ramsdell@mitre.org (John D. Ramsdell) (07/10/90)
In article <112780@linus.mitre.org> ramsdell@mitre.org writes: >It seems that NeXT is committed to the use of antiquated CPU's. I think my use of the word antiquated was inappropriate and overstated my point. I really had two points in mind. [1] It makes little sense to replace the CPU of a NeXT Computer when instead you can simply add a CPU. Is that not one of the advantages of using Mach as your kernel? Why does the mother board have to have the major processing power? [2] Let the market place decide if the new processor should be RISC or CISC. If vendors think they can make a better profit by selling CISC coprocessor rather than RISC coprocessors, so be it. Let the competition begin! John
mdeale@vega.acs.calpoly.edu (Myron (the one in Calif.) Deale) (07/11/90)
In article <RAMSDELL.90Jul10070345@frieda.mitre.org> ramsdell@mitre.org writes: >In article <112780@linus.mitre.org> ramsdell@mitre.org writes: >>It seems that NeXT is committed to the use of antiquated CPU's. > >[1] It makes little sense to replace the CPU of a NeXT Computer when >instead you can simply add a CPU. Is that not one of the advantages >of using Mach as your kernel? Why does the mother board have to >have the major processing power? > [stuff deleted] >John Yeah! Why indeed? It'd be nice to have access to some of that RISC stuff. However if the 040 could off-load all the screen re-draw to a coprocessor and have an upgraded DSP chip handle the array processing (not to mention I/O) ... and the 040 already looks promising. This extra silicon doesn't show up in the box scores though (ie. Dhrystones, etc. ) and marketing has a hard time selling such. Excellent question, though I may be putting a different stress on it. Time to read up on threads and distributed processing. -Myron
edwardj@microsoft.UUCP (Edward JUNG) (07/14/90)
In article <RAMSDELL.90Jul10070345@frieda.mitre.org> ramsdell@mitre.org writes: >In article <112780@linus.mitre.org> ramsdell@mitre.org writes: >>It seems that NeXT is committed to the use of antiquated CPU's. > >I think my use of the word antiquated was inappropriate and overstated >my point. I really had two points in mind. > >[1] It makes little sense to replace the CPU of a NeXT Computer when >instead you can simply add a CPU. Is that not one of the advantages >of using Mach as your kernel? Why does the mother board have to >have the major processing power? Good point, but mixing CPU instruction sets is a pain w.r.t. software distribution (eg to do symmetric MP transparently you would need to have multiple binaries or something slightly vaporous such as ANDF; to do asymmetric MP you make assumptions about the target environment that Mach cannot currently abstrac away). But if your point is just that NeXT should enable MP (multiple motherboards), then I agree. If it's that NeXT should enable coprocessors running Mach in an MP manner (minimal Mach kernel on each "card"), then I also agree. If it's that NeXT should split its efforts along RISC and 680x0 boundries, then I disagree -- I think it would be a bad strategic move at this point in their company lifecycle. They need focus before division. Why complicate the market at this point? > >[2] Let the market place decide if the new processor should be RISC or >CISC. If vendors think they can make a better profit by selling CISC >coprocessor rather than RISC coprocessors, so be it. Let the >competition begin! > >John CISC actually can have advantages in MP situations. I think that NeXT should leverage CISC MP before jumping into the RISC fray, unless they want to solve the difficult compatibility problems. As a side note, Microsoft has OS running on RISC. The issues of multiple CPU architectures are by no means trivial, even at the applications software level. Edward Jung Systems Architecture Systems Strategy Task Force Microsoft Corp. "I speak for myself, not my company."