The install procedure is similar for Mac, Linux, and Windows. Before starting GPU work (in any programming language) realize these general caveats:
- I/O heavy workloads are generally less suitable for GPU
- Consumer GPUs (GeForce) are > 10x slower than workstation class (Tesla, Quadro)
You must have a discrete Nvidia GPU in your laptop or desktop. Check for existence of an Nvidia GPU in your computer by:
lspci | grep -i nvidia
a blank response means an Nvidia GPU is not detected.
If you have have a Compute Capability 2.x Fermi GPU, you can fallback to Nvidia Toolkit 8.5.
- Download Nvidia Toolkit 9, choosing → Linux → x86_64 → Ubuntu (pick version ≤ your install) → deb(network) and download the base installer
- Install CUDA Toolkit, from Linux Terminal:
dpkg -i cuda-repo-*.deb apt update apt install cuda
- Install Anaconda Accelerate. Setup a distinct conda environment for Cuda, since it requires specific module versions (some of which are not the latest).
conda update conda conda create -n cuda conda activate cuda conda install accelerate
These commands are issued within Python.
import numba.cuda.api,numba.cuda.cudadrv.libs numba.cuda.cudadrv.libs.test() numba.cuda.api.detect()
- Even if you already have Continuum Accelerate (formerly NumbaPro) for Anaconda Python installed, it won’t work unless you have an adequate GPU with the CUDA drivers installed.
- Try fixing error about having gcc too new by switching gcc version.