Weitek

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

~ the last words ~

Weitek was started in 1981 by several engineers who had left Intel, and for the first decade its business was a single clean idea: general-purpose processors were bad at floating-point arithmetic, so sell a separate chip that was good at it. Intel and Motorola both sold their own coprocessors and Weitek's were faster, so Weitek got the customers who measured.

The parts list reads like a map of 1980s computing. The WTL 1064 and 1164 went into Motorola 68000 systems. The 1167 arrived for the 386 as a daughtercard, then the 2167 put it on one chip, then the 3167, sold as Abacus, in 1988. The 4167 followed in 1989 for the 486 at around 4 MFLOPS. There were 3170 and 3172 parts for SPARC machines. If you had bought a workstation in that decade there was a fair chance Weitek silicon was doing the hard sums in it, under someone else's name.

The problem with selling the part that the main chip is bad at is that the main chip gets better. Intel put the floating-point unit on the same die as the processor in the 486DX in 1989, and from that moment the question was not whether Weitek's chip was faster. It was whether anyone would pay separately for a function that now arrived free in the box. Weitek's own best customer, the 486 market, was where the integration landed first.

The company had been making substantial operating losses for several years by the time it admitted the position was finished. On 16 December 1996 it filed for Chapter 11 protection and announced it was selling assets and a technology license to Rockwell International for three million dollars, subject to adjustment depending on what could actually be transferred. Rockwell got the Universal Memory Architecture work and the floating-point technology, and made job offers to about twenty engineers. The remaining intellectual property was still looking for a buyer.

~ the record ~

~ the hole in the motherboard ~

Open any 386 machine and you saw the second socket, square and empty, sitting beside the processor. That socket was Weitek's entire business and on most boards it stayed empty for the life of the machine, because the chip that went in it cost more than most buyers had spent on everything else combined. In Turkey it was never a live choice. The question here when buying a machine was SX or DX, which is the same question one decimal place down, and the answer was usually SX. By the time I had any say about what was inside a computer the socket had been designed off the board entirely.

~ the second act ~

Weitek did try to move. In the early 1990s it went into SVGA graphics under the POWER name, shipping the P9000 and later the P9100, and for a couple of years those were respectable high-end Windows accelerator chips that ended up on cards from several board makers. It was a reasonable read of where the money was going. The trouble was that graphics in the early 1990s was the most crowded market in the industry, with S3 and Tseng and a dozen others already fighting on price.

Rockwell folded what it bought into its Brooktree division and the Weitek name disappeared from products almost immediately. The firm had been public and had shipped genuinely superior parts for a decade. It was worth three million dollars at the end. Nothing went wrong with the engineering. The category it had been built on was absorbed into a competitor's die, which is a way of losing that leaves no one to blame and nothing to fix.

@}-->--

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

« previous  ·  random  ·  next »