While Microsoft Corp. attempts to sway buyers on the ease-of-use and features of Windows XP, both Intel and Advanced Micro Devices will be using the intricacies of computer code to sway OEMs.
At the heart of the estimated $500 million joint marketing program between Intel and Microsoft is an effort to convince customers the two companies are inextricably linked. Intels main argument is that the WinXP kernel and other software contain specific optimizations for the Pentium 4 that improve performance.
Executives at Microsoft confirmed that Microsoft developed and optimized the Windows kernel for both the AMD Athlon as well as the Pentium 4. On the surface, however, it appears Microsoft has done more work on the Pentium 4, as so far the new instructions in the Athlon XP are not supported, Microsoft officials said.
At press time, whether or not specific optimizations exist for Transmeta Corp.s Crusoe and the Via Technologies Inc. C3 processor line could not be confirmed. According to David Ditzel, vice-chairman and chief technology officer of Transmeta, Santa Clara, Calif., the companys code-morphing software will provide all the optimizations users will need.
In and of itself, Windows XP is designed to be faster than both Windows ME and Windows 2000, regardless of the microprocessor used.
"Microsoft has been working for more than 20 years to make software the best possible tool for users, which has culminated in the release of Windows XP today, an operating system that takes Windows and the PC to new heights," Bill Gates, chairman and chief software architect for Microsoft, told several hundred attendees at the Marriott Marquis Theatre in Manhattan on Thursday.
Although the new operating system may not be the panacea for the American economy some hope it will be, courting the new OS could bring a lucrative dowry to both Intel and AMD.
On one hand, Microsoft will partner with Intel and hold "demo days" together at local retailers on weekends, touting the performance of Windows XP on the Pentium 4. The two companies will also hold joint seminars, continue to brief press and analysts, and publish information on the Web. A joint logo describes the collaboration: "Windows XP with Pentium 4 processor optimizations".
On the other, executives at AMD say the Athlon XP will provide "extended performance" to users of Windows XP, and that its processors are faster overall.
Intel boasts Windows XP optimizations in seven specific areas, according to executives speaking at a workshop Intel and Microsoft held recently in Burlingame, Calif. Within XP, Microsoft supports Intel-specific code in the DirectX 8.0 API; Intels SSE and SSE-2 instruction set, used by the Pentium 4 and within the core OS; a special "denormal to zero" mode; the system call interface; the TCP/IP interface; memory page zeroing and memory buffer copy instructions; and within Microsofts extended machine-check architecture support.
When notified of Intels marketing efforts, a spokesman said AMD executives were "incensed", and that countering Intels claims would be "top priority".
Those executives at AMD claim that an AMD-specific "code path" has been architected within the Windows XP kernel, with specific optimizations within the kernel, the DirectX API, and most recently an updated Windows Media Encoder. In addition, WinXP laptops running AMD processor can use AMDs PowerNow! feature. Microsoft executives confirmed these basic optimizations, but said a more detailed explanation will be forthcoming.
Measuring performance in XP, however, can be tricky.
Not only do boot times decrease, but the OS self-optimizes the way in which applications are loaded, "caching" the most frequently-accessed applications on the outside tracks of the hard disk, where they can be loaded faster.
Because of this, the performance increases may be somewhat subjective. For example, although PC Magazine tests show that a 1.4-GHz Athlon starts up under Windows XP a second slower than a 2.0-GHz Pentium 4, analysts viewing a demonstration a week and a half ago were unimpressed with the Pentium 4s performance.
"It doesnt take nearly as long when running on an Athlon," commented Nathan Brookwood, an analyst with Insight 64 in Saratoga, Calif. Another analyst, speaking privately, complained of his frustration testing Windows XP systems, as the OS seems to persistently interrupt the automated processes used in benchmarking with dialogue box queries.
Currently, the Athlon XP and Pentium 4 are about neck-and-neck in performance under Windows XP, with the two processors jockeying for the lead in ExtremeTech tests of various content-creation, media, and game applications.
Compared to Windows ME, Intel executives say theyve seen anywhere from a 7 percent to well over 30 percent improvement simply by changing the operating system. Applications can also take advantage of the SSE optimizations in the OS kernel, improving their performance. Since the Athlon XP uses SSE-1 compatible instructions, it can take advantage of this as well; however, the Athlon XP does not include an SSE-2-compatible instruction set.
Within DirectX, calls to transformation and lighting operations can be assisted by the second generation of Intels Streaming SIMD Extensions, found within the Pentium 4. The Pentium 4 also accelerates the "fast system calls using the sysenter/sysexit instructions first introduced in the Pentium II, and accelerated within the Pentium 4, according to Intel.
In networking, the Windows XP OS uses SSE-optimized checksum routines that will be accelerated by prefetching them using either the Pentium III or Pentium 4.
The Windows XP operating system also uses the SSE-2 instructions for non-temporal memory storage using the "movnti" instruction, for "Store Doubleword Using Non-Temporal Hint". "Non-temporal memory storage refers to a situation where a programmer wants to keep certain data in the L1 and/or L2 (caches) and therefore issues the movnti instruction to retire the result of an operation directly to main memory without keeping a copy of the result in the internal caches," according to Intel documentation.
In certain cases, the operating system will also attempt to make calculations with very small numbers. The Pentium 4 eliminates these calculations, returning an answer of "0" back to the application where the exact numbers themselves may not be needed.
AMD executives, meanwhile, hotly deny the Athlon architecture is in any way being shortchanged by Windows XP. "Let me be very clear—our optimizations are in there," said Patrick Moorhead, vice-president of customer advocacy for AMD.
Specifically, Moorhead claimed that DirectX 8.0 has been optimized for Athlon microprocessors, and that support exists for the 3DNow! and 3DNow! Professional instructions included in the Athlon XP. PowerNow! power-management technology is now supported natively in Windows XP, Moorhead said, and an enhanced version of the Windows Media Encoder 8.0 application is now available.
Thats true, to a point, agreed Mark Croft, lead product manager for the "PC experience" at Microsoft, Redmond, Wash. Windows XP supports 3DNow! PowerNow! and even multi-processor support for the Athlon MP, but not the 3DNow! Professional instructions of the Athlon XP. Windows XP will run on the Athlon XP, but will not take advantage of the additional 3DNow! Professional instructions.
"We had planned to ship (3DNow! Professional) inside the box, but now well have to make them available in a future version," Croft said.
On Thursday, Croft said a list of specific AMD code optimizations will be released, most likely within the next couple of days.
Executives at Transmeta, meanwhile, said the debate is meaningless since both the OS and the Transmeta Crusoe self-optimize themselves over the course of running an application. The Crusoe does not support the SSE instruction set. "All the (software) vendors, they hate it," said David Ditzel, the companys CTO, of processor-specific software optimizations. "They—I—think thats the wrong approach. All the old software is useless that way."
The companys code-morphing software version 4.3, due next year, will be designed to "profile" Windows XP, Ditzel said. The CMS software analyzes the code, translating it into internal instructions and then rewriting it to optimize it in future uses. An unnamed next-generation Crusoe, due at about the same time as the integrated TM6000 chip, will have "very different optimizations," he said.
In any event, it seems like software vendors have a choice, Ditzel added. "How many (applications) are you going to rewrite to do processor optimizations?" he asked. "One says roll your (code) loops, the other says unroll your loops. Which are you going to do?"
Croft said that it was his opinion that Windows XP did not contain any optimizations for either the Crusoe or the Via C3 processor, but that he would recieve confirmation within the next day or so.