| Last Modified On : | July 2, 2008 5:13 PM PDT |
Rate |
|
by Andrew Binstock
Increased clock speeds and cache sizes will combine with chipsets to provide unprecedented performance at low price points.
Of Intel’s primary chip families, the Pentium® 4 processor and Itanium® processor have recently been favored by considerable press coverage. Meanwhile, their workhorse sibling, the Intel® Xeon® processor, has been quietly driving more 32-bit workstations and servers than any other chip in history.
The Intel Xeon processor’s position as the chip of choice for departmental and small enterprise servers means that it enjoys frequent upgrades from Intel and the regular addition of performance-enhancing features. This article discusses the most recent releases in the Intel Xeon processor family, the new features that should ship this year, and the roadmap for 2004. As will become apparent, Intel has plenty of ideas on how to extend this architecture.
With each generation of microprocessors since the Pentium® II processor’s release in the mid 1990s, Intel has released a sibling chip under the Intel® Xeon® name that has targeted server systems. Hence, there were Pentium® II Xeon® processors and Pentium® III Xeon® processors.
With the release of the Pentium 4 architecture in 2001, however, Intel made a clean separation in the naming of the processor families: the Pentium brand targets single-processor desktops, while the Intel Xeon brand targets multiprocessor workstations and servers.
The Intel Xeon family has been further segregated into two categories: the Intel Xeon processor (for dual-processor machines) and the Intel Xeon processor MP (for systems with four or more processors). Different pin-outs on the physical chips prevent the two models from being used on the wrong systems.
The Intel Xeon chip was the first to ship with Hyper-Threading technology, Intel’s unique feature that enables two execution threads to run simultaneously on the same physical processor. The Intel Xeon chips were also Intel’s first 32-bit processors to sport a 512KB cache. In time, both Hyper-Threading technology and the capacious caches migrated to the Pentium 4 processor, as well.
The most important changes to the Intel® Xeon® processor line will come from acceleration of the processor clock speed and expansion of on-board caches. These features will be supported by some remarkable advances in the chipsets—the group of chips on the motherboard that support the system’s interaction with the central processor. We examine these shortly.
The first major change is the advent of a 1MB L3 cache. Historically, Intel Xeon processors have sported two levels of cache: Level 1 (or L1), which is part of instruction-execution core; and level 2 (L2), which is the 512KB cache referred to previously. The L3 cache is an additional layer that expands cache capacity to a full megabyte. This added cache makes a significant difference in p erformance.
Note that L3 cache is in addition to the L2 cache. Because the L3 cache’s sole function is to feed the L2 cache, however, these levels of cache are not additive. Hence, this processor is correctly spoken of as having 1MB of cache. Larger caches enable faster transactions because more records can be held in cache at the same time, so more transactions can be executed without fetching data from memory or disk. Larger caches also accelerate raw computing because more of a program’s code and data can reside in cache and need not be fetched as the program executes.
This first generation of L3 caches will run at the current top speed for Intel Xeon processors: 3.06GHz. In benchmark results published in the August 2003 issue of Microprocessor Report, early versions of the new Intel Xeon chips obtained scores on tests designed by the SPEC council* that were 15% higher on integer operations and 11% faster on floating-point operations than identical chips without the L3 cache. Big caches make a difference.
Intel previously announced that by year’s end, new Intel Xeon processors will see clock speeds scaled to 3.2GHz, which will provide an additional 5% performance boost. Analysts in the semiconductor industry expect that Intel Xeon processors will exceed 3.5GHz by the middle of 2004. Per its long-standing policy, Intel does not comment on unannounced products.
The front-side bus (or FSB) on these processors handles the movement of data back and forth from main memory to the processor caches. Intel Xeon processors today operate at either 400MHz or 533MHz. These speeds are viewed as sufficient for server applications. On the desktop, where processing of large amounts of multimedia data—especially audio and images—is common, the Pentium 4 processor uses a FSB running at 800MHz. Analysts expect that the Intel Xeon processors will enjoy 800MHz FSBs in 2004 as well—a function of anticipated needs as I/O subsystems we discuss shortly accelerate the movement of large volumes of data.
The high-end Intel Xeon processor MP is used in systems with four or more processors. Because of its orientation toward transaction processing, its design has emphasized large caches, rather than faster clock speeds. For more information on the trade-off between cache size and clock speed, see my article, Intel® Xeon® Processor MP: Bigger Caches, Slower Clocks—What’s the Deal?
The current model of this high-end chip sports a 2MB L3 cache and a clock speed of 2.8GHz. This model will be revved again in the first half of 2004. To this end, Intel has announced an upcoming version with an enormous 4MB of L3 cache and a clock running at an undisclosed speed greater than 2.8GHz. Intel also offers a variety of Intel Xeon processors MP with lower clock speeds and smaller caches. These variants are intended for use in value-priced multiprocessor servers.
Last year, Intel began selling a third line of Xeon processors, the Low Voltage (LV) Intel Xeon processor. These low-voltage microprocessors are intended for use in blade systems, those small PCs that fit into racks—dozens at a time—and consume less power than traditional PCs. This line runs at 2GHz wit h a 400MHz front-side bus. Analysts speculate that Intel will rev this product line as well during the remainder of this year and next year as well.
While processors are the computing core of a system, much of the functionality that supports this work is delivered in the system’s chipset. Historically, Intel has designed and sold chipsets for its single- and dual-processing systems, while helping OEMs design the chipsets for larger machines.
Much of the benefit Xeon® processors will deliver during the next year will be attributable to the chipsets’ ability to deliver data faster through the use of new I/O standards and faster memory. These chipsets will support PCI-X, the 64-bit, high-speed version of the PCI standard that is used today for PC expansion boards and plug-in cards.
In addition, the chipsets will sport the next-generation of double data rate (DDR) RAM, called DDR2. This recently approved memory design is expected to start shipping in quantity in early 2004; it has double the data rate of today’s DDR chips. Systems using these chipsets will tend toward larger amounts of RAM as well, by offering more memory slots that can accommodate higher-density memory chips.
Workstations will support 4GB of main memory, while servers based on Intel Xeon processors will tend toward considerably higher RAM volumes, with 16GB emerging as a defining minimum.
Finally, the chipsets will all support the Serial ATA (SATA) standard. This new specification enables hard disks to move data over disk cables 50% faster than standard IDE disks can today. The next release of SATA is expected to double this capacity again.
The results of these innovations in successive releases of Intel's chipsets during 2003 and throughout 2004 is that systems using them will perform considerably faster than the processor clock speed alone would suggest. For more information on the capabilities Intel expects in next year’s PC and workstation systems and the technology Intel plans to deliver to support those features, see the Desktop Platform Vision Guide.
It’s clear that plenty of upgrades are in store for the Intel® Xeon® processor line and for the supporting chipsets. Beyond performance, what makes this story compelling is pricing. The Intel Xeon processors, in particular, cost little more than their Pentium 4 brethren, and their ability to be doubled up in a system provides some key benefits.
Through Hyper-Threading technology, Dual Intel Xeon processors actually provide four paths of execution. This makes the chips particularly attractive for small servers, such as those that run departments and small businesses, where thread parallelism delivers important performance benefits.
Machines based on Intel Xeon processors also make a lot of sense in clustered systems where they can deliver high-performance computing (HPC). Such clusters are highly available systems consisting of several machines that appear to applications as a single server.< br />
HPC clusters distribute computing among themselves in a variety of ways and then fold back results into a single completed computational event. (This distinguishes them from traditional clusters that tend to divide work in simpler ways or serve as redundant fail-over systems for purposes of high availability).
Due to its high performance and large caches, clusters based on Intel Xeon processors MP currently hold all ten of the top ten places in the industry-standard TPC database benchmarks in the categories of performance and price/performance. Third-party vendors are also using the dual-processing Intel Xeon chip as the basis for very low cost/high-performance HPC.
In view of Intel’s continued expansion of the capabilities of these chips and its long-term roadmap for their development, it’s clear that parallel, distributed, and multiprocessor computing will increasingly become a standard staple of business systems at price points that were inconceivable only a few years ago.
The following pages are useful resources for more details on the topics discussed here:
Intel, the world's largest chipmaker, also provides an array of value-added products and information to software developers:
- Intel® Software Partner Program provides software vendors with Intel's latest technologies, helping member companies to improve product lines and grow market share.
- Intel® Software Network offers free articles and training to help software developers maximize code performance and minimize time and effort.
- Intel Software Development Products include Compilers, Performance Analyzers, Performance Libraries and Threading Tools.
- Intel® Solution Services is a worldwide consulting organization that helps solution providers, solution developers, and end customers develop cost-effective e-Business solutions to complex business problems.
- IT@Intel, through a series of white papers, case studies, and other materials, describes the lessons it has learned in identifying, evaluating, and deploying new technologies.
- The Intel® Software College provides a one-stop shop at Intel for training developers on leading-edge software-development technologies. Training consists of online and instructor-led courses covering all Intel® architectures, platforms, tools, and technologies.
Andrew Binstock is the principal analyst at Pacific Data Works LLC. Previously he was the director of PriceWaterhouseCoopers&r squo;s Global Technology Forecasts. His book “Practical Algorithms for Developers” co-written with John Rex is in its 12th printing at Addison-Wesley and in use at more than 30 university computer-science programs. Binstock can be reached at abinstock@pacificdataworks.com.
