---------------------------------------------------------------------------- The Florida SunFlash The 3G Machine SunFLASH Vol 27 #15 March 1991 ---------------------------------------------------------------------------- This article was written for Workstation News by Andreas Bechtolsheim, vice president of technology at Sun Microsystems. He designed the original Sun workstation, the SPARCstation, and many other systems. It is reproduced with permisssion. -johnj -------------------------------------------------------------------------------- The 3G Machine By Andreas Bechtolsheim Sun Microsystems Inc. What is the future of workstation technology? In short: very very fast. Those of us in the industry are seeing system performance double every 12 to 18 months while costs are declining 33% or more at the same time. As a result, workstation cost/performance is doubling each year. There have been very few things in the history of mankind that have sustained this particular kind of exponential increase over time. To realize these expectations, RISC architectures are essential; the older CISC architectures are inadequate. Why? CISC is inherently defined by the personal computer -- a system that is very different from a workstation in important ways. From a personal computer perspective, the top three requirements for next-generation CISC chips are binary compatibility, high integration, and low power. It is obvious that by 1993, today's personal computer will be one chip plus memory that fits on a credit card, which plugs into a portable keyboard/display. Perhaps it has video and audio as well, but clearly the big push is for the smallest and lowest-power implementation that gives standard PC functionality. There are several highly integrated chip sets for PC portables available today, and the future will see a lot more of them. On the workstation front, in comparison, the top four requirements are performance, performance, performance and performance. This is due to the nature of the applications that workstation solve, which are basically large programs dealing with huge amounts of data in workgroups involving many users and heavy-duty networking. Thus the workstation market places fundamentally different product requirements on the central processing unit and system design than do traditional personal computers. This is why all major UNIX workstation vendors have made a transition to RISC technology. Since UNIX need not be burdened by binary compatibility with antiquated instruction sets, system designers created CPU architectures for UNIX that allow for the highest possible performance. Without going into the technical details, the real driving force for RISC was the ability to cut the ties with the past and design CPUs architectures that utilize our best theoretical understanding of how to maximize performance. In practice, this means that RISC offers two to three times the performance of CISC, assuming the same chip technology, die size, and so on. But even more important, RISC chips are being driven to the highest performance levels on all fronts, since that is what their application requires. A good example is multiprocessing. While there are now multiprocessor chip sets for personal computers, the simple fact is that MS-DOS does not support it. On the UNIX side, however, virtually every system vendor is working on multiprocessing systems that in the near future will provide total performance levels that are higher than any general-purpose computer system on the market today. One future concept we are investigating at Sun in this area is called the 3000M machine --- or the 3G machine, for short. The early Sun workstations in 1982 were really "3M machines", they had 1 MIPS of compute power, 1 Mbyte of memory, and a 1 Mpixel display. A 3G or 3000M machine would correspondingly provide 1000 MIPS of compute power, 1000 MBytes of memory, and one thousand times the graphics power of the original 3M machine. In other words, it's 1000 times as powerful on all fronts than the first Sun workstation, and more powerful than most supercomputers on the market today. But this new computer will not require some magical development in semiconductors or any radical new concepts beyond the logical extension of next-generation RISC CPU and system design. Superscalar techniques, which will soon replace vector processing, will result in effective clock speeds of well over 100 MHz, and future process technologies will further increase clock frequencies. Thus the 3G machine is simply the natural evolution of innovation and investments that we and others have made over the last many years. What could you do if you had your personal 3G machine? Truly incredible kinds of things. Imagine for a second that you are a genetic scientist who is deciphering the human DNA code. As it happens, the human DNA is about 2^30 genetic codes long, where each genetic code fits into 1 byte of memory. So you can store the entire DNA in 1 Gbyte of memory and use the processing power to compare gene sequences along the entire DNA code in a few seconds. Or imagine that you are a chip designer who's trying to simulate a one million-gate chip. That would be very difficult to do today on a current workstation. Not so with a 3G machine, which performs simulations about 100 times faster and can store the entire design. Or imagine that you are an aeronautical designer who wants to derive the optimal airplane wing shape, factoring in weight, strength, speed and fuel economy. Again, this kind of task is not possible on today's workstations, but quite feasible on a 3G machine. The common characteristics of these applications are that they deal with compute-bound, large, complex problems, which are inherent in all design and engineering activities. In fact, we see basically all of engineering move to simulation-based environments, because they will be more cost effective than traditional laboratory approaches. In summary, I see UNIX workstations as the driving force in high-performance computing. There is a literally unsatiable demand out there for faster and more powerful machines to design better, higher-quality, and lower-cost products in the shortest possible time. In the end, this is what will be fueling worldwide innovation for years to come. ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ For information send mail to info-sunflash@sunvice.East.Sun.COM. Subscription requests should be sent to sunflash-request@sunvice.East.Sun.COM. Archives are on solar.nova.edu and paris.cs.miami.edu. All prices, availability, and other statements relating to Sun or third party products are valid in the U.S. only. Please contact your local Sales Representative for details of pricing and product availability in your region. Descriptions of, or references to products or publications within SunFlash does not imply an endorsement of that product or publication by Sun Microsystems. John McLaughlin, SunFlash editor, flash@sunvice.East.Sun.COM. (305) 776-7770.