To: Tony Viola who wrote (56554 ) 6/1/1998 10:25:00 PM From: rudedog Respond to of 186894
Tony - Here's some info from Client-server news ClieNT Server NEWS | New York and London May 29, 1998 Issue Number 252 FLASH The Independent Observer of Microsoft, Windows NT and Other Phenomena ----------------------HEADLINE---------------------- MERCED DELAYED; Revolution Halted Before It Starts By Maureen O'Gara Friday, May 29 - Intel today confirmed reports that the Merced chip will be delayed until at least the middle of the year 2000. The news makes the company look like it's suddenly suffering from a bad karma, given that the government is bearing down on it and is expected to file antitrust charges against it any day now. Intel has been notifying customers of the setback for the last few days, telling them that sample volumes, once set to begin dribbling out in 4Q98, are now not expected until next year sometime, with production volumes slipping to mid-2000, roughly a year off course. The decision was made after an internal review of the program. A spokeswoman said the company had simply "underestimated" how difficult the project would be. Intel's official statement, which went up on its web site this afternoon after the story began to leak, is less direct. According to Stephen Smith, VP and general manager of Intel's Santa Clara Processor Division, "We have reached a point in the development cycle where the product is really starting to take shape. At this point, we have more precise schedule and product data, which allows us to better understand the scope of the product development, and we have communicated the new schedule to customers and the industry." Competitors such as Samsung, which is destined to become the merchant supplier of the DEC Alpha chip, had already been tipped to the news. A story running in today's ClieNT Server News quoted Samsung's key man in the states, YJ Kim, as saying the Merced had slipped at least six to nine months and wouldn't debut until mid to late 2000. Intel today resisted the idea of late 2000. Merced's delay could represent a golden opportunity to Alpha. Potentially it is Merced's most potent rival but it has languished in DEC's hands, failing to establish any following. That could change now that Compaq, which has always wanted to be more independent of Intel, is in the Alpha picture. It could decide to really push it. Merced's delay is also good news for Sun's Sparc RISC chip and Intel's x86 rivals Advanced Micro Devices and National Semiconductor and perhaps even for other reputed x86 wannabes like the Transmeta start-up which has yet to acknowledge that it even exists. Intel claimed the delay isn't attributable to any one thing in the highly complex chip design. The 64-bit Merced is based on a novel, untried instruction set architecture called EPIC, an Explicitly Parallel Instruction Computing technology that breaks with Intel's past and creates serious backward compatibility issues. It relies heavily on concepts of speculation, prediction and explicit parallelism to overcome the performance limitations of both CISC and RISC architectures. Such concepts, which drink from the same well as that notable market-failure Very Long Word Instruction (VLIW) technology, are not universally accepted in silicon circles. In addition, one of the burdens the little chip carries is that it must not only be backward compatible with the x86 but also with HP's PA-RISC line. Some time back, perhaps anticipating today's news, HP indicated that it could get through the year 2000 using its PA-RISC chips before it abandons them for IA-64. The basic design, to the extent that the secretive Intel has revealed it, is regarded as very difficult, even by Merced's architects. Intel researcher John Mills once remarked that "microprocessor design is like putting a gun to your head, pulling the trigger and then finding out three years later if you killed yourself." His words may come home to haunt him now. All industry development has been centered lately on the Merced, which has been responsible for most of the recent positionings and repositioning among operating systems vendors. It will take some time to assess the full impact of the slippage. Immediately though it would appear to work to the benefit of Windows NT, buying Microsoft more time to develop a true 64-bit operating system. Unix vendors had hoped to recover ground from Microsoft with 64-bit versions of their operating systems, which are far more advanced and robust than NT. NT's next release, 5.0, which isn't expected now until next year, won't even be fully 64-bit. Reports last week suggested that Intel's problem with Merced may be more a manufacturing one than anything else. In its own defense, Intel noted that the Merced project had:  Defined the 64-bit instruction set;  Completed the fundamental microarchitecture;  Completed the functional model and initial physical layout;  Completed the mechanical and thermal design and validation with system vendors;  Completed the specification and had designs for the chipset and other systems components underway;  Had made progress on the all-important 64-bit compilers and IA-64 software development kits;  Had in hand real-time software emulation capability for writing optimized applications;  Had multiple operating systems, presumably an early version of NT and the various Unixes, running on Merced simulations;  Had all required platform components planned for alignment with 1999 samples for initial system assembly and testing. At close of business today Intel shares were off two points to $71.43. Further slippage may occur as the news spreads but Intel was not expected to do a thumping great business in Merceds next year. Initial volumes were estimated to be only in the hundreds of thousands. It was not considered to be the stuff of volume shipments until its second generation in 2001 or beyond.