Matimex Group Datahub is equipped with Industrial servers with Graphic Processing Units (GPU) onboard. Colocation space is organised in Estonia, Tallinn (European Networks, www.en.ee)
Using GPU card in servers give us a massively parallel-able processing power. A single GPU unit can have many hundreds or thousands of cores. Where a CPU’s strength is in a single or a few complex threads, a GPU can handle many streams of data at once and process them in parallel. Deep Learning and High Performance Computing of big data sets, any application requiring highly parallel processing workloads can benefit from enabling GPU instead of CPU processors.
Medical data sets are basically have much data from connected data sources that need to be fed into a server and processed. By having a highly parallel capable system on board, this data is analyzed simultaneously on GPU modules.
Matimex Group Data hub overview
Data hub is located in Tallinn, Estonia in TIER-III data center. Estonia is the best place to work with high consumption GPU processes. The data center is certified by TIER rating and ISKE standards. Additional certifications: Bureau Veritas applied ISO:9001 standard to all processes and management system, also PCI and ISO:27001 certification was done.
Our data hub, expanded in APC racks inside the data center is guaranteeing by power supply from two separate Schneider Galaxy UPS units and independent feeder lines. Power lines are connected to two different substations. If one of line is breaking down, then an automatic system switch immediately activate the second line. If both lines are down, then power generator will automatically start feeding the data center.
All our servers are installed in the 19″ cabinets and equipped with two Schneider/APC Smart UPS units and different PDUs (Power Distribution Units). The servers have Dual PSU (Power Supply Units) and in case of PSU failure or power line failure will remain operational.
Wiring and PDUs in the 19″ cabinets are organised for GPU servers colocation. Each cabinet have dual feeders, each of 32 Amp (ampere). Total of 2x32Amp per rack.
We use HP and Supermicro professional 19″ rackmount servers optimized for Deep Machine Learning. Neural Network servers are equipped with GPU cards (Nvidia).
Data hub is connected by 3x separate fibre-optic cables to Baltic communication fiber rings with a total uplink speed of 24 Gbit/s. This upstream connections is linking data hub to all main points of interconnects in Europe and USA. The overall network speed allows realtime operations with big data sets and live migrations of VM involved in Deep Learning cloud.