After FPGA pre-processing, there are many streaming data sources customers want to capture or record, but achieving at high speed (100 Gb/s) is challenging.
BittWare offers Reference Designs for users who have their own FPGA engineering team and application IP, but would benefit from having a proven application framework—the Reference Design. This allows customers to focus on their application-specific IP.
These Reference Designs use a small part of the FPGA, leaving room for customer IP like filtering, decimation, signal processing and more. Our customers have requirements of 40 Gb/s or greater and minutes to hours of recording time at line rate.
Learn more about 100G Data Capture with our white paper and download the detailed App Note describing its operation. For BittWare customers with Xilinx-based products, you can also download the Capture project on the Developer website.
For Data Capture, sensor data passes through the FPGA and CPU to host DDR4 DRAM memory.
The capture depth is typically limited to a few seconds for high data rates. DDR4 DRAM is relatively expensive.
For direct Data Record, sensor data passes through the FPGA directly to the SSD NVMe FLASH via a Molex cable assembly.
The record depth can be many hours at high data rates using standard off-the-shelf SSDs.
BittWare’s Data Capture and direct Data Recorder reference designs are based on a network packet generator and capture design using the 250-SoC accelerator card. This is a dual 100GbE product, however BittWare has a rich portfolio of other FPGA cards allowing data ingress from other sensor sources. Examples include: digital I/O, multi-channel RF ADC and video. Please contact BittWare for details.
This approach uses the standard PCIe interface of a BittWare FPGA card for data capture to host memory, and record to FLASH NVMe SSDs at 100 Gb/s.
This reference design is included free with applicable BittWare hardware as described in the Deliverables. The majority of FPGA resources reserved for customer application IP. Customers can program their application using their preferred tool flow.
Click here to learn more about Data Capture in our white paper.
This is an extension of Design #1: using a second PCIe interface, we can achieve up to approximately 200 Gb/s to host memory. This is achieved by using BittWare’s on-board expansion ports to add a second PCIe connection to the one FPGA card. The second connection goes in an adjacent PCIe slot.
Customers can readily build this configuration through consultation with BittWare.
The Data Recorder takes a different approach to Data Capture. The sensor data is streamed directly from the FPGA to the NVMe FLASH SSDs via a cable assembly that taps into the server mid-plane.
BittWare Recorder IP (subject to license), implemented in the FPGA, deals with NVMe access for the host application running on the CPU. This offloads significant processing burdens from the CPU and implements them efficiently in the FPGA. Customers can then record for many hours at high data rates. The CPU becomes supervisory in function, and therefore can be low cost, low power.
The NVMe drives are still exposed to the host application for optional read/write to the recorded data using standard NVMe drivers.
This reference design is included (conditions apply) with applicable BittWare hardware as described in the Deliverables. The majority of FPGA resources reserved for customer application IP. Customers can program their application using their preferred tool flow.
Details on our NVMe High-Speed Data Capture and Recorder are available upon request. Get in touch with us to learn more!
Watch the recording (including Q&A) to learn about FPGA acceleration for NVMe storage; case studies and a new product announcement!
Watch the recording (including Q&A) to learn about FPGA acceleration for NVMe storage; case studies and a new product announcement!