Collaboration between academia, hardware and software companies heralds breakthrough in GPU processing speeds and ease of use
A coalition of European hardware and computer software companies and the University of Bielefeld unveiled their prototype at the GTC Conference in Amsterdam last week, which they say will herald a transformation in the way enterprise and data centres can use GPU units in the future.
The collaboration between Toradex, Brytlyt, Christmann and the University of Bielefeld means that for the first time companies using a specialised Tegra GPU carrier will be able to deploy to data centres much more processing speed and power while reducing energy consumption. The technology can also, according to their showcase, eliminate data transfer bottlenecks and the result is an ability to utilise more data and more sophisticated analytics in the future.
Until now Tegra GPUs have been designed for use in embedded units such as mobile phones, tablets, autonomous vehicles, drones and remote processing. The new prototype extends the Tegra GPU use case into the datacentre; increasing GPU memory footprint and removing the data transfer bottlenecks in standard GPU (Tesla) systems. This means SQL databases like the one developed by Brytlyt are able to use Tegra GPUs effectively.
Richard Heyns, CEO of Brytlyt commented “We are delighted to be working on this project with our European colleagues. Faster, more efficient processing will open up a new world for those enterprises and data centres that until now have not been able to access this kind of processing power and energy efficiency.”
A number of technology challenges had to be overcome by the coalition of hardware and software companies including developing micro-server modules, a module carrier, exposing the modules via network, optimising power consumption and developing the software to run across the modules. The software has been designed to federate memory and compute so that the micro-server modules can be addressed and used as a single entity. Tegra is different in that it combines ARM cores and NVIDIA GPU cores rather than typical x86 architecture.
“We are very pleased to be collaborating with Brytlyt, Bielefeld University and Toradex. The GTC has presented a great platform to exhibit and demonstrate the exceptional performance our hardware is able to deliver” – Micha vor dem Berge (Christmann)
“It’s really exciting to be at the NVIDIA GTC and showcase our hardware running the World’s fastest SQL database with Christmann, Brytlyt and Toradex” – Jens Hagemeyer (Bielefeld University)
“We are really excited to see our modules being used in this way.” – Marcel Ziswiler (Toradex)
The prototype was unveiled at the GTC conference last week and showed how the collaboration was able to build a unit that allows the user to have an easy to consume interface by adopting PostgreSQL as the host. Now, legacy code and reporting infrastructure can run out the box using the new hardware.