NVMe IP Core
PCIe Gen4 / Gen3 · UltraScale+ · high-rate sequential writes
An NVMe host engine built for extreme ingest: deep, continuous writes from AXI streams direct to drive, no OS or CPU in the write path. Proven at roughly 85 Gb/s sustained in a real camera application with RAID0 striping across multiple drives.
ZCU106 · Aper-Oculus — build scripts, constraints, and a working RAID0 writer pipeline.
| Quick reference | |
|---|---|
| PCIe | Gen4 / Gen3 endpoint |
| Targets | Xilinx UltraScale+ |
| Sustained writes | ≈85 Gb/s (RAID0 ref.) |
| Ingest | AXI4-Stream → NVMe |
Table 1 — Key parameters. Full electrical characteristics under Specifications.
PCIe Gen4 / Gen3
Optimized for Xilinx UltraScale+ devices using the native PCIe hard block, tuned for high-throughput sequential transfers.
NVMe queue engine
Submission/completion queue pipeline with PRP/SGL handling for large, contiguous writes from streaming sources.
Camera → RAID0 writes
Proven in a real ≈85 Gb/s camera application with RAID0 striping across multiple NVMe drives for sustained record throughput.
Low overhead
Tightly pipelined queue and DMA handling with tunable deep FIFOs and backpressure for continuous streams.
Documentation
Need extreme record bandwidth?
Integrate a hardware-proven NVMe pipeline on UltraScale+ for sustained ~85 Gb/s RAID0 writes.