unlocking the future
Heterogeneuous Computing Clusters Stand-alone network-attached FPGA
Management of cloud FPGA at scale
Heterogenes Computing Cluster
The concept of stand-alone network-atteched FPGA is based on two pillars. First, FPGA are no longer attached to a CPU via PCIe attachment as in the past but via a network attachment. Second, FPGAs are promoted to the rank of remote peer processors by disaggregating them from the servers and making them available as independent and self-managed resources in the cloud.
The cloud FPGA infrastructure is the most important prerequisite for such a large-scale deployment of FPGA in data centers. It was designed from the ground up to provide the world's densest and most energy-efficient rack unit of FPGAs.
A stand-alone, network-connected FPGA can be requested independently from a host via an FPGA Cloud Resource Manager. This provides an API for integration into the management stack of the data center (e.g. OpenStack).