Intel's six-core and eight-core server CPU architecture analysis

March 17, 2008 - Today, Intel is the first to talk about its upcoming industry-leading microprocessors and technologies. Intel's leading 45nm high-k metal gate process technology is driving products from all segments of the industry to join the era of multicore processors. Intel today introduced its future products and technologies in quad-core, six-core, eight-core and more computing cores.

Pat Kissinger, Senior Vice President and General Manager, Digital Enterprise Division, Intel:

Kissinger talks about Intel-based six-core processors (R&D code "Dunnington" ) The details of the multi-channel (MP) server and the new Itanium processor (R&D code "Tukwila"). Kissinger also explores current hot topics in enterprise IT applications such as virtualization and the new SPECpower benchmark. The benchmark is designed to measure server performance, while the top 20 servers in the review are based on Intel architecture. He also introduced a series of technical features of Intel's next-generation processor family Nehalem and future multicore products Larrabee.

"Dunnington for Scalable (Multi-Channel) Servers - Intel's existing Xeon 7300 series server platform (codenamed "Caneland") is ideal for multi-channel (MP) server virtualization platforms in the industry The platform is built on the new quad-core Intel® Xeon® 7300 series processor and Intel 7300 chipset. The Dunnington platform, which will be available in the second half of 2008, is fully compatible with the Caneland platform slot. Dunnington is the first. A six-core processor based on the IA architecture, based on a 45nm high-k process with a large shared cache. In addition, the processor supports FlexMigration technology, allowing a single compatible virtual pool at 65nm and 45nm high -k Intel Core Microarchitecture Server and real-time migration of virtual machines (VMs) based on different generations of 45nm processor servers. This not only provides investment protection for customers, but also allows users to flexibly choose the most suitable server platform to maximize optimization. Performance, cost, power and reliability.

" Tukwila delivers excellence for the world's most powerful computers Yes - Tukwila is Intel's next-generation quad-core Itanium processor with up to 30MB of cache, high-speed interconnect with QuickPath, dual integrated memory controllers and mainframe-class RAS features. It is the world's first microprocessor with 2 billion transistors, and we expect its performance to be more than double that of existing Itanium processors.

" Nehalem is Intel's new dynamic and expandable processor microarchitecture - Nehalem can significantly improve the performance and energy performance of Intel's current industry-leading microprocessors. With the release of future versions, Nehalem The architecture will be applied to processors including dual-core, quad-core, six-core and eight-core, and with 4 to 16 thread processing capabilities via Simultaneous Multi-threading technology. Nehalem will be available Four times the memory bandwidth of systems based on the current best-performing Intel Xeon processors. With up to 8MB of L3 cache, 731 million transistors, Quickpath high-speed interconnect (up to 25.6 GB/s), integrated memory controller and With optional integrated graphics, the Nehalem architecture will eventually be applied to all processors from laptops to high-performance servers. The architecture also supports DDR3-800, 1066 and 1333 memory, SSE4.2 instruction set, 32 KB instruction cache, 32 KB data cache, 256 K two-level low-latency data and instruction cache per core and a new second-level TLB (Translation Lookaside Buffer) structure. These technical improvements can greatly improve the performance and flexibility of various processors based on the Nehalem architecture. In addition, Kissinger also discusses the new Tylersburg platform, which can be configured to support a single high-end desktop (HEDT) ) and the operation of dual-channel (HPC and two-socket server) systems.

"Visual Computing: Graphics Redefinition--Visual Computing is rewriting the visual experience and immersive HD experience of computer users. The next generation of technology will provide a natural and realistic gaming experience, graphics effects and high-definition video and audio, placing even greater demands on the performance and architecture of the computer. For example, illuminating techniques such as ray tracing can be used to provide accurate shadows and lighting effects, which puts higher performance requirements on computers than traditional graphics cards. The fidelity of behavior in the application (such as realistic actions in the game or the real performance of human motion in medical imaging) also drives the need for more general computing. In the end, people will enjoy a completely different interactive experience. For example, a new game controller that understands human motion will make the user the favorite player in the game; in the medical imaging field, the sensor carried by the patient will transmit information in real time, helping the doctor to perform interactive calculation-assisted medical work. To provide visual computing, we need a complete platform that includes multi-core CPUs, chipsets and graphics cards, as well as software and related developer tools. To this end, Intel continues to increase its investment in developing related technologies, products and platforms and accelerate the development process to meet the needs of visual computing forward stepping.

"Larrabee architecture for visual computing - Larrabee architecture will be Intel's next architecture for developing visual computing platforms, and is scheduled to be demonstrated for the first time later this year. The Larrabee architecture includes a high performance A wider single instruction multiple data (wide SIMD) vector processing unit (VPU), and a new set of vector instructions (including integer and floating point algorithms, vector memory operations and conditional instructions). In addition, Larrabee also uses a new based Hardware coherent cache design to support more processing core architectures. The architecture and instruction set are designed to meet the requirements of workloads that require parallel computing, including visual computing, and It provides sufficient performance, excellent power consumption, and general programmability. Development tools are critical to the success of this architecture, so we also need to optimize key Intel® software products to support Larrabee Architecture and unparalleled freedom for developers. Larrabee-based products will support Dire Industrial Application Programming Interface (API) such as ctX? and OpenGL.

"Intel AVX: The Next Step in the Intel Instruction Set - Kissinger also talks about the Intel Advanced Vector Extensions (AVX), Advanced Vector Extensions ). By using this instruction set, software programmers can improve the performance of running floating point, media, and processor-intensive software. AVX also improves system performance and is backward compatible with existing Intel processors. Its main features include a wider vector, and the width is increased from 128 bits to 256 bits, which doubles the floating point peak output. . Enhanced data reorganization capabilities support more efficient drag and drop of data, three operands, and non-destructive syntax, giving you a range of benefits. Intel will announce detailed specifications at the Intel Information Technology Summit in Shanghai in early April. The instruction set is planned to be deployed in the microarchitecture codenamed "Sandy Bridge" launched in 2010.