Revolutionary changes await industrial production: the use of augmented and virtual reality, artificial intelligence and rapid adaptation to the current needs of the customer or available resources. All this will be brought about by the RICAIP research project.
And M Computers has become a provider of solutions for research in the field of artificial intelligence.

Research and Innovation Centre on Advanced Industrial Production

The RICAIP project (Research and Innovation Center on Advanced Industrial Production) was launched at CIIRC ČVUT in Prague last year: It is a strategic research cooperation between the Czech Republic and Germany, which will bring great changes in industrial production in the future. This platform will examine, among other things, the use of artificial intelligence and industrial robotics. The goal is to shift the understanding of automated industrial production as we know it today to production that is extremely flexible and distributed, and sometimes referred to as Industry 4.0.

The partners of the international RICAIP project are its main coordinator CIIRC ČVUT, then CEITEC VUT and on behalf of the German side DFKI and ZeMA based in Saarbrücken.

The significance and potential of the project is also evidenced by the participation of Tomas Mikolov (AI Expert Tomas Mikolov is coming from Facebook AI to join RICAIP team), the world’s leading expert in artificial intelligence research with experience from Google, Microsoft or Facebook.

Supercomputer for AI workloads

A necessary part of the research is the computational power in the background. However, choosing the right one is not easy – whether in terms of technology, performance or efficiency. M Computers therefore delivered a solution based on NVIDIA DGX-1 MaxQ technology. It is a version of the classic NVIDIA DGX-1 solution specially designed for research institutions.

The most effective solution

The uniqueness of the DGX-1 MaxQ lies in the optimization of the power / consumption ratio: With half the energy consumption, it reaches 80% of the performance of the classic DGX-1!

This not only brings significant savings in terms of operating and acquisition costs, but also contributes to the sustainability of the research.


The main features of DGX-1 MaxQ

The RICAIP project has been using a rack with five NVIDIA DGX-1 MaxQs since May 2020.

In each of these devices, there are eight NVIDIA V100 accelerators connected by NVLinks, so the whole solution uses 40 accelerators with a total performance of 4.3 TFLOPS (mixed precision) for AI applications.

A total of 204,800 computing (CUDA) cores work in the supercomputer using 2.5 TB of RAM. The parameters of one unit are clearly shown in the table:

How to leverage Multi-GPU systems?

This is one of the most common questions our customers ask us. There are couple of techniques you can use as described in webinar to Multi-GPU topic. Or you can attend Fundamentals of Deep Learning for Multi-GPUs workshop organized by us together with NVIDIA Deep Learning Institute (DLI).


But what is much more interesting is the already mentioned software equipment of the offered NVIDIA machines. All of them offer pre-installed and, above all, performance-tuned environments for machine learning (eg Caffe or Cafe 2, Theano, TensorFlow, Torch, or MXNet) or an intuitive environment for data analytics (NVIDIA Digits). All elegantly packed in Docker containers. According to NVIDIA, such a tuned environment provides 30% higher performance for machine learning applications than applications deployed purely on NVIDIA hardware. The main advantage of a pre-installed environment is the speed of deployment, which varies in units of hours.


The strong point of the offered NVIDIA solution is the support of the whole system. Hardware support (in case of failure of any of the components) is a matter of course. Software support for the entire environment is essential in case something does not work as it should. The customer has hundreds of developers ready to help. Support is part of the purchase of NVIDIA DGX systems. It is available for 1 or 3 years and can be extended after this period.



Petr Plodik

Leave a comment

Your email address will not be published. Required fields are marked *