melling@cs.psu.edu (Michael D Mellinger) (04/03/91)
In article <14575@life.ai.mit.edu> petrilli@geech.gnu.ai.mit.edu (Chris Petrilli) writes:
I don't know that I agree with teh RISC approach in general, and I
think a multi-processor approach is much more sound, and in the end
better. A quad-'040 machine would dust the new HP (which, BTW, is
VERY impressive), especially if they attached 4 56001 chips to it. I
4 * 15 = 60. The new low-end HP's that cost $12,000 will do 57 mips.
I think NeXT would be much better off with a quad-RISC Cube.
think this is they way NeXT is going, and I think it is the right way.
If they decide to go RISC (God forbid), then I hope they stay away
from Sparc, it is not "infinitely extensible" as it's proponents
claim, and I see it as the least suitable choice. My preference is
for the MIPS set, perhaps the new R4000 due out next year I belive, or
perhaps the PA-RISC from HP (also quite good). The 88K is also a
pretty good processor, but I'm not sure its the best, but then who
knows what kind of deal Motorola would cut, if they could get 50MHz
88K processors out the door, they would be a force to be reckoned
with.
I agree, MIPS is the best choice. It is estimated that the R4000 will
perform 50 mips while only running at 25MHz. MIP's mips always seem
to be close to the actual SPECmark rating too. Motorola has made big
promises for the 88K, but who knows when they're going to deliver. In
the meantime, Intel is readying the 50MHz 486 for this summer, and a
486SX is due out soon. Moto. is getting thoroughly trounced.
However, Intel is going to release the N11, the sucessor to the i860,
in June, and it will have double the performance of the i860. I don't
suppose NeXT will ship NeXT Dimension in July with the N11 attached
instead of the i860?
-Mikesonger@orchestra.ecn.purdue.edu (Christopher M Songer) (04/04/91)
melling@cs.psu.edu (Michael D Mellinger) writes: > >4 * 15 = 60. The new low-end HP's that cost $12,000 will do 57 mips. >I think NeXT would be much better off with a quad-RISC Cube. > Well, um, actually... Not that I claim to be an expert or anything, but I don't think you can add the numbers like that. There is a certain inefficiency in multi-processor systems. Does anyone have an idea of what the speed ratings might actually be if 4 040's were used in a multi- processor fashion under NeXT Mach? Also, there is alot to be said for one big chip over a bunch of small ones. Really depends on what you are doing. Certainly multi- processing is not parallel processsing, (ie, a multi-processing machine and a single processor machine are going to do a striaght line calculation in the same time given both machines are unloaded.) I'd like to see a multi-processing Next, but there are some good points to moving to one really fast chip. -Chris