Myrtle.ai MAU Accelerator for AI Financial Trading Models
REFERENCE DESIGN MAU Accelerator for AI Financial Trading Models Ultra-low Latency, High Throughput Machine Learning Inference Well suited to a range of applications in financial
Corundum
Open-Source 100G NIC IP Core
Jump-start your SmartNIC project with the open-source Corundum IP core! It’s got the basic framework for FPGA-based network acceleration, ready for you to adapt and extend. The expectation is for users to add their own custom logic and host software to build SmartNICs, saving time by using the foundational network packet processing elements in Corundum. A user application area allows for easier targeting of multiple card types.
What sets Corundum apart from other open-source FPGA NIC frameworks are features like scatter/gather, PTP, and robust host DMA integration. The overall performance targets general-purpose packet processing applications at line rates up to 100 Gb/s.
The IP core targets BittWare FPGA cards with AMD Virtex UltraScale+ and Altera Agilex 7 devices. Specific supported cards are listed under specifications. As an open-source project, customers may use the IP to port to other BittWare cards.
Open Source for Commercial Projects
Fine-Grained Traffic Control
Designed for Customization
Product Description
Features
- Open-source FPGA-based NIC
- PCIe Gen3 x16, multiple 10G/25G/100G Ethernet ports
- Fully custom DMA engine; Linux driver
- Fine-grained traffic control
- 10,000+ hardware queues, customizable schedulers
- Application block for custom logic
- Access to network traffic, DMA engine, on-card RAM, PTP time
- PTP timestamping and time synchronization
- Management features (FW update, etc.)
Data Sheet and Product Details
Applications
- Offload application-specific processing
- Datapath for novel transmit schedulers
- Instrument Corundum for performance measurements
- Direct transceiver access permits physical-layer measurements and development of new wire protocols
- Use core logic as a packet DMA engine in a larger system
Detailed Feature List
Fine-Grained Traffic Control
- 10,000+ transmit queues
- Each queue is an independent channel between SW and HW
- Classify in SW, control in HW
- Fine-grained, per-flow or per-destination control
- 128 bits/queue -> 4096 queues in 2 URAM on US+
- Transmit scheduler
- Determines which queue to transmit from
- Default scheduler is round robin
- Can be used to implement traffic shaping, rate limiting, etc.
Open Source
- Corundum core logic 100% open-source Verilog
- Including 10G/25G MACs, PCIe DMA, AXI, AXI stream, etc.
- Device hard IP used for interfacing – PCIe, 100G MAC, serdes
- Simulation uses open-source tools
- Cocotb + Icarus Verilog/Verilator
- Cocotb extensions for AXI, Ethernet, and PCIe
- Tox + pytest for CI
- Makefiles for build automation
Ports and Interfaces
- Hardware support for multiple uplinks
- Multiple physical ports appear as single OS-level interface
- Ports have separate schedulers
- Migrate or stripe flows across ports by changing scheduler settings
Traditional NIC: Assignment in Software
Corundum NIC: Assignment in Hardware
Modular DMA Engine
- DMA engine split between interface and client modules
- Interface connects to host – PCIe, AXI, etc.
- Client modules form internal ports – AXI stream, memory-mapped AXI
- Clients connected to interface with dual port RAMs
- Support both servers (PCIe) and SoCs (AXI) with same core logic
Corundum
About the Project
Corundum is an open source FPGA-based NIC and platform for in-network compute.
Interested in Pricing or More Information?
Our technical sales team is ready to provide availability and configuration information, or answer your technical questions.
