Kontron, a global leader in IoT/Embedded Computing Technology (ECT), has been certified as a ‘Preferred Partner’ under the NVIDIA partner program and uses NVIDIA’s powerful Graphics Performance Units (GPUs) in its AI platforms, such as the Kontron rackmount server KISS V3 4U SKX, which was developed for AI and Inference applications.
Applications in the field of artificial intelligence (AI) are increasingly gaining importance in almost every industry. Technology providers such as Kontron are creating the platform for AI applications that can only be realized via data- and computationally-intensive processes.
Examples can be found in the fields of automation, the energy industry, pharmaceuticals and healthcare. In healthcare, AI can ease the time pressure in prevention, diagnosis and therapy, since AI applications structure and analyse huge amounts of data in a short time.
Powerful hardware and software components are essential for AI solutions to fulfil their role as system-critical applications in any industry. The systems must analyze terabytes of data in a very short time. This work relies to some degree on high-performance CPUs, but the critical parallel processing work is done by Graphics Processing Units (GPUs), which provide the necessary processing power for deep learning, machine learning and inferencing.
The NVIDIA® V100 Tensor Core GPU, based on the NVIDIA Volta architecture, offers 640 Tensor Cores and 5,120 NVIDIA CUDA cores with a double precision performance of 7 TFLOPS.
NVIDIA GPUs are part of the currently most powerful Kontron rackmount server, the KISS V3 4U SKX-AI. The scalable server is equipped with two powerful Intel® Xeon® SP series CPUs that can be expanded with twelve DIMM DDR4-2666 modules and up to 768 GB of system RAM with ECC support. Up to three double-width high-end GPU cards (e.g. NVIDIA V100 or an NVIDIA T4 GPU) ensure extremely high GPU performance, which is essential for AI applications. GPUs carry out the parallel multitasking that is required, for instance, for rendering graphics or for the efficient training and inference of neural networks. For software development, users can deploy the powerful NVIDIA CUDA toolkit.