T
Terje Mathisen
- Jan 1, 1970
- 0
Richard said:[Please do not mail me a copy of your followup]
John Larkin <[email protected]> spake the secret code
But there are few problems that need more than 64 or 80 bit
floating-point math, and fewer still that need it fast.
....which makes it perfect for reconfigurable computing.
Not really.
The critical building block, as Nick have already mentioned, is the
NxN->2N multiplier (and/or MAC/FMAC), which means that you don't need a
lot of uncommitted gates, but instead a lot of multipliers and enough
(reconfigurable) routing between them to setup the required operations.
As I understand it, that's what people are doing these days. Where
previously you would have been required to develop a custom ASIC at
great NRE cost, nowadays you can develop custom hardware using a
relatively inexpensive FPGA peripheral card. Configuring the FPGA
circuitry is very much like writing software when you use something
like VHDL or Verilog.
In fact, you can even get these in a portable form factor -- there's
an FPGA card you can buy that plugs into the Nintendo GameBoy Advance
and it comes with a complete tool chain included. All for about $200
from Charmed Labs <http://www.charmedlabs.com> as the Xport 2.0 which
has 150,000 gates.
Nice.
Terje