Virtualization of GPU cards - who needs it? | Copernicus Computing
chat

IT WORKS!

It was just yesterday that, at the very last moment,  we finished our project “Virtualization of GPU cards for sharing computing power in the cloud for the needs of scientific research or creative industries”. I use the expression “at the very last moment”  because we had to speed up the work in relation to the originally planned dates. If you have ever implemented the IT project, you can easily imagine the challenge to shorten its implementation time by 30%… However, we managed to do it and all the tests carried out yesterday indicate that the code aggregation was successful. It is mainly a credit to our invaluable team of programmers, who have lived on coffee and energy drinks over the last few weeks.


Now let me present briefly what it gives us. Thanks to the virtualization of the hardware resource, you can separate it from the place where it is located and assign it to another. Thanks to this we can create a computer with 150 GPUs running simultaneously without any major problems (The Witcher 3 runs great). With more serious and business applications, we can use it to make calculations 150 times faster than on a single machine without having to queue a job on multiple machines. All this is remote and done, as it is said,  with “one click”. If it was not enough, this system will be implemented to our Karpiński supercomputer, which is currently being built. There, the system will allow you to connect (more sensitive nerds- please sit down and fasten your seatbelts) 800 graphics cards to a single computer.

 

Michał Organiściak – he started his adventure with 3D graphic in the 90’s. Originally he used LightWave, creating more or less complex models and visualisations. He is currently managing the users’ scenes, supervises the programming and supervises the spending of subsidies. He likes simple, fast and effective solutions. In private he is a fan of 4×4 cars and medieval sword fighting.