FlashPaper: Merced will be late – should HP 3000 shops care?

June 1998

Click here for Adager sponsor message

Merced will be late – should HP 3000 shops care?

Sounding a lot like Hewlett-Packard did a dozen years ago with the Spectrum RISC project, Intel admitted at the end of May that its Merced 64-bit chip will ship six months late, now sometime in mid-2000. Merced, the most tangible evidence of the IA-64 RISC architecture designed and manufactured by HP and Intel, is having problems in the manufacturing and performance departments, according to insiders on the project. The delay of the project first announced in 1993 makes Merced one of the longest-awaited technologies in HP history, but its impact will be slight on the HP 3000 community. That’s because as of presstime, Merced and IA-64 still aren’t officially on CSY’s adopted technology list for the HP 3000s of the future. However, the division has begun to show off performance charts for its product line like the one flashed at IPROF this spring, charts with an interesting spike in performance for systems around the year 2002. And it’s no secret at all that MPE/iX is being redesigned in a major project by the division for 64-bit compatibility – with PA-RISC processors in the 8200 and 8500 series. HP is saying that its 64-bit work for the 3000 will arrive with the same minimal impact that every other generation of PA-RISC delivers: no scrapping old code, no house-divided between the older 32/48 bit 3000s and the newer PA-RISC of the 8000 series chips. CSY’s work is proceeding for HP’s own RISC processors, while Intel efforts will stay in the labs a half-year longer.

The May 29 announcement reminded us of HP’s 1986 admission that Spectrum, the first commercial RISC system from HP, wouldn’t be in any 3000 site’s shop until sometime the next year. These major architecture shifts usually take longer than expected, and Merced has clearly outlasted most predictions of its arrival. Nobody ever promised working Merced silicon before 1998, but not long ago the predictions slipped for sometime in 1999. Now the chip will arrive in the new millennium, if Intel and HP can figure out how to make the processor run the X86 software that will be its bread and butter a lot faster. Rumors are circulating, no doubt fueled by Merced competitors like Advanced Micro Devices and IBM/Motorola, that the Windows 95/98 software is mighty slow on Merced. The part of the processor that runs things like HP-UX software, and other truly-RISC environments that have been announced for it like Sun’s Solaris, is reported to be a screamer. It’s hard to tell, because there simply haven’t been any released speed figures on Merced in all of the five years we’ve been waiting for it.

No, the real problem with the chip is reportedly in the X86 emulation, a part that Intel is even more keenly interested in than its partner HP. The Pentiums have been outpaced by other X86 chips lately, and Intel has been forced to keep cutting prices on them to keep their market share. That’s really tough on earnings, something Intel disappointed investors with a few weeks back. Intel really needs a chip that will run all that Windows software, NT and 98 alike, a lot faster than Pentium. This is of some interest to HP, of course, but Hewlett-Packard’s bottom line won’t shift dramatically if Pentiums keep getting beat and while Merced learns to race Windows applications. In fact, the delay may even help HP, since Intel will have to continue to reduce prices on components that HP sells in its PCs and NetServers. HP has been counting on getting hot chips ready to roll out before 2000 for its Unix systems, especially the high end servers that are still fighting off Sun offerings. With no Merced for an extra six months, HP-UX will have to defend itself with the PA-8500 and PA-8700 chips. This was far from the HP 9000 division’s first game plan for sales in early 2000. It’s going to be longer before Merced starts to earn the speed marks that will prompt software companies to do the porting to it.

Some surprising software suppliers are announcing that Merced and IA-64 ports are underway. The most surprising to us was IBM, one of the only companies that still has an active hand in an alternative processor for high-end and midrange systems. IBM is still marketing its PowerPC chips to other suppliers, mostly in the embedded systems market and to Apple for Macintoshes. New Activec technology in the PowerPC promises a vast performance jump for multimedia and networking applications next year. But the place PowerPC is most likely to bump up against HP 3000 interests is when it’s installed in an AS/400. Like the 3000, IBM’s venerable war horse got a new lease on life with RISC inside, and it’s having a sales resurgence in double digits this year. Despite having all that technology in house and cheap, IBM still admitted in late May that it’s building a version of the AIX operating system for Merced. When you’re IBM, you can apparently afford to hedge your bets even if the wager is expensive and causes some confusion among your software partners. If you write software for the RS/6000, what do you develop for – 1999’s 64-bit PowerPC model at 600MHz on copper traces, or Merced? And forget about Alpha making a comeback. Intel recently signed to buy the chip foundry where all the Alpha processors are built, and run it for Digital. Wonder how many Alphas Intel will turn out with Merced running late? The deal is really about getting another stream of software for Merced; Intel, Digital and Microsoft will work on creating a common programming environment for 64-bit NT that covers both Merced and Alpha chips. Digital retains the rights, but essentially steps out of the chip race as it drops its lawsuit with Intel that it launched last summer. See, money can buy friends, or at least allies who need some cash and can sell off assets that aren’t a major part of their future.

Happily, the 3000 community will be suffering little from Merced’s delay. Just remember, we haven’t heard yet that CSY is passing up Merced; it was one of the candidate technologies for 3000s listed at last year’s HP World briefings on futures. For the moment, CSY is taking a “we’ll see” look at adopting Merced and IA-64. We can only wonder what they might see in the news of the delay, but we know that PA-RISC remains the division’s first-choice game plan for the next three years at least. That should cover any more delays, problems that could well be making HP’s Unix and NT fans sweat a little with this hot summer news.
Copyright 1998 The 3000 NewsWire. All rights reserved



		Click here for Adager sponsor message
		Merced will be late – should HP 3000 shops care? Sounding a lot like Hewlett-Packard did a dozen years ago with the Spectrum RISC project, Intel admitted at the end of May that its Merced 64-bit chip will ship six months late, now sometime in mid-2000. Merced, the most tangible evidence of the IA-64 RISC architecture designed and manufactured by HP and Intel, is having problems in the manufacturing and performance departments, according to insiders on the project. The delay of the project first announced in 1993 makes Merced one of the longest-awaited technologies in HP history, but its impact will be slight on the HP 3000 community. That’s because as of presstime, Merced and IA-64 still aren’t officially on CSY’s adopted technology list for the HP 3000s of the future. However, the division has begun to show off performance charts for its product line like the one flashed at IPROF this spring, charts with an interesting spike in performance for systems around the year 2002. And it’s no secret at all that MPE/iX is being redesigned in a major project by the division for 64-bit compatibility – with PA-RISC processors in the 8200 and 8500 series. HP is saying that its 64-bit work for the 3000 will arrive with the same minimal impact that every other generation of PA-RISC delivers: no scrapping old code, no house-divided between the older 32/48 bit 3000s and the newer PA-RISC of the 8000 series chips. CSY’s work is proceeding for HP’s own RISC processors, while Intel efforts will stay in the labs a half-year longer. The May 29 announcement reminded us of HP’s 1986 admission that Spectrum, the first commercial RISC system from HP, wouldn’t be in any 3000 site’s shop until sometime the next year. These major architecture shifts usually take longer than expected, and Merced has clearly outlasted most predictions of its arrival. Nobody ever promised working Merced silicon before 1998, but not long ago the predictions slipped for sometime in 1999. Now the chip will arrive in the new millennium, if Intel and HP can figure out how to make the processor run the X86 software that will be its bread and butter a lot faster. Rumors are circulating, no doubt fueled by Merced competitors like Advanced Micro Devices and IBM/Motorola, that the Windows 95/98 software is mighty slow on Merced. The part of the processor that runs things like HP-UX software, and other truly-RISC environments that have been announced for it like Sun’s Solaris, is reported to be a screamer. It’s hard to tell, because there simply haven’t been any released speed figures on Merced in all of the five years we’ve been waiting for it. No, the real problem with the chip is reportedly in the X86 emulation, a part that Intel is even more keenly interested in than its partner HP. The Pentiums have been outpaced by other X86 chips lately, and Intel has been forced to keep cutting prices on them to keep their market share. That’s really tough on earnings, something Intel disappointed investors with a few weeks back. Intel really needs a chip that will run all that Windows software, NT and 98 alike, a lot faster than Pentium. This is of some interest to HP, of course, but Hewlett-Packard’s bottom line won’t shift dramatically if Pentiums keep getting beat and while Merced learns to race Windows applications. In fact, the delay may even help HP, since Intel will have to continue to reduce prices on components that HP sells in its PCs and NetServers. HP has been counting on getting hot chips ready to roll out before 2000 for its Unix systems, especially the high end servers that are still fighting off Sun offerings. With no Merced for an extra six months, HP-UX will have to defend itself with the PA-8500 and PA-8700 chips. This was far from the HP 9000 division’s first game plan for sales in early 2000. It’s going to be longer before Merced starts to earn the speed marks that will prompt software companies to do the porting to it. Some surprising software suppliers are announcing that Merced and IA-64 ports are underway. The most surprising to us was IBM, one of the only companies that still has an active hand in an alternative processor for high-end and midrange systems. IBM is still marketing its PowerPC chips to other suppliers, mostly in the embedded systems market and to Apple for Macintoshes. New Activec technology in the PowerPC promises a vast performance jump for multimedia and networking applications next year. But the place PowerPC is most likely to bump up against HP 3000 interests is when it’s installed in an AS/400. Like the 3000, IBM’s venerable war horse got a new lease on life with RISC inside, and it’s having a sales resurgence in double digits this year. Despite having all that technology in house and cheap, IBM still admitted in late May that it’s building a version of the AIX operating system for Merced. When you’re IBM, you can apparently afford to hedge your bets even if the wager is expensive and causes some confusion among your software partners. If you write software for the RS/6000, what do you develop for – 1999’s 64-bit PowerPC model at 600MHz on copper traces, or Merced? And forget about Alpha making a comeback. Intel recently signed to buy the chip foundry where all the Alpha processors are built, and run it for Digital. Wonder how many Alphas Intel will turn out with Merced running late? The deal is really about getting another stream of software for Merced; Intel, Digital and Microsoft will work on creating a common programming environment for 64-bit NT that covers both Merced and Alpha chips. Digital retains the rights, but essentially steps out of the chip race as it drops its lawsuit with Intel that it launched last summer. See, money can buy friends, or at least allies who need some cash and can sell off assets that aren’t a major part of their future. Happily, the 3000 community will be suffering little from Merced’s delay. Just remember, we haven’t heard yet that CSY is passing up Merced; it was one of the candidate technologies for 3000s listed at last year’s HP World briefings on futures. For the moment, CSY is taking a “we’ll see” look at adopting Merced and IA-64. We can only wonder what they might see in the news of the delay, but we know that PA-RISC remains the division’s first-choice game plan for the next three years at least. That should cover any more delays, problems that could well be making HP’s Unix and NT fans sweat a little with this hot summer news. Copyright 1998 The 3000 NewsWire. All rights reserved