The Multicore CPU Race Is On

Data center consolidation and virtualization may be driving server implementations these days, but there's a much more fundamental shift occurring at the microprocessor platform level.

Systems based on the industry-standard x86 architecture are taking on a predominant role in the overall infrastructure, rivaling the performance and capability of higher-end processors. At the same time, new generations of RISC-based enterprise platforms that run improved multithreaded implementations of Unix are making huge strides in throughput and performance.

The latest in server processors is a constant barrage of next-generation platforms--and a robust pipeline, to boot.

It's important to note that the open-source revolution is redrawing the server-processing landscape. Because customers now can run software on various processor platforms, VARs have the option of moving customers off higher-end Unix systems.

x86/x64 Marks the Spot

Make no mistake. More systems will migrate toward the x86/x64 platform, says Jim McGregor, analyst at In-Stat. "The majority of the market really is geared toward x86, because you have more and more capability coming out of the platform from both AMD and Intel," he says. "My belief is that there's still a market out there [for Itanium, PowerPC and SPARC], but if customers have the resources to convert to x86 now, it would save them more money in the long term."

In 2003, AMD's Opteron was the first 64-bit server chip capable of running 32-bit apps. It propelled the Linux revolution, providing a scalable option to Windows.

Afterward, AMD enjoyed healthy leadership in the x86/x64 field, beating Intel to market with dual-core systems and leading its much larger rival in price/performance. AMD's technical edge helped it win mind share among some key server vendors--most notably, Hewlett-Packard, IBM and Sun Microsystems. Even Dell broke ranks with Intel, its exclusive supplier of CPUs.

Last summer, though, Intel stymied its archrival by releasing its Xeon 5100 processors, the first ones based on Intel's next-generation Core 2 Duo, which supports up to four full instructions simultaneously. The Xeon 5100 also supports Intel's new Advanced Smart Cache, which allows for dynamic allocation of data between the two separate cores. For the first time in years, Intel had a server platform that outperformed Opteron. AMD took a further hit late last year when Intel released the first quad-core x64-based processors, the Xeon 5300.

Now AMD is priming the pump for its quad-core chips, code-named Barcelona. Last month, the vendor expanded its Opteron line with low-power, dual-core Models 1218 HE, 2218 HE and 8218 HE. The new line gives AMD a portfolio of three power bands--120 watts, 95 watts and now the 68-watt thermal envelopes. "We actually think we have the leading performance," says Steve Demski, Opteron product manager at AMD.

But Shannon Poulin, Intel's director of performance and competitive marketing, begs to differ: "We've got quad-core; they don't," he notes.

And Intel says it will continue to trump its competition. Later this year, the vendor plans to go to production with 45-nanometer chipsets (the current is 65 nm), which it says will lead to more power-efficient CPUs early next year.

What's more, the Intel camp is about to receive a benefactor in Sun. The only major server vendor not supporting Intel's x64, Sun shook things up in January, saying it plans to offer its first Xeon-based servers and workstations this year. As reported in "Sun's Rekindled Star," Sun will ship Xeon-based servers this summer. While it will join a crowded field of server vendors offering Xeon-based systems, Sun has a reputation for adding value in highly commoditized markets, especially in the way of cooling and energy efficiency.

While offering Opteron-based x86 servers for nearly four years has given Sun a new foothold in the broader server market, the vendor needed to align itself with Intel in order to maintain credibility and gain market share.

"This is important, because customers do have preferences in terms of which x86 processors they use," says Gordon Haff, an analyst at Illuminata. "At any given time, one of the two vendors [AMD and Intel] will have an edge over the other."

Under the new partnership with Sun, Intel will fine-tune Xeon for Solaris. As a result, Sun will be vying to position Solaris as a viable option to Linux and Windows on the x86 platform.

NEXT: Multithreading with SPARC

With x86/x64-based systems rivaling older RISC-based CPU platforms, it should come as no surprise that enterprise computing has seen some major performance boosts as well. Slightly more than a year ago, Sun started shipping systems based on its UltraSPARC T1 processor, code-named Niagara. Using a multithreading capability that Sun calls CoolThreads, the company says it can double transaction throughput while cutting power utilization in half over the prior SPARC processors.

Sun targets its SPARC processors at mission-critical, transaction-oriented applications, as well as midrange to high-end Web-tier systems. Its core customer base consists of telcos, large companies in the financial-services industry and firms running so-called Web 2.0 applications.

"The demand on infrastructure for such companies is way above Moore's Law, so the traditional sort of approach of adding more megahertz and increasing the memory size on the processors doesn't fit those applications," says Fadi Azhari, Sun's director of marketing for SPARC and chip multithreading, or CMT, technologies.

That's why Sun pursued the course of scaling its SPARC processors using multithreading capability, Azhari says. The next iteration of the T1 processor, dubbed UltraSPARC T2 (Niagara 2), will have the same number of cores--eight--but it will also have double the number of threads per core. As a result, T2 will offer twice the throughput in the same thermal and power envelope, according to Sun. In addition, the T2 will have one floating point unit per core, rather than one per processor. The FPU performs mathematical computations on a CPU.

Itanium, Take Two

Since its release earlier this decade, Intel's Itanium has gotten off to a slow start. But with its second generation, code-named Montesito, it seems to be gaining some momentum. Although Itanium's biggest supporter has always been HP, others that offer servers based on the platform include Bull, Fujitsu Siemens, Hitachi, NEC, Silicon Graphics and Unisys. In 2005, the Itanium Solutions Alliance was formed to develop and promote Itanium-based solutions. Its charter members include BEA, Microsoft, Novell, Oracle, Red Hat, SAP and Sybase.

HP recently gave Itanium momentum by adding the DL8 60C, its first Integrity blade for the new c-Class Blade System.

"We're going after our competition with high-end capabilities at entry-level price points," says Nick van der Zweep, director of virtualization and business-critical systems at HP. "Our objective is to [get] Sun market share in the entry class."

One of Itanium's benefits is its ability to run HP/UX. But Karl Freund, vice president of Systems P marketing at IBM, says much of HP's legacy customer base, including those on PA/RISC running HP/UX, is migrating to the Power platform. "Customers see Power providing the performance they want," Freund says. "Plus they've been burned by a road map that came to an end. They see Power as a viable architecture for the long term."

Sales of Power-based servers are distributed evenly among low-end, midrange and large data-center implementations. At the low end, P Series servers are attractive at branch offices in retail and banking, where reliability is critical.

IBM has offered a quad-core implementation of its Power5 processor since October 2005. IBM positions Power5 for applications that require a lot of throughput on multiple threads, such as Web-facing systems. Conversely, applications that are not highly threaded will not perform as well on a quad-core processor as they will on a processor with a higher clock speed, Freund says.

Clock speed will be the emphasis of IBM's next generation of the Power Processor. Due out later this year, Power6 will have clock speeds ranging from 3 GHz to 4 GHz, and that will optimize workload management specifically as it relates to transaction processing. It will also demonstrate a major boost in integer performance, according to Freund.

In the end, customers are likely to stick with the platform most closely aligned with their existing infrastructures. Software will almost always drive platform decisions. "The value isn't always in the architecture," says In-Stat's McGregor. "It's in the software."

NEXT: The top 5 RISC/Itanium Unix vendors.