Kalray unveils its Kalray Target Controller and Kalray Neural Network
Kalray announced Kalray Target Controller (KTC80) with inline mirroring to offload the network while securing the data and the Kalray Neural Network (KaNN) performing live object recognition using artificial intelligence.
The KTC (Kalray Target Controller) solution has been certified earlier this year by the NVM Express organization through the University of New Hampshire InterOperability Laboratory (UNH-IOL), an independent testing provider of standard conformance solutions and multi-vendor interoperability. This certification is the first of its kind for a fully integrated system.
Randall Skelley, Kalray’s VP Data Center Business Unit, declared: “This is an important milestone for Kalray. We contribute to advancing NVMe technology and helping storage companies build intelligent systems for optimized data centers. The KTC80, which combines the Konic80 board, the intelligent MPPA processor and associated software, delivers a unique all-in-one system solution.”
Rex Chu, VP sales AIC Europe, added: “The NVMe-oF certification demonstrates the level of maturity of the Kalray solutions in the market. They address customers’ needs for systems that bring analytics and other computing capabilities closer to storage.”
Besides supporting the NVM Express protocols, KTC offers extra resources in situ (i.e. more than 100 cores out of 288 of the Kalray Bostan2 processor), providing customers with smarter storage solutions.
In-situ computing also saves network bandwidth, running Input/Output-intensive applications closer to the storage capacity.
Such tasks as RAID/erasure coding, deduplication, compression, encryption or analytics, deep learning and artificial intelligence algorithms can be directly offloaded from the storage server.
In addition, Kalray provides the Kalray Neural Network (KaNN), a tool that allows customers to take full advantage of the performance and flexibility offered by the MPPA’s unique architecture.
This solution imports the trained models from usual frameworks (Caffe, TensorFlow, etc.) to ensure an optimal execution of artificial intelligence algorithms such as GoogLeNet, ResNet, YOLO and others.