For some 30 years prior to the introduction of CMOS technology in the early ’90s, IBM mainframe processors had been increasing in capacity by 15 percent per annum, with the price declining at a similar rate. The correlation between the capacity increase and price decline was due to the economics of processor manufacturing, where any capacity increase delivers an identical cost reduction. Now, although this was a significant annual price decline, it didn’t allow users to “scrap” mainframe technology as early as was the case in the mini-computer and PC world. This resulted in users requiring upgrades to installed mainframe processors and even additional processors of the same type for some years after they were removed from IBM marketing. This created a healthy used market for IBM processors.
However, with the introduction of CMOS with its 40 percent per annum capacity increase and a 40 percent per annum price decline, the mainframe processor life was shortened and the value and demand for “used” systems was minimal. This was also driven by IBM’s clever marketing programs, which allowed for upgrades from one technology level to the next, restricting the availability of used processors and some manipulative maintenance pricing, that greatly increased the TCO of older processors.
Around 2000 this changed, and since then, the average annual price decline is closer to 10 percent—well below even the pre-CMOS level—and consequently, the second user market is once again growing as users are keeping their processors longer and, therefore, require upgrades and additional processors. This price decline slowdown is undoubtedly a result of the Plug Compatible Mainframe (PCM) suppliers exiting the market—leaving IBM with no competitive pressure to lower prices. In fact, IBM increased the price per MIPS slightly with the introduction of the z990—the first time in the history of the mainframe.
The result of the slower price decline is that the price per MIPS is around 10 times higher today than the level that should have been achieved with CMOS technology, yet the performance of the top-end z990 is almost exactly on the technology curve.
To minimize the number of used systems (or upgrades) being purchased, IBM is today selling the older z900 at up to 25 percent higher prices per MIPS than the z990, is charging higher maintenance prices per MIPS, and is offering lower software pricing for most users on the z990.
The net result is that for any user requiring more than a very small z900 upgrade, the three-year TCO of a z990 solution from IBM will be around 10 to 15 percent less than the z900 from IBM.
While the slower price decline may sound like bad news, market forces always come into play when a monopoly attempts to exploit its domination; in this case, second user systems and third-party maintenance. These are both well-tried solutions from an earlier era and will grow in importance over the next few years.
To illustrate the used processor opportunity, users looking for either a net new z900 processor (which will be a used system) or an upgrade to an existing z900 can today purchase such systems from third-party suppliers for around 600 euros or $750 per clothed MIPS. This is less than half the current IBM price. For new systems, this involves a supplier delivering a complete z900 system, but for upgrades, it normally involves replacing the installed system with a larger model—a two-truck upgrade as it is known in the trade.
To illustrate the potential savings, the following figures are from a recent situation where a user considered purchasing two new systems and wanted to evaluate the three-year costs. With the z990, the solution was based on models 302 and 304, with the 304 growing to a 306 over three years. For the used z900 alternative, z900 models 1C4 and 1C8 were chosen, with the 1C8 growing to a 113 over three years.
Despite maintenance and software savings with the z990, the used z900 solution from a third-party was still $1 million (5 percent) cheaper over the three years. But more important, even greater savings were seen in the first year, which is highly attractive to many users. With the IBM used z900 proposal, the three-year z900 cost was $3 million (14 percent) above the z990 cost, and even the one-year z900 cost was $1.8 million (20 percent) higher.
Any user requiring an upgrade rather than a new system is likely to save even more with a third-party used z900 solution due to the “book” pricing issue of the z990. For example, a user upgrading from around 1,250 to 2,000 MIPS would save more than $380,000 (28 percent) in the first year by opting for a second user z900 upgrade—although for some users, some of this savings could be offset by the lower software costs on the z990.
In this case, even a z900 solution direct from IBM would still have been less expensive than the z990 upgrade, but the advantage of the z900 would have been reduced to less than 10 percent.
While this has focused on the z900 used processors, a similar case could be made for a used z800. However, earlier CMOS processors offer little opportunity for all but the smallest users with no planned growth.