Groundbreaking Threaded Solution Maximizes Throughput with 32 Processor Resources and Highly Integrated Network and Application Acceleration
CUPERTINO, CA, - May 16, 2005 – Raza Microelectronics Inc. (RMI™) today announced the XLR™ family of Thread Processors™, the industry’s first high performance, general purpose processor designed to meet the growing throughput and multi-tasking demands for connected computing applications. RMI’s new XLR processor is the most powerful, throughput-optimized processing solution shipping in silicon today.
The XLR processor is based on a unique threaded architecture supporting 32 threads on RMI-enhanced MIPS64® CPU cores. With an operating frequency of up to 1.5GHz, it’s further supported by highly integrated networking and application-aware accelerators. The XLR processor is a key enabler for the growing convergence of networking and computing applications -- such as integrated security (Firewall, VPN, Anti-Virus, Intrusion Prevention), web services, virtualized storage, load balancing, server offload and intelligent routing and switching -- in which content-awareness and application intelligence are becoming increasingly important.
“The next generation of Information Infrastructure solutions is creating a new era of high-touch, connected computing systems in which the need for intelligence, flexibility and greater processing power to support and manage heavy workloads and large numbers of flows is outstripping the capability of today’s costly and cumbersome array of processors,” said Aamer Latif, Chief Strategy Officer at RMI. “With the XLR processor, we have created an innovative threaded architecture, with a general purpose programming model, capable of delivering the tremendous throughput and intelligent functionality which, until now, has been unavailable.”
The Optimal Processor Solution
Today’s increasingly network-centric connected computing environment is creating a shift from compute-bound to I/O driven processing. The result is a greater emphasis on efficient access and utilization of processor and system memory. But even the fastest memories simply cannot keep up with today’s high-speed microprocessors. The result is a tremendous amount of idle processing time while the processor is left waiting to access data from the much slower memory. Overall throughput and performance, therefore, are not scaling with the increase in frequency.
Until now, a mix of proprietary and semi-programmable NPUs, ASICs and co-processors have been used to better enable systems for more I/O intensive applications. However, these solutions have created a hardware and software environment that is increasingly complex, expensive and difficult to scale, thus creating a tradeoff between performance and usability. The XLR processor bridges this gap with a highly parallel, single-chip, MIPS64 solution, providing a true C/C++ programming environment that enables wire speed, software-driven applications across multiple platforms using well understood development tools.
Unique Thread Advantage
Key to the XLR processor’s performance and one of its major innovations is its extensive set of 32 threads enabled among 8 enhanced MIPS64-compatible cores at up to 1.5GHz. Each of the cores implements fine-grained, 4-way multithreading. While new processors are clearly moving to multicore, the extension to thread processing is required for a versatile, scalable means of hiding memory latency prevalent in today’s I/O-centric applications. Threading makes this possible by providing a simple means to utilize the available processing time when today’s processors are normally stalled waiting for requested data, making them very inefficient.
“Multiprocessing system-on-chip designs are the wave of the future. To meet the needs of today's increasingly sophisticated multi-service platforms and applications, processing power now scales by enabling extensive parallelism and providing next-generation supporting architectures," said Kevin Krewell, Editor in Chief, Microprocessor Report (an In-Stat publication). "With the XLR Thread Processor, RMI has raised the bar by delivering 32-way threading, outstanding integration, and innovative scalability which will enable a new generation of throughput-optimized connected systems.”
To support its high-performance threaded architecture, the XLR processor incorporates a range of innovative features and design elements enabling optimal throughput and scalability. This includes large sophisticated Level-2 caches, full speed high throughput interconnects, networking accelerators, security acceleration and an interchip messaging complex. In addition, the processor eliminates the bottlenecks of older shared bus architectures by using point-to-point interconnects that scale in frequency and throughput. The benefit of this design is not only in the ability to operate the interconnect at the same frequency as the core but also in its support for multiple transactions per clock.
The XLR processor architecture is designed with a unique eight-banked 2MB cache that can process up to 8 simultaneous requests from any of the processing cores. The XLR processor incorporates a high-performance Security Acceleration Engine, which delivers up to 10Gbps of bulk cryptography and operates autonomously from the core CPUs to maximize processing efficiency. The processor also contains an innovative solution for optimized data communications among all processing agents with a unique high-performance, low-latency Fast Messaging NetworkTM (FMN). The result is a highly scaleable system that can distribute workloads across the processing threads and accelerators.
Complete Product Family
RMI’s new XLR family of Thread Processors includes the XLR732, XLR716, XLR532, XLR516, XLR508 and the XLR308. The XLR300 series integrates 3 Gigabit Ethernet interfaces, a dual 36-bit DDR1/DDR2/RLD2 DRAM controller, as well as a 64-bit PCI-X interface and general purpose I/O. The XLR500 series supports 4 Gigabit Ethernet interfaces with four 36-bit memory controllers along with a HyperTransport interface. The XLR700 series includes all these features plus a standard LA-1 interface and 2 high-speed ports supporting either 10 Gigabit Ethernet (XGMII) or SPI-4.2 interfaces. The last two digits in the product number indicate the number of threads. The XLRn32 processors support 32 threads and 2MB of level-2 cache while the XLRn16 provide 16 threads and 1MB of level-2 cache. Finally, the XLRn08 processors support 8 threads and 512KB of level-2 cache memory.
Price and Availability
The entire XLR processor family has been sampling for 6 months along with evaluation development boards, a complete board support package and software development kit. The XLR processor is available now. Pricing is available upon request.
RMI (Raza Microelectronics, Inc.) is an innovative leader in silicon solutions designed to transform the Information Infrastructure. The company’s portfolio of products provide the foundation for the next generation of scalable Enterprise, Access and Digital Imaging systems. The XLR™ Processor is the first Thread Processor™ for Connected Computing while the Orion™ Intelligent Access™ Processor provides the industry’s first integrated solution delivering advanced packet services over existing transport infrastructure. RMI’s XL™ Processor is the most power-efficient solution optimized for advanced digital imaging applications. As a silicon execution machine, RMI’s unique ability to rapidly deliver semiconductor solutions enables system manufacturers to quickly respond to changing market needs. RMI is headquartered in Cupertino, California with offices and development centers in China, India and Japan. For more information, please visit www.razamicroelectronics.com.