Company overview

 

Product information

Markets

Kalray is presently focused on two main markets with a very strong short-term potential:

• Intelligent data centers:Kalray’s processors enable new-generation storage systems and networking cards.
• Intelligent embedded intelligent systems: Kalray’s processors enable new-generation embedded systems such as autonomous vehicles, healthcare equipment, as well as drones and robots.

By 2020, the commercial potential of these markets is estimated at $ 3,25 billion with the emergence of artificial intelligence and rapid memory storage (Sources: Gartner/IDC).

Data storage

With the adoption of high-speed, low-latency SSDs, new performance bottlenecks are created at different points along the data path. The deployment of the emerging NVMe-oF protocol helps alleviate these bottlenecks and enable a disaggregated storage model to scale compute and storage resources independently.

With its MPPA offering more than 100 cores towards end-user processing, Kalray offers customers the chance to make their JBOF smarter while exploiting the performance of SSDs to the fullest. Equipped with Kalray’s MPPA processor, Kalray’s NVMe-oF Target Controllershave fully integrated CPU + NIC functions which enable high-density/high-performance storage solutions with in-line processing capabilities such as deduplication, encryption, compression, RAID, etc. The KTC solution is the first fully integrated system certified by the NVM Express organization.

Data networking

Virtualization has changed the face of many industries, but none so much as networking for telecommunications and data centers. Data centers are quickly adopting two new techniques, Network Function Virtualization (NFV) and Software-Defined Networking (SDN), which offer software and Cloud-based solutions in order to avoid costly and frequent hardware updates.

Kalray offers several programmable SmartNIC solutions, which allow telecommunications and data centers to completely offload their compute-heavy CPUs. With real-time processing, reliably low-latency and standard C/C++/OpenCL programmability, customer can maintain line-rate performance implementing solutions such as Juniper Contrail and RedHat Openstack.

Autonomous vehicles

Present-day cars are brimming with electronics since more autonomy requires more capabilities and much more computing power. At the same time, car infrastructures are moving from a multitude of disseminated small processors (ECU) to a much more centralized platform (“SuperECU”)supporting a wide number of assistance functions. Deep learning, computer vision, sensor fusion are all compute-intensive functions, yet they are just a subset of what needs to run in parallel.

Kalray’s MPPA is the ideal optimized processing solution for autonomous vehicles, offering:

• Extreme computing:an unreached level of heterogeneous compute power and the ability to execute simultaneously multiple independent applications or multiple threads of the same application so that all time-critical heavy-computing functions of the Autonomous Driving System can be integrated on the same die. In addition, the MPPA offers an optimized tool and libraries allowing the best performances of deep-learning or vision-type algorithms. The inherent scalability of the MPPA architecture further allows the use of one, two or more MPPA processors, depending on the level of performance they require.
• Safety:architectural determinism ensures reliability, security and a high level of safety. The MPPA processor supports ASIL B/C and fulfils the ISO 26262 standard.
• Power consumption:as low as one tenth of that of existing solutions so that architects can develop dense systems with no, or limited, cooling needs, thus reducing the cost of the solution and of maintenance.
• Programmability:the MPPA offers an easy-to-program platform, that can be easily customized and updated. Not only can standard C/C++/OpenCL code libraries, standard libraries (OpenCV, OpenVX…) or run-time environment (AUTOSAR) be used, so can plugins to existing code generation tools, allowing the reuse of an easy legacy code.

Contact us