Talk:Weitek

FP precision?
A query over the comment that weitek floating point coprocessors were single precision only: certainly not true in the 1167 (for the 386), and presumably not true for later compatible successors. There was some weirdness in the instruction set (for instance subtract instruction was a=b-a where other two address machines have a=a-b). But it did both single and double precision. I'm pretty sure this was also true for the FPs weitek did for high end SUN 3s about the same time.

Oven controlled
The Silicon Graphics Inc. "Power Challenge" multiprocessor (IP21) used a device as its Floating Point Unit ("R8010") that had been developed for SGI by Weitek. I strongly believe that the IP21 FPU was capable of handling double precision numbers. Marty Deneroff was the project chief and I'm sure could provide an accurate primary source account. I have no idea how secondary sources appropriate for a Wikipedia article can be found for such things.

The unusual thing about this FPU was that it was what we would now call "oven controlled". It worked reliably in a narrow temperature range and so in place of the thermal pad that one would have expected to find between the device package and the heat sink, we installed a heater and thermostat. Rt3368 (talk) 22:32, 13 January 2016 (UTC)

Laser printers
I had an old Qume CrystalPrint "laser" printer that had two different Weitek chips on its mainboard. Instead of a laser, the CrystalPrint series had a long LCD shutter panel directly between a halogen lamp and the OPC drum, with a single LCD element for each dot in the line.