Breakthrough NVMe Performance on Zynq UltraScale+: 7 GB/s with DG NVMe Gen4 IP on PetaLinux

High-speed storage is no longer optional — it’s mission-critical.

At Design Gateway, we are proud to introduce a game-changing NVMe storage solution for embedded systems: Achieve up to 7 GB/s throughput on Zynq UltraScale+ FPGA using our NVMe Gen4 Soft IP running on PetaLinux.

🎯 Performance that sets a new benchmark Using our proprietary PCIe Gen4 Soft IP, dual-channel DMA engine, and custom NVMe driver, we’ve pushed NVMe Gen4 SSD performance to the limit — reaching 7,500 MB/s read and 6,900 MB/s write speeds.

🧠 This solution is not just fast — it’s flexible, scalable, and works with any FPGA, even those without built-in PCIe Hard IP.

📽️ Watch the demo now: ▶️ Breakthrough NVMe Performance on Zynq UltraScale+

⚙️ How We Achieved 7 GB/s on Zynq UltraScale+ (ZCU106)

Most embedded Linux systems using standard NVMe drivers and PCIe Gen3 Hard IP struggle to break 2 GB/s due to software overhead and architectural bottlenecks. Our solution completely redefines the pipeline:

  • We replaced the PCIe Hard IP with our PCIe Gen4 Soft IP, fully synthesized into logic fabric.
  • Data transfer is accelerated via dual DMA channels, offloading CPU tasks entirely.
  • Custom NVMe drivers eliminate Linux kernel bottlenecks, enabling full-speed data transfers.
  • Our compact IP core supports any FPGA with PCIe resources, ensuring platform flexibility.
Engineers relying on traditional NVMe access via PCIe Gen3 Hard IP on Zynq® UltraScale+ MPSoC platforms have typically seen performance capped around 2 GB/s due to straightforward hardware setup, and software overhead from embedded Linux environments like PetaLinux.
With Design Gateway’s PCIe Gen4 Soft IP + Custom NVMe Driver, we’ve redefined the performance boundary—reaching up to 7 GB/s, even on the same FPGA platform.
NVMe Gen3 vs NVMe Gen4 Performance Comparison
The NVMeG4 IP core features a standalone NVMe protocol stack, integrated PCIe Gen4 Soft IP, and utilizes RAM as an internal buffer—eliminating the need for external CPUs and memory. It connects directly to an NVMe Gen4 SSD via a 4-lane PCIe interface implemented on the FPGA, delivering high-speed data transfer. This architecture simplifies NVMe Gen4 integration and is backed by our proven reference design on Zynq UltraScale+ MPSoC with PetaLinux.
Performance

📌 Measured performance:

  • Write: 6,900 MB/s
  • Read: 7,500 MB/s (Tested on ZCU106 with AB17-M2FMC adapter and NVMe Gen4 SSD)

📦 Use Cases

There are many types of applications that need high-speed access to NVMe storage.
• High-speed data logging and acquisition
• Real-time video processing and storage
• High-performance Edge Computing
• Edge Server requiring ultrafast storage
NVMe Storage Application

This solution is ideal for:

  • 🔬 AI & Edge computing
  • 🛡️ Defense & Aerospace
  • 🚗 Automotive data logging
  • 🎥 Real-time 4K/8K video processing
  • 📉 High-frequency trading
  • 🧪 Scientific research platforms

📚 Resources

🌐 Learn more about DG NVMe IP Core
👉 Free evaluation file
👉 AMD Partners
🛒 Buy Adapter Boards at Mouser

✅ Ready to Accelerate Your Storage?

Whether you’re building a real-time data acquisition system or designing high-performance embedded platforms, our NVMe Gen4 Soft IP unlocks true PCIe Gen4 speed on FPGA — without the usual hardware constraints.

📩 Contact us today to evaluate or integrate this breakthrough into your next project.
🌐 Visit: https://design-gateway.com

🔁 Like | 💬 Comment | 🔗 Share
Let’s drive the future of high-speed FPGA storage — together.