6

I wanted to accelerate pandas on my GPU so I decided to use cudf library. Please do suggest other libraries(if any).

I tried to install cudf using pip by pip3.6 install cudf-cuda92. The pip version is 19.2.3(latest).

When I run pip3.6 install cudf-cuda92 on my cmd it says:

Collecting cudf-cuda92
  ERROR: Could not find a version that satisfies the requirement cudf-cuda92 (from versions: none)
ERROR: No matching distribution found for cudf-cuda92
talonmies
  • 70,661
  • 34
  • 192
  • 269
rahul_5409
  • 71
  • 1
  • 1
  • 5

4 Answers4

2

You can try installing using conda.You can get a minimal conda installation with Miniconda or get the full installation with Anaconda.

Install and update cuDF using the conda command:

# CUDA 9.2
conda install -c nvidia -c rapidsai -c numba -c conda-forge -c defaults cudf

# CUDA 10.0
conda install -c nvidia/label/cuda10.0 -c rapidsai/label/cuda10.0 -c numba -c conda-forge -c defaults cudf

Find out more from cudf.PyPi

Dennis
  • 307
  • 2
  • 10
2

You can install with pip, but be aware that the cuda version dependency can be quite strict and may need to be matched to your cudf version. Different cudf versions have been provided for this purpose, but they may lag behind latest.

e.g. I'm running cuda 10.0 so installed cudf with:

pip3 install cudf-cuda100

and then added a file in /etc/ld.so.conf.d to point to /usr/local/cuda-10.0/lib64

(I'm running on Ubuntu 18.04)

2

cudf pip install is officially NOT supported nor maintained. Check out the discussion on cudf github and the official rapids announcement.

The way to go is either through conda or by building the source code.

tronic
  • 220
  • 1
  • 10
0

try:

!pip install cudf-cu11==22.12 rmm-cu11==22.12 dask-cudf-cu11==22.12 --extra-index-url=https://pypi.ngc.nvidia.com/

https://github.com/rapidsai/cudf/issues/12762

martin
  • 862
  • 9
  • 28