Install KDB-X and the GPU Module¶

This page explains how to install KDB-X with a GPU-enabled license, download the GPU module, and run the application on a host or in an NVIDIA container.

Install KDB-X¶

Install KDB-X using the KDB-X install guide.

Download the GPU module¶

Retrieve a KDB-X GPU-enabled license and download the GPU module from the KX downloads portal.
Save the GPU module in $HOME/.kx/mod/kx.

Note

These instructions use the default KDB-X installation location $HOME/.kx. If you installed KDB-X elsewhere, adjust the code examples to match your location.

The folder structure looks similar to:

~/.kx/mod/kx/
├── ai
├── gpu.li64.so ← (gpu module)
├── kfk
├── kurl
├── objstor
├── pq
└── rest.q_

For more information on modules, refer to modules.

Run the application¶

With KDB-X and the GPU module installed, run the application directly on the host or in an NVIDIA container (requires Docker).

Non-containerized¶

If you run directly on the host, install the CUDA Toolkit on the host and ensure the CUDA runtime and shared libraries are available to the application.

Load the KDB-X GPU module and inspect the available APIs:

q).gpu:use[`kx.gpu][`init] ([copy:`async;alloc:`simple])
q)key .gpu
`cfg`to`from`sync`type`ndev`gdev`sdev`mdev`select`append`bin`iasc`asc`xasc`dbg`hex`zero`Qajf0`ajf0`aj`aj0`ajf

Containerized app with CUDA dependencies¶

We recommend using NVIDIA containers. To run the app, use the following Dockerfile examples (covering ubuntu24.04 and rockylinux10 for CUDA 13.1 built apps):

Ubuntu example

FROM nvidia/cuda:13.1.0-runtime-ubuntu24.04

RUN apt-get update && \
    apt-get install -y --no-install-recommends \
        curl \
        unzip \
        zip \
        ca-certificates \
        build-essential

RUN mkdir /opt/kx
WORKDIR /opt/kx/app
ENTRYPOINT ["/opt/kx/bin/q"]

Red Hat example

FROM nvidia/cuda:13.1.0-runtime-rockylinux10

RUN dnf install -y \
        curl \
        unzip \
        zip \
        ca-certificates \
        gcc \
        gcc-c++ \
        make \
        libubsan && \
    dnf clean all

RUN mkdir /opt/kx
WORKDIR /opt/kx/app
ENTRYPOINT ["/opt/kx/bin/q"]

To build the image:

docker build -f Dockerfile -t kdbx-gpu .

To run the image, mount the KDB-X directory to the container, along with any data sources, dependencies, and ports specific to your application:

docker run --gpus all -it -p 1234:1234 -v /home/mmcphillips/.kx:/opt/kx kdbx-gpu -p 1234

Load the KDB-X GPU module and inspect the available APIs as shown in the Non-containerized section.

Warning

Ensure the --gpus flag is passed to docker run, otherwise, you will see the following error:

'no kexport: libcuda.so.1: cannot open shared object file: No such file or directory