Re: PCI card current loads

[png@woof.net (Peter N. Glaskowsky)]

> The worst case dynamic load for an add in card will most likely be found
> on a card with large amounts of memory, like a graphics or printer
> interface. Both of these types of cards use large amounts of RAM, some
> over 256 Mbytes. Most of the dynamic load comes from cycling the memory.

I'd have to agree that this can be very bad, but frankly this implies some
very bad board design.

I designed a NuBus product consisting of two full-size NuBus cards stacked
together for installation in a single slot. Between those two boards there
was a total of 288 4-Mbit DRAMs-- 144 MBytes. They covered about two-thirds
of each of the four available board sides.

With careful attention to detail, the whole board (including a 32-bit RISC
processor consuming a few watts itself and several ASICs) would never
exceed a measured 12W peak power consumption. There were four sets of
address and data buffers, staggered refresh, and an awful lot of
capacitors: low-ESR 0.1uF and 0.33uF ceramic caps (values empirically
selected) for each pair of DRAMs, plus one 10uF tantalum cap for each nine
DRAMs, plus about thirty more 10-50uF tantalums spread around each board.
(This is from memory; I did this four years ago.)

I've never seen such a dense packaging of memory devices for a
personal-computer application, but it worked fine. This board set was
manufactured and sold in some moderate volume as the SuperMac Splash (later
the Radius Splash, and possibly still available from Splash Technology).

I think it's just plain irresponsible to build a PCI card that could
consume the kind of power you describe. 100W?? Ridiculous. That's enough to
risk damage to the motherboard or PCI connector, and I know it's not
necessary.

.                 png

ÅD1