0

I just created a new env using minicoda with py3.9 and cuda. While trying to install cudf with: conda install -c rapidsai cudf i get the following error message.


Output in format: Requested package -> Available versionsThe following specifications were found to be incompatible with your system:

  - feature:/linux-64::__glibc==2.35=0
  - cudf -> __glibc[version='>=2.17,<3.0.a0']
  - cudf -> cupy[version='>7.1.0,<10.0.0a0'] -> __glibc[version='>=2.17']
  - python=3.9 -> libgcc-ng[version='>=7.5.0'] -> __glibc[version='>=2.17']

Your installed version is: 2.35

I've found a post with similar error and was asked to use pip instead. If there a way to rectify this while still using conda? Kindly advise.

talonmies
  • 70,661
  • 34
  • 192
  • 269
ManOnTheMoon
  • 587
  • 2
  • 11
  • please link similar post – merv Jun 14 '22 at 14:11
  • 1
    The problem with Conda solver failures is that they are context-sensitive to the environment being installed into. You would be better supported if you followed [the official installation instructions](https://rapids.ai/start.html#get-rapids), which means creating a new environment and specifying all the channels that the documentation shows. E.g., `conda create -n rapids-22.04 -c rapidsai -c nvidia -c conda-forge cudf=22.04 python=3.9 cudatoolkit=11.2` – merv Jun 14 '22 at 14:16
  • @merv, thank you for the solution. It worked, however, it took over an hour to load which could be due to incompatible nvidia. I've only found out that maxwell cards are deprecated from CUDA 11 onward after trying to run some old codes. I was running CUDA a couple of years back on windows, hence, wasn't aware of the change. This where I read about installation with pip [link](https://stackoverflow.com/questions/71835906/conda-reports-conflicts-but-there-seems-to-be-no-conflict) – ManOnTheMoon Jun 14 '22 at 19:31

1 Answers1

0

As merv said, your install command isn't speific enough to reduce your time, so my addition is to please use mamba to install RAPIDS. First you conda install mamba, then you use mamba create -n rapids-22.04 -c rapidsai -c nvidia -c conda-forge cudf=22.04 python=3.9 cudatoolkit=11.2. It should be much faster and is anaconda's recommendation as well

TaureanDyerNV
  • 1,208
  • 8
  • 9