GPU Accelerator Installation


Related links: Jedox GPU Accelerator RequirementsJedox GPU Accelerator on VMware, Using the Jedox GPU Accelerator

Please install the NVIDIA driver before installing or updating the Jedox Suite. Use of the latest driver is recommended.


The following table lists the CUDA Toolkit versions used in latest Jedox releases and the minimum required driver versions:

CUDA Toolkit version

Jedox version

Minimum driver version






2018.4, 2019.1, 2019.2



2018.2, 2018.3



7.0 – 2018.1



6.0 SR2, 6.0 SR3


Driver installation:

To install NVIDIA graphics drivers, first download the driver, then start the driver installation process:


After installing the drivers, make sure that TCC (Tesla Compute Cluster) mode is enabled for each Tesla™ GPU. To check, start the tool nvidia-smi ( which comes with the driver installation) via command prompt:

cd C:\Program Files\NVIDIA Corporation\NVSMI

For each GPU, the driver mode is shown to the right of its name. In the screenshot above, the GTX™ card runs in WDDM driver mode and the Tesla™ card runs in TCC driver mode. If you see any Tesla™ card running in WDDM mode, change its mode with the following command:
nvidia-smi -g 0 -dm 1 

The -g option refers to the ID that appears to the left of the GPU name. Change it accordingly. Reboot the system after performing the change.


Download the .run file driver installer.

Make binary file executable with:
chmod +x

Stop the Xserver by changing to run level 3 with:
init 3 


  • You have to first install the Linux kernel headers and the build tools of your distribution to run setup.
  • If the “nouveau” module is loaded on startup, this module has to be blacklisted.

Then run executable as ‘root’ with

For more information about driver installation on Linux, please read the NVIDIA CUDA Installation Guide for Linux:


Jedox Suite Installation

Use the Windows Jedox Setup file from the official Jedox website.

During the Windows Setup, select the “Advanced Setup” type:

In the “Final options” setup dialog, select “Enable GPU Accelerator”. This option is only selectable if a supported GPU card was found.

Activation will be done automatically during the installation process.


Download the current Jedox Premium download for Linux and stop any running Jedox Suite instance. At this point, we recommend backing up existing files. Extract all files in the tar archive to a temporary directory. Install it by executing

After installation add the option


to the palo.ini configuration file. Start the Jedox Suite. 


When not specified otherwise, Jedox starts with all instances of the most powerful GPU found in the system according to the licensed number of devices (for questions on licensing, please contact Jedox). If only specific GPUs should be used, then those must be explicitly listed in the configuration file (palo.ini) in the following way:

device 83.0
device 84.0

The device IDs can be taken from the OLAP server log file after a successful startup:

… INFO: GPU: Device Count: 3 – CUDA driver / runtime version: 8.0 / 8.0
… INFO: GPU: Device 0 (Tesla K40c) has ID 83.0
… INFO: GPU: Device 1 (Tesla K40c) has ID 84.0
… INFO: GPU: Device 2 (Tesla K40c) has ID 85.0

Performance optimization

To further enhance the performance of the Jedox GPU Accelerator up to 100%, we recommend modifying the operating system’s power options. 

To prevent Windows from setting the system to power saving mode, thereby slowing down the OLAP Server, follow these steps:

  1. Go to Control Panel > System and Security > Power Options.
  2. Choose “High performance” (maybe hidden under “additional plans”) or generate your own plan and enter the “Change advanced power settings” dialog.
  3. Set “Processor power management > Minimum processor state” to “100%”. In addition, you can set the “PCI Express > Link State Power Management” to “off”.

For every CPU core with index <X>, run the following command to always set the CPU frequency to the maximum:

echo performance | tee /sys/devices/system/cpu/cpu<X>/cpufreq/scaling_governor

nvidia-smi -ac <memory,graphics> (e.g. 2000, 800)

This command specifies maximum <memory,graphics> clocks that define the GPU’s speed while running applications on a GPU. Note that the command only works on Tesla devices from the Kepler+ family.
See NVIDIA System Management Interface for details.