AGH UST Cyfronet offers a new computing system for research with the use of artificial intelligence methods

The AGH UST Academic Computer Centre CYFRONET has launched the fastest Polish academic computational system dedicated to the needs of artificial intelligence, with the power of over 4 Pflops (petaflops). The new infrastructure has been added to Prometheus, the largest Polish supercomputer, thanks to which it is possible to enrich traditional simulations with methods based on machine learning carried out in the same computing environment and on the same data sets. 

The new system consists of four HPE Apollo 6500 servers, each equipped with two Intel Xeon Gold 5220 processors, eight NVIDIA Tesla V100 accelerators, and 384 GB of internal memory. The system’s computing power is over 4 Pflops (petaflops) for tensor operations and 256 TFLOPS (teraflops) for standard operations on double precision numbers, which makes it the fastest dedicated solution for artificial intelligence available for the needs of Polish science.   

The applied NVIDIA Tesla V100 accelerators have the form of modules taking advantage of the specialised SXM2 interface, replacing the conventional PCI Express connector, which makes it possible to use the NVLINK bus offering a transfer capacity of up to 300 GB/s. Each module has a highly efficient HBM2 memory offering a capacity of 32 GB and a transfer capacity of 900 GB/s. The servers are connected with one another by an InfiniBand network offering a transfer capacity of 40 Gb/s, and have been added to the existing infrastructure of Prometheus. Thanks to it, they can make use of over 10 PB of disk memory, offering access speed at the level of 180 GB/s, which is particularly important for obtaining the highest efficiency of machine learning. 

The new infrastructure will make it possible to conduct research with the use of artificial intelligence methods, which can also be used to enrich research projects carried out on the basis of HPC simulations (high performance computing).  

AGH UST Cyfronet is currently engaged in a number of projects that take advantage of artificial intelligence methods in various domains of science. An example can be calculations in medical diagnostics, where the automatic recognition of images is applied to the classification of microscope pictures for the purpose of defining the types of antibodies present in blood samples taken from patients, or the application of machine learning techniques for the purpose of obtaining missing medical information about a particular case of treatment in the conditions of insufficient or contradictory hospital records. The system will be used for research projects in the fields of chemistry, biology, medicine, and the development of algorithms for autonomous vehicles. The system was launched in December 2019, and is available through the PLGrid infrastructure.