Products

Silicom Denmark | FPGA Solutions

fbSmartNIC1.0 Samrt Card

Dual Port Smart Card

The quest for ever increasing system performance is propelling the adoption of FPGA technology in Network Interface Cards (or NICs). As the network speed has been increasing from 1GE to 100GE, FPGAs have proven their ability to process massive amounts of network traffic in real-time and bring a level of determinism and consistency that is not possible with other technologies.

The Silicom Denmark fbSmartNIC product family has the most comprehensive selection of programmable Ethernet cards designed to competitively enhance performance in a host of networking, financial and big data solutions. The fbSmartNIC consists of highly optimized functionality well suited for e.g. financial algorithmic trading platforms where sub-micro second latency is crucial.

The complete transport layer offload can be combined with customized logic interacting directly with the Silicom Denmark network cores. This is an efficient way for the advanced user to achieve even lower latency and to implement any filters or acceleration that are necessary for a specific application. The Silicom Denmark solution comes with a full Software API for a seamless integration and monitoring.

 

All fbSmartNIC products are supported with our hand-optimized SDK and device drivers which allow our customers to start developing their high-performance applications right out of the box. Layers 3 and 4 network capabilities are provided using our own industry-proven MAC and TCP/IP solutions. The fully integrated offering makes Fiberblaze a true one-stop solution provider and helps customers save on development and total cost of ownership.

Fiberblaze cards are available in variety of form factors and support 1G to 100G Ethernet network speeds, both Xilinx and Altera FPGAs, latest generation of PCI Express host interface, and a variety of memory configurations suitable for most applications.

fbSmartNIC1.0 Samrt Card

Dual Port Smart Card

Host Interface:

  • 64 logical channels connected to DMA

Latency:

  • Wire to wire latency of 380 ns: Receiving on UDP port, through User Logic, and sending from TCP port
  • Full 10G throughput on TCP port

MAC:

  • Combined MAC (10GbE) and PCS (10GBASE-R) IP core
  • Fixed latency of 101 ns: User Logic to wire, Tx + Rx

Performance:

  • Data rate: 10 Gbit/s
  • TCP/UDP throughput at line rate

Time Stamping & Sync:

  • Resolution = 1ns
  • Accuracy down to 20 ns

Environment:

  • At least half of the FPGA resources are available for User Logic

UDP Offload Engine:

  • 64 UDP multicast streams accessible to Host and User Logic
  • IGMP support

TCP Offload Engine (TOE):

  • Up to 8 TCP connections accessible to Host and User Logic
  • Support up to 2K bytes Ethernet frames

User Logic:

  • 8 dedicated DMA Channels to/from Host System at full PCIe speed
  • 256KB of register access
  • All TCP and UDP streams accessible
  • Access to all physical ports

Software API:

  • Multi DMA channel streaming (64 channels) with host dependent DMA buffer
  • C based API (DLL/Shared library). Windows, Linux & FreeBSD (on request)

On Board Sensor Readings:

  • Current fbSmartNIC1.0 System status available in Software API
  • Temperature with present minimum, maximum card operating temperature
  • Link status
  • TCP State information

Standard NIC:

  • Full Standard NIC Support
  • Configurable to all available ports running alongside TOE and UDP Engine
  • Data path also accessible to User Logic

Statistics:

  • Statistics are based on a subset of RFC2819 RMON1
  • Statistics are provided on a per second basis:
  • Network counter: Number of Octet, CRC Align Errors, Undersize Packets, Oversize Packets, Jabbers, etc.

Monitor Channels:

  • All outgoing packet can be looped back to Host through PCIe for monitoring or logging purposes

fbSmartNIC1.0 Samrt Card

Dual Port Smart Card

General Technical Specifications

Host Interface: 64 logical channels connected to DMA
Latency: Wire to wire latency of 380 ns: Receiving on UDP port, through User Logic, and sending from TCP port
Full 10G throughput on TCP port
MAC: Combined MAC (10GbE) and PCS (10GBASE-R) IP core
Fixed latency of 101 ns: User Logic to wire, Tx + Rx
Performance: Data rate: 10 Gbit/s
TCP/UDP throughput at line rate
Time Stamping & Sync: Resolution = 1ns
Accuracy down to 20 ns
Environment: At least half of the FPGA resources are available for User Logic
UPD Offload Engine: 64 UDP multicast streams accessible to Host and User Logic
IGMP support
TCP Offload Engine (TOE): Up to 8 TCP connections accessible to Host and User Logic
Support up to 2K bytes Ethernet frames
User Logic: 8 dedicated DMA Channels to/from Host System at full PCIe speed
256KB of register access
All TCP and UDP streams accessible
Access to all physical ports
Software API: Multi DMA channel streaming (64 channels) with host dependent DMA buffer
C based API (DLL/Shared library). Windows, Linux & FreeBSD (on request)
On Board Sensor Readings: Current fbSmartNIC1.0 System status available in Software API
Temperature with present minimum, maximum card operating temperature
Link status
TCP State information
Standard NIC: Full Standard NIC Support
Configurable to all available ports running alongside TOE and UDP Engine
Data path also accessible to User Logic
Statistics: Statistics are based on a subset of RFC2819 RMON1
Statistics are provided on a per second basis:
Network counter: Number of Octet, CRC Align Errors, Undersize Packets, Oversize Packets, Jabbers, etc.
Monitor Channels: All outgoing packet can be looped back to Host through PCIe for monitoring or logging purposes

fbSmartNIC1.0 Samrt Card

Dual Port Smart Card

P/N

Description

Notes

fbSmartNIC1.0    

 

1V0