From the very beginning of the computer era in the early ’60s, there was one major rule—the price per unit of performance of mainframe processors would decline directly in proportion to the performance increase. So, if the performance from one processor generation to the next doubled, then the price per MIPS (or KIPS in the early days) would halve. This meant that from one generation to the next, despite a doubling of the capacity, the largest system consistently cost around $3 million— not unlike the PC scenario where despite huge performance increases each new model was roughly the same price.
Behind this was the basic manufacturing process where performance increases were largely driven by the miniaturization of components, so that to double the performance the component size had to be halved. Consequently, a production line that produced x units of performance after halving the size of the components could now produce 2x units of performance for the same cost. Therefore, assuming the full capacity could be sold, the price to the user could be halved at no loss of profit to the supplier.
Obviously, the supplier could increase its profitability by reducing the price by somewhat less, but until recently, the existence of competition with the same economic scenario ensured all suppliers reduced prices in line with the technology improvements.
This model resulted in the price per MIPS falling from $10 million in 1960, to $400,000 in 1980, as technology advanced at a rate that allowed the price to fall by 15 percent per annum. This rate of decline accelerated to 26 percent per annum during the ’80s with the price per MIPS falling from $400,000 to $20,000 as competition increased with the arrival of PCM suppliers Amdahl and Hitachi. This competition resulted in technology advancing more rapidly to sustain the higher level of price decline. This was good for the users, as only more rapid technology advancement could enable IBM to sustain its market share. As a result, both technology and price declines were accelerated, allowing an explosion in the use of technology.
However, as software costs became more important in the ’90s, IBM began to exploit its control of the software to limit the competitiveness of the PCM suppliers (for example, through the Parallel Sysplex pricing model). This enabled them to slow the speed of technology advance and therefore the consequent price decline. The result was that during the ’90s, prices fell by only 20 percent per annum from $20,000 to $2,000 per MIPS.
This was obviously not so good for the users, but worse was to come as IBM’s use of software pricing to reduce the competitiveness of the PCMs eventually forced them out of the market in the late ’90s. This has allowed IBM since 2000, now with no mainframe processor competition, to completely separate the technology advancement and price decline curves for the first time in the history of the mainframe.
To illustrate just how dramatic the change has been, over the 2000 to 2005 period, the technology improvement has been more than 300 percent, which would support a price decline of more than 20 percent per annum. Yet, IBM has delivered closer to 13 percent per annum price decline to lower the price from $2,000 to $1,000. This represented a significant slowdown for the user. But it became even worse when the z990 was introduced at the same price as the z900 predecessor, despite a doubling of the technology that would have supported a 50 percent price reduction. This has been followed with the z9 being introduced at around 10 percent higher pricing than the z990 despite a 30 percent technology improvement that should support a 15 percent price reduction.
The net result of the slowdown that has occurred since the ’90s, when IBM first began to limit and ultimately eradicated all mainframe processor competition, is that the price per MIPS today is approximately six times higher than the $165 per MIPS that the traditional technology/price decline link would have produced. As a result, the largest systems today cost closer to $18 million than the $3 million they have cost throughout most of the mainframe’s history—a very high price to the user for the elimination of competition.
Interestingly, if you go back to the PC comparison at the beginning of this column, the opposite has occurred with PCs where increased competition has actually meant that in addition to large performance increases from one model to the next, the price of the typical PC is now declining.