Bill Broadley writes:
>I've talked to a few dec engineers and they seem to think that a very
>cheap motherboard using the ALPHA 21066 should be possible. They sent
>me I believe almost everything I need to make a motherboard. (Not
>that I have any skill at such things.)
>Big features of the 21066:
> Embedded graphics controller (on chip)
> L2 cache support (directly connects to sram.)
> PCI I/O controller on chip
> 8k/8k L1 cache
> 3.3 V chip (cooler)
> 70 Specint 105 Specfp (better then any 100 Mhz mips)
>This has several big advantages of the mips design:
> Cheaper, faster, PCI.
>Of course the disadvantages:
> No cheap production at Waldorf
> Software will be pretty tricky getting started/booted.
I think this is worth looking into further. I think we should check
the prices of available motherboards, though, while we consider one of
our own, as it may be just as cheap overall to get an existing board.
DEC intends for this chip to be in strong competition with the
pentium, so prices for the whole motherboard could be pretty low, even
for a pre-assembled one. I would hope that we could get 21066
motherboards for < $1000 without ram, but with CPU, at some point. If
not, DEC is going to have trouble competing with intel, unfortunately.
Then we just have to port Linux to it... I wonder if DEC would be
interested in helping us with that? There was someone from DEC who
posted recently saying that if there was enough demand, he'd talk to
his management about a Linux port. Maybe if there isn't enough
demand for that, but some demand, they'd help others do a port.
By writing PALcode that approximates the MM of the 386, a porting
effort could be made a lot easier. We would also be able to use
Drew's PCI scsi driver, with few or no changes, which would be a big
win. A RISC chip with 386-like MM and all the major devices already
supported in Linux/86 would make for relatively easy porting, other
than places where going from 32 to 64 bits causes problems, which
shouldn't be too many places. Hopefully, GCC will have improved code
generation for the ALPHA by that time...
I vote that we look into this, at least.