3

The earliest CMOS microprocessors (RCA 1802, HP Stirling RISC, et.al.) were slower than contemporaneous NMOS microprocessors and Bipolar logic computers. (IIRC, both the 1802 and the 6502 could be clocked at 1 MHz, but the 1802 required 8 cycles for the shortest instruction, much more than the 6502.)

When did commercially available CMOS microprocessors become faster than NMOS microprocessors? And when were CMOS IC based processors first used for computers fast enough to be on, or to top the Top-500 supercomputers list, outperforming computers based on bipolar logic (various CDC and Cray models, et.al.)?

8
  • 2
    Never? Not the chip technology but computer design changed. Bipolar faded as it became less expensive to build a supercomputer from a larger number of slow but cheap and ready available processors than to continue developing the faster ones which would only be needed in low numbers. It wasn't the higher performance per core that made CMOS the survivor, but a lower Dollar per MIPS ratio they did provide (by simply using more of them)
    – Raffzahn
    Mar 16 at 23:42
  • 7
    I think it is fair to say that an NMOS version of any modern (last 20+ years) microprocessor would melt down almost instantly.
    – Jon Custer
    Mar 17 at 0:54
  • 2
    @JonCuster: Emitter-coupled logic based on bipolar transistors was historically faster than CMOS, and required insane amounts of cooling. Static currents are orders of magnitude greater than CMOS, but since voltage swings were much smaller, I think dynamic currents would have been reduced.
    – supercat
    Mar 17 at 14:55
  • 1
    In the beginning, you didn't choose CMOS for its speed - you chose it for its ultra low power draw, in things like digital watches. Speed didn't become a priority until people started realizing that power draw was the limiting factor for speed in the older technologies. Mar 18 at 3:55
  • 1
    @MarkRansom: Power dissipation didn't become a limiting factor in older technologies until feature sizes started falling. The passive pull-ups in something like a 6502 are smaller than suitable PMOS transistors would have been.
    – supercat
    Mar 18 at 15:19

2 Answers 2

5

In DEC, the VAX 9000, due to ship in around 1989 but actually delayed somewhat, was a hugely-expensive ECL-based computer.

By 1988 it had become clear to some that the NVAX processor, a CMOS implementation despite the codename, would be competitive with the 9000's CPU.

In 'VUPs' (VAX unit of performance, with the original 11/780 as 1 VUP), the 9000 was rated at 40 VUPs. NVAX+ was 35 VUPs, not quite matching the 9000, but on the other hand it sat on your desk rather than requiring an air-conditioned room. A later variant reached 50 VUPs.

All of this puts the 'crossover time' at around 1990.

(There was of course more to the VAX 9000 than raw CPU power, but not enough to keep it alive in the face of competition from desktop systems)

2
  • The question asked about fastest microprocessors, and by 1990, there were plenty of NMOS RISC chips faster than any VAX. Also, the VAX 9000 CPU was not a microprocessor, but a large circuit board with multiple packages, each containing multiple gate arrays. The first two VAX microprocessors (V-11 and MicroVAX 78032) were NMOS, and didn't go much above 1 VUP, but all the successors were CMOS. Mar 19 at 13:58
  • 2
    The question mentions 'bipolar logic computers', of which the VAX 9000 was one. It also asks when CMOS processors were fast enough to be on the 'supercomputer list' - I'm not claiming the VAX was a supercomputer, but certainly none of the mentioned supercomputers were built with microprocessors. So, my answer addresses the subpart about CMOS microprocessors outperforming bipolar logic. Mar 19 at 15:35
2

It's hard to put an exact date on it, but the release of the Intel Pentium Pro in 1995 is a good approximation. This was fabricated in Bipolar CMOS (BICMOS) which is now mostly used for mixed analogue-digital chips, but was used for the Pentium, Pentium Pro and Sun SuperSPARC.

The Pentium Pro didn't make a huge impact in the consumer market, because the dominant OS in that market at the time was Windows 95, which still included a lot of 16-bit code. The Pentium Pro was designed primarily for 32-bit code, and its 16-bit performance was lower. This problem was fixed in the Pentium II and Pentium III families.

However, some builders of 32-bit RISC processors. such as MIPS and Sun, found themselves needing to improve performance drastically to compete with Pentium Pro. They did this, for a while, but were eventually defeated by the amount of money that Intel made from the consumer market and invested in chip development.

The Pentium Pro was used in ASCI Red, which was the fastest computer in the world from its completion in September 1997 until late 2000.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.