By Timothy Prickett Morgan
It’s a rare occurrence when IBM Corp kills a big, strategic product announcement that has been in the works for more than a year, but last Thursday Big Blue did just that. The people running the AS/400 business, which has been hammered by Y2K lockdowns and ERP slowdowns, decided that the only way they were going to be able to stimulate demand in the last quarter of 1999 and the first quarter of 2000 was to kill the widely anticipated I-Star PowerPC refresh of the current Northstar AS/400e product line.
IBM’s Jennifer Clark, who has the title of IBM Americas Technology Team Manager, broke the news to midrange analysts and business partners. There will be no refresh of the 7XX family in the first half of 2000, she said bluntly. When pressed as to why IBM would cancel the I-Star announcements, she replied with her own question: Why would we announce a new product when folks are just coming through Y2K? Plus, we just announced the 7XX line this year and it is going strong. We are not going to throw a new product in the street in three months and put customers in a frenzy. Customers can feel comfortable investing in 7XX servers today because there will not be a refresh announcement.
It’s hard to say whether or not this development (or perhaps insufficient development) will cause IBM’s AS/400 customers as well as RS/6000 customers who were counting on I-Star-based servers early next year to feel comfortable. While they and their IBM business partners both understand that the AS/400 business is on the rocks at the moment because of its dependence on ERP software sales to drive hardware sales, it can be argued just as logically that what IBM needs to do is move up the I-Star announcement or push Pulsar-based AS/400s out of the door to stimulate demand. IBM doesn’t plan to cut AS/400 prices to stimulate demand, and says that with the price cuts announced in February 1999 with the 7XX models it is already difficult enough for AS/400 sales reps and business partners to make their numbers. That is why IBM is sticking with the Northstars as is until the second half of 2000. The Northstar line started selling in late 1998 at the high end and in early 1999 at the low end of the AS/400 line; the high end 740 AS/400 models are essentially identical to the RS/6000 S7A Blackbird line, except that they can be equipped with more memory than the RS/6000s (thus boosting benchmark performance) and they run IBM’s proprietary OS/400 operating system rather than AIX.
The Apache AS/400 servers, launched in August 1997, were rebranded as the RS/6000 S70 Raven servers, and marked the first time IBM sold the same hardware platform under the RS/6000 and AS/400 names. This year’s just-announced RS/6000 S80 Condor server, which is smoking all the Unix vendor in benchmark tests, is more or less a prototype for what the high-end AS/400 I-Star models. While the S80 uses the new 450 MHz Pulsar processor (which is basically a Northstar processor re-implemented in IBM’s CMOS-7S copper chip fabrication process) and a whole new memory subsystem and high-bandwidth I/O subsystem to match it, the AS/400 I-Stars were going to use a faster 560 MHz processor in the same S80 frame. The I-Star chip, which was to be IBM’s first server processor to use its CMOS-8S chip process, was supposed to run at 560 MHz in February’s AS/400 servers and at 800 MHz in servers to be delivered at the end of 2000.
I-Star is virtually identical to the 450 MHz Pulsar, so it is probably not a chip design problem that is holding up the I- Stars. CMOS-8S marries IBM’s copper chip process with its similarly touted silicon-on-insulator (SOI) chip process, and it is SOI that is allowing IBM to crank clock speeds up from 450 MHz to 560 MHz. The Y2K problem may just be giving IBM a perfect cover to hide SOI problems, although IBM hasn’t admitted this yet. When asked directly if IBM could have announced the I-Star iron – meaning it was ready technically even if IBM didn’t think it was a good time to market it – Clarke said she didn’t know the exact status of I-Star development. In IBMspeak, this translates to I don’t want to talk about it right now. So don’t read too much into it.
The RS/6000 division has been mum about the I-Star so far, and has only implied that it would use the fastest chips that IBM could deliver in future RS/6000 machines, be they Pulsar, I-Star or Power4. There’s no question that if IBM wants to keep the heat on Sun, it has to use I-Star in RS/6000s when the chip is ready, not when the AS/400 division is ready. Ditto for the Power4 processors in 2001.
IBM has only officially committed to doubling the scalability of the next line of AS/400s sometime in 2000, incidentally, although that official statement is so recent that the people who made it to me a few months ago obviously knew that IBM was considering pushing out the I-Stars already. If IBM dropped OS/400 on today’s Condor S80 boxes, it would offer just about twice the performance of the AS/400 Northstar servers, which are still using slower 262 MHz processors rather than the 340 MHz Northstars that IBM is using in the RS/6000 H70 line. The Pulsar processor itself is about 60% faster than the old Northstar chip, and doubling the number of chips in the box from 12 to 24 would yield about twice the aggregate processing capacity of the top-end AS/400 740 models. The I-Stars running at 560 MHz would be about 2.5 times as fast as the 262 MHz Northstars, and the resulting 24-way servers would have about three times the aggregate processing capacity. The 800 MHz I-Stars would have about 3.5 times the uniprocessor power of the 262 MHz Northstars and the 24-way servers using them would have about five times the aggregate processing power of the current 740s.
This is exactly the kind of processing power that IBM needs to bring to bear in the Unix market, and it may be that this is one of IBM’s long-term motives. The vast majority of AS/400 customers have uniprocessor machines with processors that are not even as fast as the Northstar chip going full tilt. Only the biggest AS/400 shops need the I-Stars. But if IBM is to compete against Sun and HP, the RS/6000 division needs to be top dog for a while, as it is now with the Pulsar-based S80s. If IBM is having technical problems bringing faster midrange processors and servers to market, as Sun is having with its UltraSparc-III processors and the Serengeti servers that will use them, this does not bode well for the RS/6000. But the fact that the AS/400 division has pulled a product announcement for five months until demand picks up again is not necessarily a disaster for the RS/6000 division now, either. With the Pulsar-based S80 just out the door, IBM is obviously not going to be keen on talking about its future RS/6000 plans, but we’ll try to get the inside scoop for next week.
