Dataflow Development Engineer - LPU Hardware DataFlow

NVIDIA is known as the "AI Computing Company." Our GPUs power modern Deep Learning software frameworks, accelerated analytics, data centers, and autonomous vehicles. We are looking for a Dataflow Development Engineer - LPU Hardware to join our team and develop, build, and improve dataflow systems at the hardware–software boundary. You will work on FPGA accelerator dataflow: implementing and tuning dataflow pipelines, creating host-side drivers and runtimes that collaborate with programmable logic, and jointly inventing hardware and software for deterministic, low-latency execution. Dataflow development engineers at NVIDIA connect FPGA and custom hardware with our software systems. You will implement dataflow graphs and streaming pipelines in hardware. You will build efficient host–device interfaces (PCIe, DMA, VFIO) and collaborate with compiler and architecture teams to map high-level dataflow onto FPGA and accelerator fabrics. Your work directly affects latency, efficiency, and resource usage for inference at scale. The ideal candidate has a proven hardware approach, including experience with FPGA development, HDL, or hardware/software co-design. They can analyze timing, resource usage, and data movement. We seek engineers comfortable working from RTL to runtime. They consider pipelines and hardware performance and enjoy implementing dataflow architectures in silicon and programmable logic. What you'll be doing: Build and implement dataflow pipelines and streaming architectures in FPGA or programmable logic. Develop host-side software, drivers, and runtimes that collaborate with FPGA and accelerator hardware (e.g. PCIe, DMA, VFIO). Partner with compiler and hardware groups to allocate dataflow graphs onto hardware resources; improve latency, processing efficiency, and area/utilization. Build and maintain hardware–software co-design flows: from high-level dataflow specs to synthesis, place-and-route, and validation. Build tooling and methodologies for debugging, pr...