'm not exactly sure how to update JAX within the localcolabfold conda env |
|
I have figured this out. As per a hint on the jax documentation site I unset The next problem I had was this error: I was able to solve this problem by installing the Now |
|
Hi I was having the same error "WARNING: no GPU detected, will be using CPU". I upgraded the Nvidia driver to 545.23.08 with cuda version 12.3. I checked this with both commands: "nvidia-smi" and "nvcc --version". Nonethelss, every time I tried to run "colabfold_batch" I always got the same error as mentioned before. Following the instructions of @YoshitakaMo I saw that JAX could not load cuda. Despite having the right version of cuda installed, JAX was still complaning about the version of cuda being 11.7 and not 11.8. I could fix this issue by going into the installation script (install_colabbatch_linux.sh) and changing line 25: "$COLABFOLDDIR/colabfold-conda/bin/pip" install --upgrade "jax[cuda11_pip]==0.4.23" Basically instead of installing the cuda11 version of jax I installed the version for cuda12. After that the localcolabfold envioroment has the right cuda installation andcolabfold_bacth is running as nice and smooth as before. Best regards, |
|
BTW, I had this same error recently after a fresh install where both In the end the issue was with the tensorflow installation which seems to default to using CUDA 12. The solution for me was that after running the install script I then entered the conda environment and essentially did a I actually used the script below to accomplish this, based on the linux install script, your mileage may vary. |
|
I have an AMD GPU in my workstation. Is this GPU going to work with Alphafold? |
|
No, AMD GPUs are not supported currently. |
|
I have to this day the same problem.
|
|
I'm having the same problem. 2024-05-03 06:24:18,051 WARNING: no GPU detected, will be using CPU I'm using the exact same version of nvcc version as you said. this is my computer's gpu version $gcc --version $nvidia-smi +-----------------------------------------------------------------------------+ python -c "import jax; print(f'Jax backend: {jax.default_backend()}')" |
|
@frenko @crshin Could you tell me the version of $ /path/to/your/localcolabfold/colabfold-conda/bin/python3.10 -m pip list | grep nvidia-cudnn
nvidia-cudnn-cu11 9.1.0.70
$ $ colabfold-conda/bin/python3.10 -m pip list | grep jax
jax 0.4.23
jaxlib 0.4.23+cuda11.cudnn86I suspect a mismatch of cudnn and jaxlib+cudnn versions causes the error. |
|
from my localcolabfold: cuda 12 is installed on my system so in the localcolabfold installation file I replaced jax[cuda11_pip]==0.4.23 with jax[cuda12_pip]==0.4.23. As suggested by some users |
|
@frenko In my system, I installed cuda-12.3, but |
|
sorry for late reply. from my localcolabfold, In my case, I'am using cuda 11 version. |
I solved by giving the following command: |
|
In my case, that doesn't work. . |
|
As mentioned yesterday I update my situation. After updating the jax packages with pip: and for cudnn: |
|
Hi everyone, I was having the same problem of GPU not being detected, and after updating JAX as @frenko did I got it to work. My issue now is related to GPU memory: As for the versions of CUDA, JAX, and cudnn: Thanks a lot in advance and let me know if I should provide any more info. |
|
First, check your cuda version using @crshin I recommend jax and jaxlib==0.4.23 with gpu support, not the latest one. ./localcolabfold/colabfold-conda/bin/python3.10 -m pip install "jax[cuda11_pip]==0.4.23" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
# or use this one for cuda 12
# ./localcolabfold/colabfold-conda/bin/python3.10 -m pip install "jax[cuda12_pip]==0.4.23" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
./localcolabfold/colabfold-conda/bin/python3.10 -m pip install jax==0.4.23@f-meireles |
|
To start anew, I uninstalled the existing localcolabfold file and reinstalled it. And now, these are my nvcc version and gcc version. (also nvidia-smi) But I found some points that presumed to cause the problem. And during the installation of "jax[cuda11_pip]==0.4.23", This warning kept popping up. I'm looking for a solution to this in my own way, but it's not easy. Do you have any guesses or solutions? |
|
I've noticed that "cuda11/12-pip" installation is not recommended now: https://jax.readthedocs.io/en/latest/installation.html $ ./localcolabfold/colabfold-conda/bin/python3.10 -m pip list | grep jax
jax 0.4.23
jax-cuda12-pjrt 0.4.23
jax-cuda12-plugin 0.4.23
jaxlib 0.4.23+cuda12.cudnn89
$ jaxlib 0.4.23+cuda12.cudnn89
nvidia-cudnn-cu12 9.1.0.70@crshin How about updating CUDA and Nvidia-driver to the latest version? It might improve that problem. I'm using CUDA 12.4 and NVIDIA-SMI 550.54.15. |
|
@YoshitakaMo Thanks for the reply. Even after using the latest version of the As for the Which should be compatible with my jax version: Any idea what else could be the issue? Thanks again! |
|
@f-meireles Which is your OS, Windows (WSL2) or native Ubuntu 22.04? |
|
@YoshitakaMo It is native Ubuntu 22.04. I just tried the solution you sent but also didn't have success. I'll keep trying things here. |
|
I updated CUDA version 11 .8to 12.4 as you adviced. from this state, the same problem occured as with other users, so, I installed 'jax-cuda12-pjrt 0.4.26' and 'jax-cuda12-plugin 0.4.26' like @frenko 's comment: after that, my localcolabfold successfully detected the GPU: However, from now on, I'm experiencing the same problem as @f-meireles I tried the solution |
I was running into the same issue and this solution worked for me, just updating to nvidia-cudnn-cu12 and jax[cuda12] as recommended in the reply above. I am in an Oracle Linux 9.4 and Tesla 4 GPUs. |
|
@crshin Your problem is more likely to be a JAX/CUDA issue rather than localcolabfold itself... It may be more helpful to ask in a JAX issue forum: jax-ml/jax#15361 |
|
@YoshitakaMo @crshin I was finally able to make it work! I had to get ColabFold directly from the original repository. In an new, empty conda env with Python3.10 I did: And now it works normally. Unfortunately, I didn't figure out what was wrong, but I hope this works for you as well. |
|
That works on my computer too!! @YoshitakaMo @fatpmeireles |
I met the same problem, and this workaround just worked! Thanks! |
hello,I'm having some problems running it with a “WARNING: no GPU detected, will be using CPU” error, how can I fix this?
The text was updated successfully, but these errors were encountered: