In a typical cloud server, the CPU must run user VMs and also process network storage packets from the NIC. As cloud workloads grow, software-based NVMe/TCP processing can consume CPU cycles, increase power usage, and serialize storage request handling.
NVMe/TCP protocol processing is moved entirely into FPGA logic, so the host CPU is no longer burdened by storage networking tasks and can dedicate its full capacity to tenant virtual machines.
Unlike sequential CPU software, dedicated FPGA logic handles multiple storage operations simultaneously — eliminating the queuing bottleneck that limits throughput under heavy I/O loads.
The NVMeTCP-IP core operates without an attached CPU or DDR memory, reducing system complexity, board cost, and power draw compared to conventional SmartNIC offload approaches.
Unlike fixed-function ASICs, the FPGA fabric is fully reconfigurable — the same physical hardware can be restructured to meet new protocol specifications, updated standards, or custom offload requirements without replacing any equipment. This means the NVMe/TCP offload engine can evolve alongside your infrastructure.
