By Timothy Prickett Morgan
Sun Microsystems wants to remind everyone out there in server and workstation land that it is still making headway getting its UltraSparc-III processor out the door. Mel Friedman, president of Sun’s microelectronics division, told the Wall Street Journal that Sun was on track to get the so-called Cheetah processor in volume production by the end of the year. Friedman told the Journal that it got its hands on the first Cheetah samples from its fabrication partner, Texas Instruments, in early May.
The Cheetah, which has 23m transistors built using a 0.18 micron CMOS process and which will run at 600MHz, taped out in mid-April – about six months late. The lateness getting Cheetah off the drawing boards (or rather, out of the EDA supercomputers), has pushed out the delivery date for the 1,000-way Serengeti follow- ons to the current Starfire+ line to mid-2000 or so. Since mid- 1998, Sun has been telling customers that a 1,000-way Serengeti server will be able to process 2.4 trillion operations per second, 2500% more than the first generation Starfire servers.
Obviously, a lot of that power is going to come through huge (some might say ridiculous) scalability. As far as anyone can figure, the 600MHz Cheetah should have a SPECint95 rating of 35+ and a SPECfp95 rating of 60+, which is roughly double that of the current top-end 450MHz UltraSparc-IIs (19.6 SPECint95, 27.1 SPECfp95). (The current Starfire+ servers announced in March use 400MHz UltraSparc-IIs.) Sun’s original roadmap from September last year called for the delivery of a 480MHz UltraSparc-II part by now, and a 750MHz Cheetah part in early 2000. The former has been clocked down to 450MHz and still isn’t in Starfires, and the latter probably won’t get that out the door much before Sun had hoped to get the 1GHz UltraSparc-IV to market, which was due in mid-2000 more or less concurrently with the much-belated IA-64 Merced chip from Intel and Hewlett- Packard. There are clearly easier businesses to be in than the microprocessor racket.
