While this chip adresses many of our needs (ie, address/data DEMUX,
DRAM / VRAM control, video timing signal generation, AT keyboard
interface) but falls short in other areas.
Most notable is the chips abysmal DMA support :
- There's no support for slave mode DMA. This makes it impossible to
use the low cost NCR53c90 for synchronous SCSI, severely
limiting your bandwidth. It also makes it impossible to
get slave mode DMA from IDE, etc. devices, meaning we'll
have to waste CPU cycles doing polled transfers when we
recieve an appropraite interrupt.
- Support for bus mastering *is* there, Bus grant / requests could
be daisy chained to get something other than the AMC ILAC
on the bus. However, if we use other busmastering devices
(the NCR53c700 SCSI controller comes to mind immediately) we
increase the cost of the rest of the system, not to mention
hardware and software development time. Also, the onboard
support for busmasters wants to see a MUX'd address/data bus
with signals like those from the ILAC / 3041/3051/3081.
Also, bus sizing is limited to 32 / 64 bits wide, and we want
to use 8 and 16 bit peripherials.
Obviously, we can design arround these limitations with external
hardware, but the cost of the necessary internal hardware must be
considered when comparing the MOM chips with others like the
3700.
|