NexGen

» see nexgen.com as it lived, on the wayback machine

~ the goodbye ~

NexGen was founded in Milpitas in 1986 by Thampy Thomas, with funding from Compaq and Kleiner Perkins. The Japanese software house ASCII was in as well. The plan was to build an x86-compatible processor by reading the published instruction set and implementing it from scratch, with no license from Intel and no shared heritage with anything Intel had built. That is a legal position as much as an engineering one, and it took eight years to turn into a product.

The Nx586 shipped in 1994. Internally it was ahead of what Intel was selling: it broke x86 instructions into simpler internal operations and executed those out of order, an approach Intel itself would adopt with the Pentium Pro a year later. It ran at 75 and 83.3 MHz in its P80 and P90 grades, and the naming was a direct claim against Pentium parts at those numbers.

It also could not go in your computer. The Nx586 was not pin-compatible with anything, and required a custom NxVL-based motherboard that only NexGen and a handful of partners made. AMD and Cyrix were both selling chips that dropped into a Socket 5 board next to whatever else was already there. Alaris shipped the first Nx586 machines in the fall of 1994. Compaq, which had helped fund the company, never put the chip into volume production.

~ the chronicle ~

~ how it landed ~

The early Nx586 had no floating-point unit at all. There was an optional Nx587 coprocessor beside it, which in 1994 was already an answer from the previous decade, and later revisions brought the x87 unit onto a multichip module with IBM's packaging. The chip was therefore fast at integer work and poor at everything Quake was about to make important.

Meanwhile the Nx686 was in design, and it fixed the things that had killed its predecessor. It had a proper floating-point unit and it was meant for a standard socket. It never shipped as a NexGen product.

~ the decline ~

AMD's own K5 had gone badly. It was late, it underperformed against the Pentium parts it was priced against, and AMD needed a credible next chip more than it needed the pride of having designed one. NexGen needed a fab relationship and a socket that other people's boards already had. The two companies announced terms on 24 October 1995: 0.8 AMD shares for each of NexGen's roughly 41 million shares, valuing the deal at something in the region of 857 million dollars, all in stock.

It closed on 17 January 1996, structured as a tax-free reorganization with NexGen merged directly into AMD. About 29.3 million AMD shares went out, with another 4.3 million reserved against assumed NexGen options and warrants. AMD stopped work on its internal K5 successor and continued from the Nx686 instead.

~ what shipped anyway ~

The K6 came out in April 1997, Socket 7, floating-point unit included, and it was the chip that made AMD a serious competitor rather than a discount alternative. The design under it was NexGen's. Every K6 and K6-2 sold carried the work of a company that had ceased to exist fifteen months before the first one was boxed.

That matters here more than it did in America. AMD was the cheap option in a market where price settled nearly every purchase, and from 1997 onward the budget machines sold in Turkey ran K6 and K6-2 parts far more often than the US market did. A Pentium cost what it cost and an AMD machine cost visibly less, and in Turkey that gap was the entire argument. The instruction decoder doing the work inside those chips was drawn in Milpitas by a company already dissolved into AMD before I ever made a dial-up connection. Nobody gets told this. Silicon carries no note about who designed it.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »