Palo GPU - Accelerator

With the 3.1 release of Palo in April 2010, Palo GPU was introduced for selected customers in a ramp-up edition. In the following we will explain what Palo GPU is, what it can do and how it is best implemented in your business.

GPU Whitepaper Download click here

What is Palo GPU?

Palo GPU is a technology that was developed in cooperation with researchers from the University of Freiburg and the University of Western Australia,
co-funded by the German Research Foundation (DFG) and Jedox. It is a technology that uses Nvidia GPU hardware to speed up large data processing jobs.

What does Palo GPU do?

In addtion to the CPU of the server, Palo GPU uses the tremendous computing capacity of Nvidia graphics cards (GPUs) to speed up large OLAP aggregations. Modern GPUs come with hundreds of so-called stremaing processors and with several GB of graphics memory that allows us to store cube data directly on the GPU.

The way of parallel computation used in Palo GPU is explained below using the simple example of Newspaper Delivery .

The delivery truck in the following figure represents the CPU (processor):

The delivery truck can only deliver to one house at a time. Consequently, it will try to find a possibly short route passing all subscribers and then deliver to one household after the other.

The Paper boys in the following figure represent the GPU (grafic card):

A couple of Paper Boys (representing the thousands of threads running on a GPU) are more efficent, as there are so many of them. Each one simply goes to a house, checks if a subsciber lives there and if yes, delivers a paper. This way, newspaper delivery can be achieved much faster.

Palo GPU relies heavily on this kind of parallel computation, thus speeding up aggretions (i.e. the calculation of consolidated Palo cube cells).

How do I get it for my business?

At the moment Palo GPU is only available to selected partners and customers upon request. Please click here to request access to Palo GPU.
This is because we want to give the best possible to support to users in the ramp-up phase. Support includes a test whether the user's specific model is suitable for Palo GPU, what hardware to use, how to re-model the data if necessary, etc.

For best performance of Palo GPU Jedox recommends Nvidia Tesla C1060, M1060, and S1070 GPUs, which come in workstations and servers. 

More information on the Nvidia Products

  top