Sixth generation of flow processors packs 216 processing cores and 100 accelerators into fully programmable device based on Intel 22nm 3D Tri-Gate technology
Netronome, the leading developer of flow processors, today disclosed the details of the NFP-6xxx family of processors. The NFP-6xxx product line is the sixth generation of flow processors and is designed for high-performance networking applications that require programmable, line-rate, stateful processing on millions of simultaneous, complex flows. The NFP-6xxx extends Netronome’s market-leading position in flow processing and provides a software-compatible migration for customers of Netronome’s current NFP.
A series of patented architectural innovations combined with the Intel 22nm 3D Tri-Gate manufacturing process result in the industry’s first programmable 200Gbps flow processor. The NFP-6xxx product line is ideally suited for intelligent line cards, service blades and PCIe cards in applications such as carrier networking, cyber security, software defined networking including OpenFlow, mobile networking, virtualized datacenters and cloud computing. These applications share a common set of requirements that include increasing bandwidth, extensive processing per packet, stateful flow processing and a programmable approach to keep pace with the rapid rate of change in protocols and threats.
“Netronome has established a clear leadership position in flow processing with its heterogeneous architecture complementing x86 processors,” said Bob Wheeler, senior analyst at The Linley Group. “The advances in the NFP-6xxx further strengthen its position as an alternative to integrated multicore processors while also enabling it to directly compete against network processors in standalone configurations.”
The scalable NFP-6xxx product line targets designs from 10 to 400 Gbps based on the following features and supported flow processing building blocks:
- 256 Gbps Processing Complex: 216 fully-programmable, workload-specific cores, featuring 120 L2-L7 flow processing cores with 960 threads, and 96 L2-L3 packet processing cores, delivering over 307 billion instructions per second to 384M packets per second. Based on Intel 22nm process technology, the NFP-6xxx provides over six times the packet processing performance at less than half the power compared to 28nm-based processing alternatives.
- 100 Dedicated Accelerators: Programmable hardware accelerators for computationally intense networking tasks including deep packet inspection, regular expression matching, optimized I/O packet transfers, traffic management, security processing and bulk cryptography.
- Optimized Memory Architecture: Over 30 MB of on-chip storage, that includes Proximity Memory™, eliminates the need for expensive external synchronous SRAM and enables packets to be forwarded from one network interface to another without requiring the packet to be stored in external DDR3 memory. The internal memory also facilitates fast transfer between processing elements and powers the patented adaptive memory controller for highly efficient DDR3 transactions at 2133 MHz across six channels. Combined, these features provide ultra-high memory bandwidth and utilization, reducing memory latency and increasing processor efficiency.
- 50 Gbps Security Processing: Hardware-based, in-line security processing and bulk cryptography support for all major cipher suites and key sizes reduces system design complexity, cost and power while ensuring wire-speed security processing without degrading data plane performance or requiring external out-of-band hardware co-processors.
- 720 Gbps of Processor I/O: 72x10G KR-capable SERDES interfaces anchor high-bandwidth access to the chip, including 10G/40G/100G Ethernet with integrated MACs and 10G/40G/100G Interlaken. The high-density I/O removes the need for separate aggregation devices and enables line-rate 200 Gbps designs with optional support for 4-to-1 oversubscription. The NFP-6xxx family also supports dedicated 4x8 PCI Gen3 interfaces for applications demanding high bandwidth, dynamic load balancing to Intel x86 processors.
- Easy and Flexible Programing Tools: The NFP-6xxx is supported by the industry’s most comprehensive suite of software development tools, including a graphical integrated development environment, C-compiler, packet generator, and cycle- and timing- accurate simulators. Developer’s may program in C or take advantage of APIs to Netronome’s flow management software. Additionally a wide range of production-ready hardware reference designs are available to assist in both hardware and software design.
- Extensive Software Libraries: The NFP-6xxx is complimented by the industry’s broadest set of software libraries to accelerate time to market and reduce development costs. Netronome’s flow management software provides comprehensive code for leading applications such as: intrusion detection and prevention, next-generation firewall, OpenFlow, SSL inspection, IPsec, IP/MPLS carrier networking, data loss prevention, network forensics, deep packet inspection and application identification, stateful load balancing, IP test and measurement, intelligent network taps, and high frequency trading.
“We experienced tremendous growth last year as a direct result of our early commitment to flow processing and its supporting building blocks. The NFP-6xxx strengthens this commitment and extends the benefits of high throughput flow aware processing to our leading edge customers,” said Howard Bubb, CEO of Netronome. “With a 10x increase in flow processing capabilities, the NFP-6xxx will accelerate the movement from packet to flow-aware processing for a wide range of advanced network services.”
Netronome will make the first public technical presentation of the NFP-6xxx during the Linley Tech Carrier Conference, June 5-6, 2012 at the Double Tree Hotel in San Jose, California. The NFP-6xxx will be available to customers in mid-2013.