Intensivate, the datacentre acceleration specialist,  says that, early in Q1 2020, it will have first silicon  of a chip-scale coherent fabric for a PCIe card which will provide 12x more compute per dollar on platforms such as Hadoop, Kafka, Spark, NoSQL databases and other custom applications running on clusters.

First shipments of the accelerator card are scheduled for early 2021.

AI accelerators, GPUs, FPGAs, and Custom ASICs are focused on deep learning or narrow function accelerations. Intensivate is going outside this class of acceleration by addressing applications that run on clusters and imposing no specialized software requirements typically seen with accelerator technologies.

Leveraging the RISC-V ISA tools, software, and industry ecosystem, Intensivate, in partnership with SiFive, has created a chip that consumes 14W.

A high-speed 120Gbps chip-to-chip mesh network allows the twenty-one chips on the accelerator to support network-intensive scale-out applications.

Each chip is combined with 32GB of quad-channel LP-DDR4 to feed data into the 16 Intensivate designed cores inside.

Each combination of chip and memory is equivalent to a 1U server node, providing twenty-one server nodes on a single PCIe card.

The accelerator card offers ROI savings, with a 1000-Node Hadoop cluster able to be matched in performance by just 84-Nodes when equipped with Intensivate accelerator cards, claims Intensivate, delivering over $1 million in CapEx and Power savings over a 3-year period.