In the aftermath of Sun Microsystems Inc’s SuperSparc-2 RISC announcement, the Microprocessor Report has performed what it believes is a definitive post-mortem on the troubled SuperSparc chip family designed by Sun and fabricated by Texas Instruments Inc. The problems have caused Sun and its followers much concern and proved a stern warning to the rest of the industry, which has subsequently been far more circumspect when estimating the performance of future processors. SuperSparc was announced in February 1991 and scheduled to appear at 50MHz, but made first silicon at 33MHz due to internal timing problems. The design eventually got to 40MHz in 1993 and finally went to 50MHz with the SuperSparc+, in the third quarter of 1993 after a 10% process shrink, a year later than promised. Microprocessor Report believes that the design team sliced away clock speed to reduce complexity. However, the team did not build a complete timing model of the chip, as Sun’s tools were not up to it at the time. This meant that it had no idea of how damaging its actions were until critical paths that had not been apparent showed up in the first silicon it got back from Texas Instruments, which showed the design was in bad shape. Sun and Texas worked to rectify the problem, but each fix and each new iteration of the chip revealed other unseen critical paths. The convoluted pipeline and reduced yield from using BiCMOS transistors hurt the overall design, the newsletter believes. Later performance gains came from the gate shrinks. If the design team had recognised the problems it was getting itself into, the SuperSparc would have looked more like SuperSparc-2, said the report. Even now Sun does not expect significant yield at 75MHz and 90MHz from SuperSparc-2.
Power-starved
Although Microprocessor Report observed that SuperSparc-2 delivers more performance to power-starved Sparc users by unclogging the pipeline, it does not see the part improving Sun’s position compared with other RISC chip sets on performance or price. SuperSparc-2 does not make up for this lack of performance by offering lower cost: indeed it is just the opposite. According to its estimates, SuperSparc-2 is the most costly processor of its generation, 15% more than Hewlett-Packard Co’s PA-7200 and at least 60% more than the 120MHz Pentium, 200MHz R4400, 100MHz PowerPC 604 and 275MHz 21604A. The 75MHz SuperSparc-2 costs twice as much as a 90MHz PowerPC 604 but delivers less performance. The only way for Sun to cure its price-performance problems is for the SuperSparc design to be expunged from its product line, warns Microprocessor Report, unfortunately UltraSparc-1 will simply push SuperSparc-2 into the mid-range rather than displace it entirely, leaving Sun with a part that is too expensive and underpowered to compete in that market. Despite the problems, Sun has maintained its leading share of the workstation market by pricing its systems aggressively and essentially giving up the very high end of the market. SuperSparc-2’s shortfalls could dent Sun’s mid-market strength, the newsletter believes, but sees help on the way in the form of microSparc III which should finally rid Sun of the curse of SuperSparc. Most likely that is also why Sun is still shy about describing the microSparc III design in any detail.
