Kernel tuner optimization software now ready to use

“Graphics processing units (GPUs) were originally intended for playing video games, but they have turned out to be excellent at performing scientific calculations, which is why they are used for all kinds of computationally intensive tasks,” explains Van Werkhoven. However, manually programming GPUs is a complex and time-consuming task. ‘Kernel Tuner allows you to more efficiently experiment with different versions of your code for these units.’

Outdated hardware

‘You see software generally lasts much longer than hardware. We have a new computer every three or four years, but computer software is sometimes used for decades,’ says Van Werkhoven. ‘Every time new hardware is available, you have to re-optimize your software – sometimes even over the course of a two-year project.’ Kernel Tuner uses optimization algorithms to try many different versions of software on GPU hardware. Van Werkhoven: ‘I wanted to automate the process of optimizing your software over and over again. You see many people still do it by hand.’

Faster decision tree exploration

Programming a graphics processing unit is an art in itself. If you also want them to be fast, the software must use the underlying hardware efficiently. ‘There are so many choices you can make and, in fact, you have to make exactly the right choice if you really want to achieve the optimum,’ explains Van Werkhoven. With Kernel Tuner, it is possible to automate this process and take all possible paths at once, identifying the optimal set of choices. ‘You can think of it as a decision tree that you move through. You can make all kinds of choices, but you can only manually walk one path at a time. With Kernel Tuner you can explore all branches of the tree at once.

Widely applicable

Kernel Tuner enables scientists to develop scientific software to perform complex calculations more efficiently. With a team of key developers spread across organizations across the Netherlands, including the eScience Centre, Leiden University, Centrum Wiskunde & Informatica, Astron and the Radio Astronomy Institute in Drenthe, Kernel Tuner is being developed. In addition to the software’s interest from a research point of view, Kernel Tuner also has practical applications in radio astronomy and meteorology.

Sustainability

“We can optimize not only in terms of time but also energy consumption,” points out Van Werkhoven. Kernel Tuner not only measures the efficiency of different configurations, but also calculates how much power they consume. ‘Especially in cooperation with Astronom, energy consumption is an important component in the development of radio telescopes. Antennas are often placed very far from civilization on purpose to have as little radio interference as possible. This means that the telescopes are located somewhere in a meadow where there may not be much power. Especially in such a situation, it is important that something like that can be done with a little strength.’

Finding outliers

The best set of parameters can make a big difference in terms of speed compared to other configurations. There is little chance that this outlier was found by manually adjusting the software. Even if this optimal configuration were found manually, it is a time-consuming job that cannot keep up with hardware development. As soon as a new graphics processing unit hits the market, the process can start all over again. Van Werkhoven: ‘Therein lies the power of Kernel Tuner: you can automate this process.’