Nagle@cup.portal.com (John - Nagle) (11/12/89)
The i860 and i960 families are rather nice general-purpose RISC
machines. It's interesting that Intel insists on referring to them as
"embedded processors". Presumably this is to avoid annoying the
marketing people who are putting up all the "386" billboards. In
many ways, the 860 is a better machine than the 386. It has lots of
registers, almost too many. There are 32 integer registers (32 bits
each), 32 floating point registers of 32 bits each, and ten other
special-purpose registers. At last, Intel has a part with lots of
homogeneous registers. There's on-chip floating point, an on-chip
paged MMU, user/supervisor mode, and a 32-bit flat address space.
Special instructions for Z-buffered graphics are provided.
Then there's the i960. This isn't a faster 860. It's a completely
different architecture. Again, a flat 32-bit address space, and
user/supervisor mode, but no on-chip MMU except in a special model
"designed for Ada multitasking applications." The register model
is different from the 860; there are "local" and "global" registers,
and upon procedure call, the local registers are saved, on the stack
if necessary and on-chip if possible. The instructions are different
from those in the 860, and not just trivially so.
There's no visible commonality between the two families. It looks
like Intel had two projects underway to build a fast RISC machine, and
they both succeeded. Would someone from Intel care to comment?
Both are very fast machines. Numbers like 66 to 99 MIPS appear in
the documentation. I look forward to seeing numbers from third parties.
You could certainly build a workstation around either part, and it would
be much faster than a '486-based workstation.
These machines are the future. The '86 line is the past.
John Nagle