melling@cs.psu.edu (Michael D Mellinger) (09/27/90)
Forgive me if I seem to be jumping the gun, but now is the time for NeXT to consider this. I believe that NeXT's biggest advantage over Apple is that its OS isn't written in 68000 assembler. This means that they can develop a new machine around a new high performance RISC chip. The new generation of RISC's are going to be in the 40 mip range. I doubt if the 68050 will be ready before 1992. The new generation RISC's should appear by early next year. By next September NeXT could have a 40 mip machine on the market. Apple will surely have an 040 Mac by January, but they have to stop there until they rewrite their OS. I hope NeXT is hard at work building a RISC based machine, so that they can maintain the performance edge they now have over Apple. -Mike
bruce@atncpc.UUCP (Bruce Henderson) (09/27/90)
In article <F_-0h+b2@cs.psu.edu>, melling@cs.psu.edu (Michael D Mellinger) writes: > > Forgive me if I seem to be jumping the gun, but now is the time for > NeXT to consider this. I believe that NeXT's biggest advantage over > Apple is that its OS isn't written in 68000 assembler. This means > that they can develop a new machine around a new high performance RISC > chip. The new generation of RISC's are going to be in the 40 mip > range. I doubt if the 68050 will be ready before 1992. The new > generation RISC's should appear by early next year. By next September > NeXT could have a 40 mip machine on the market. Apple will surely > have an 040 Mac by January, but they have to stop there until they > rewrite their OS. I hope NeXT is hard at work building a RISC based > machine, so that they can maintain the performance edge they now have > over Apple. > > -Mike you can pretty much rest assured that this isn't going to happen. I don't think that MIPS or raw CPU horsepower is going to be a concern for a while. Reasons: 1) the 040 at 25MHz is a pretty fast chip. Any guesses how the 33MHz (or 50MHz) version will be? 2) Re-working EVERYTHING for a binary incompatible CPU would be a multi-year nightmare. Not only that but you effectively upset all of your customers who have invested big $$$ in software, peripherals, etc, when you inform them that none of it works with the new hardware (binary incompatibility) 3) NeXT doesn't need to be the fastest machine in the world. If speed is what you crave, buy a Cray... 8-) 4) NeXT has the capability to sell a 100 MIPS binary compatible machine within the next 12 months. Stay tuned for details. Bruce Henderson Ashton Potato Goon Squad
engstrom@SRC.Honeywell.COM (Eric Engstrom) (09/28/90)
In article <F_-0h+b2@cs.psu.edu> melling@cs.psu.edu (Michael D Mellinger) writes: > Forgive me if I seem to be jumping the gun, but now is the time for > NeXT to consider this. I believe that NeXT's biggest advantage over > Apple is that its OS isn't written in 68000 assembler. This means > that they can develop a new machine around a new high performance RISC > chip. NeXT has a large number of advantages over Macintoshes, some of which I, PERSONALLY, believe are more deserving of attention that what the OS is written in. Granted, the fact that the OS is Mach, written in C, and probably most of the rest of the system is also written in C or Objective-C, does give NeXT the opportunity to port it to other CPU types, which Apple may not have with the Macintosh, but that might not be the way to go. (read on...) > The new generation of RISC's are going to be in the 40 mip > range. I doubt if the 68050 will be ready before 1992. The new > generation RISC's should appear by early next year. Ok, so they have to wait a few months before they get prototype 050s. Motorola's not going to make NeXT wait until their release date to play with the 050s... In addition, diverging the NeXT architecture into two very distinct types might plague them with more problems. For example, Sun had in the Sun3 a fairly nice machine. Along comes the Sun4 (Sparcstaion) using RISC technology. Sure its faster, but trying to integrate the two types of machine into one network is no minor task. "But," you say, "Why would I ever want a Sun3 when the Sun4 is so much better?" (point debatable). What if you already own <x> Sun3s (or NeXT 030s) and you want to buy say <y> Sun4s (or NeXT-RiSCs). Not only are you going to have to purchase the machines, but you will also have to double (or more) your file system space to hold all of the doubled binaries (as they are not compatible between the two architectures). In addition, you will probably have to hire at least one more person to help with the load of maintaining the expanded file-system. Now, I must say that I think Suns and NeXTs of any denomination are nice machines (much better that those that I will leave unnamed). Also, I'm not claiming to be any marketing wizard, so perhaps this kind of move is actually very appropriate. > By next September > NeXT could have a 40 mip machine on the market. Apple will surely > have an 040 Mac by January, but they have to stop there until they > rewrite their OS. I hope NeXT is hard at work building a RISC based > machine, so that they can maintain the performance edge they now have > over Apple. Sure NeXT has a performance edge over Apple, but what kind of Macintosh are you thinking of which could compete with a NeXT in terms of MIPS? Even so, Why are you stuck in the "Magic Number Syndrome," as someone (I don't remember) on this newsgroup stated earlier. If you were to look a real applications usage, a NeXT would always out-perform a Mac. Then, if you take into account the relative prices and amount of NeXT bundled software (yes there still is some ;-) I think the NeXT is a much better buy... Eric +--------- Eric Engststrom, Honeywell SRC | ARPA: engstrom@src.honeywell.com | [DISCLAIMER: UUCP: {umn-cs,ems,bthpyd}!srcsip!engstrom | My own opinions - 3.1415 = MAIL: 3660 Technology Drive, Mpls, MN | orgainically grown] Phone: (612) 782-7318 |
azure@portia.Stanford.EDU (Lai Heng Chua) (09/29/90)
What is important for NeXT in the coming year? Right now they have to build a big enough base for sure. I'm amazed, I can't get a decent 030 Mac for the price of an 040 NeXT. Unfortunately, NeXT abandoned two of their selling points (and uniqueness) - the bundled applications and the OD. The OD can be dropped and recovered later but there should be consistency on the bundled applications for a long time frame. Take Lisp for example. If you develop Lisp applications for the NeXT that goes for $250, you don't want to have to have the buyer pay $1500 for the Lisp. Who is going to buy your app.? If you have an app that messages Mathematica and the latter doesn't come with all machines, what is the use. Same for Sybase. These are enabling software and NeXT was right to bundle them with the machines. I think it becomes much harder to develop powerful apps for the NeXT with many of the underpinings gone. Software costs will also be expected to be higher. Fortunately, Webster is still around. Every app can message Webster to help the user. Some suggestions: NeXT work on making multiprocessing on the NeXT work in anticipation of multi-CPU NeXT. That is compiler, libraries and multi-CPU architecture. The object oriented paradigm is a help but something like Gul Agha's Actors would have been better. Programmers should not have to worry too much about threads etc. Maybe they should develop Actor-C. Mach is just great. Watch non-unix machines trying to go multi-CPUs. 4 040 should yield about 60 MIPS. MIPS is not everything but I'm afraid it is still important to keep up with the competition. Applications grow to consume the MIPS available. Put together a database strategy. All these small little databases for each app. doesn't do it. Client/server allows dynamic sharing of data but a unified database provides persistence. You can interrupt your work and get back to the state you were in. Imagine, if you want to you can get back all your windows with the appropriate files open and in the previous locations. Different people can work on accounting data etc. Why do you thing companies use IBM mainframes and resist the PCs? Why are the mainframes still around and well....? I remember NeXT says that it is going for leveraging group work and database is really important. Object database may be the way to go given the object orientedness of NeXT. How about integrating all the help documentation? Chua chua@cive.stanford.edu
melling@cs.psu.edu (Michael D Mellinger) (09/29/90)
In article <340@atncpc.UUCP> bruce@atncpc.UUCP (Bruce Henderson) writes:
you can pretty much rest assured that this isn't going to happen.
I don't think that MIPS or raw CPU horsepower is going to be a concern
for a while. Reasons:
1) the 040 at 25MHz is a pretty fast chip. Any guesses how the
33MHz (or 50MHz) version will be?
DEC is rumored to be releasing a 40 mips machine based on the MIPS
R4000 within the next six months. RISC architetures that process
several instructions in one cycle are being developed by Moto., MIPS,
etc. I think the IBM RS/6000 can do 4 instructions in one clock
cycle(ideally). A CPU that runs at 50MHz and completes 4 instructions
(on average) a cycle is going to give you 200 mips.
DEC is rumored to be releasing a 40 mips machine based on the MIPS
R4000 within the next six months(MIPS has a 55 mips machine in beta).
RISC architetures that process several instructions in one cycle are
being developed by Moto., MIPS, etc. I think the IBM RS/6000 can do
4(I think. Maybe only 2) instructions in one clock cycle(ideally). A
CPU that runs at 50MHz and issues completes 4 instructions (on
average) a cycle is going to give you 200 mips( x4 and you have a Cray
in a Cube).
2) Re-working EVERYTHING for a binary incompatible CPU would
be a multi-year nightmare. Not only that but you effectively
upset all of your customers who have invested big $$$ in
software, peripherals, etc, when you inform them that none
of it works with the new hardware (binary incompatibility)
The software problem can be solved by packaging two binaries with your
product. Hopefully in a year another viable way of distributing
software will be found(20 MB floppies, DAT, Optical, CD ROM).
3) NeXT doesn't need to be the fastest machine in the world.
If speed is what you crave, buy a Cray... 8-)
Right, they just have to remain competitive. Sun, DEC, HP, and IBM
aren't just going to watch NeXT take a huge bite out of their markets.
Expect to see some faster machines from them.
4) NeXT has the capability to sell a 100 MIPS binary compatible
machine within the next 12 months. Stay tuned for details.
Does the new cube have all the hardware necessary to support
multi-processing?
Bruce Henderson
Ashton Potato
Goon Squad
Is Ashton Tate planning to port FullWrite to the NeXT? I've heard a
lot of good things about it. The main complaint was that it was slow.
Probably wouldn't be slow on an 040 NeXT :-).
-Mikeedwardj@microsoft.UUCP (Edward JUNG) (10/01/90)
IMHO the main reason to look at other CPUs is *not* principally speed, performance, or what have you. It is independence. Companies such as NeXT need to stay loose and mobile to remain competitive. NeXT has taken many steps toward this: 1. a modular, portable OS 2. a modular, portable development environment 3. standard networking But consider that currently NeXT is dependent upon a single vendor, ie Motorola, to deliver CPUs upon which their entire marketing strategy rests. A delay in delivery, which has happened more than once already, translates into a delay in shipment, loss of revenue, loss of mindshare, and loss of reputation in general. This is a dangerous exposure for a small, innovative company that needs to remain competitive in a timely manner. Much is made of NeXT's ability to design and deliver an entire new system in eight months, but this only happened because it was eight months that happened to mesh with Motorola's schedule. For other parts NeXT ostensibly could have gone with other architectures, but for CPU there was only one choice. Open CPU architectures such as SPARC and MIPS allow designers the freedom of looking toward many vendors of the CPU, more types of CPUs with varying performance and pricing, and even the flexibility of manufacturing the chip for themselves. There is less dependence upon a single company to service the needs of all clients of the architecture. You get to spread your bets, so to speak. NeXT could make this leap and have a new binary standard before it is too expensive to do so (one could argue, though, that at this point in their marketing strategy it would be fatal to switch). Witness MIPS, for example. They do not really produce chips themselves, but you can buy a variety of MIPS CPUs for all kinds of applications. A 64-bit MP-capable 50+ MIPS chip will be available along with various embedded versions (eg for a laser printer), high-speed ECL, and numerous high-speed FP units -- well-before the higher speed 68050 is available, for example. So IMHO the reason to move to a different CPU architecture has more to do with independence, freedom, and flexibility than any other reasons. -- Edward Jung Microsoft Corp. My opinions do not reflect any policy of my employer.
bruce@atncpc.UUCP (Bruce Henderson) (10/02/90)
In article <Fyeiqvd2@cs.psu.edu>, melling@cs.psu.edu (Michael D Mellinger) writes: > > Is Ashton Tate planning to port FullWrite to the NeXT? I've heard a > lot of good things about it. The main complaint was that it was slow. > Probably wouldn't be slow on an 040 NeXT :-). > > -Mike I really can't tell you wether we are or not. But if enough people like yourself ask for it, I'm sure we would see it. But if we did have it, it probably would be nice and peppy on and '040 8-) Bruce Henderson "Livin' in my own private Idaho Ashton Potato
jim@jagmac2.gsfc.nasa.gov (Jim Jagielski) (10/05/90)
In article <57835@microsoft.UUCP> edwardj@microsoft.UUCP (Edward JUNG) writes: > >Witness MIPS, for example. They do not really produce chips themselves, > Sorry, but MIPS have been designing their chips for a number of years. In fact, it's THEIR chips that people like Dec, SGI, CDC, etc are using for their machines -- ======================================================================= #include <std/disclaimer.h> =:^) Jim Jagielski NASA/GSFC, Code 711.1 jim@jagmac2.gsfc.nasa.gov Greenbelt, MD 20771 "Kilimanjaro is a pretty tricky climb. Most of it's up, until you reach the very, very top, and then it tends to slope away rather sharply."
phil@motaus.sps.mot.com (Phil Brownfield) (10/06/90)
In article <3560@dftsrv.gsfc.nasa.gov> jim@jagmac2.gsfc.nasa.gov (Jim Jagielski) writes: >In article <57835@microsoft.UUCP> edwardj@microsoft.UUCP (Edward JUNG) writes: >> >>Witness MIPS, for example. They do not really produce chips themselves, >> >Sorry, but MIPS have been designing their chips for a number of years. In >fact, it's THEIR chips that people like Dec, SGI, CDC, etc are using for >their machines Design != produce. MIPS has no silicon fabrication facilities; their MPU customers either license the rights to fab silicon themselves and/or buy from foundries. -- Phil Brownfield phil@motaus.sps.mot.com {cs.utexas.edu!oakhill, mcdchg}!motaus!phil Speaking for myself, not my employer
rca@cs.brown.edu (Ronald C.F. Antony) (10/06/90)
In article <1990Sep28.233054.2605@portia.Stanford.EDU> azure@portia.Stanford.EDU (Lai Heng Chua) writes: >Take Lisp for example. If you develop Lisp applications for the >NeXT that goes for $250, you don't want to have to have the >buyer pay $1500 for the Lisp. Who is going to buy your app.? Fortunately you can get a runtime license for Lisp and distribute it with your app. You then have to pay 2% of your price as license to Franz, inc. >If you have an app that messages Mathematica and the latter doesn't >come with all machines, what is the use. Very true, unfortunately... >Same for Sybase. >These are enabling software and NeXT was right to bundle >them with the machines. Yes NeXT WAS right. You still can get Sybase for a nominal fee (what ever this means), but many people will not do that. The good thing about the sybase server was that it could have been a replacement for all the silly file standards. Instead all progrmas that are somewhat data-driven could have accessed the same data from the data-base. No more problems with incompatible data-formats. No more being locked into one sigle application. Of course this would have meant that NeXT should have developed also a dbKit. Such a dbKit could even contain some object-oriented extensions and ... Fortunately not everything is lost yet. NeXT did not abandon so much that readapting makes them look silly. On the other hand we never know what sort of unreasonable claims were made by the producers of the software in question. If I look at the product catalog and some of the high prices there, I think I know why NeXT had to step back from some of their plans. Ronald ------------------------------------------------------------------------------ "The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man." Bernhard Shaw | rca@cs.brown.edu or antony@browncog.bitnet
keithp@xavier.tamu.edu (Keith D Perkins) (10/06/90)
On the subject of the Sybase Database... I've heard that 2.0 will have a database class. Steve Jobs used it in his demo on the 18th. Keith Perkins Texas A&M University
dd26+@andrew.cmu.edu (Douglas F. DeJulio) (10/06/90)
rca@cs.brown.edu (Ronald C.F. Antony) writes: > In article <1990Sep28.233054.2605@portia.Stanford.EDU> azure@portia.Stanford.EDU (Lai Heng Chua) writes: > >Take Lisp for example. If you develop Lisp applications for the > >NeXT that goes for $250, you don't want to have to have the > >buyer pay $1500 for the Lisp. Who is going to buy your app.? > > Fortunately you can get a runtime license for Lisp and distribute it > with your app. You then have to pay 2% of your price as license to > Franz, inc. How would this work if you want to put a program in the public domain? -- DdJ
dorner@pequod.cso.uiuc.edu (Steve Dorner) (10/09/90)
In article <Fyeiqvd2@cs.psu.edu> melling@cs.psu.edu (Michael D Mellinger) writes: >etc. I think the IBM RS/6000 can do 4 instructions in one clock >cycle(ideally). A CPU that runs at 50MHz and completes 4 instructions >(on average) a cycle is going to give you 200 mips. No, because RISC instructions do less. You have to normalize to VAX 780 mips if you want to make "real" comparisons. (If you really want to make "real" comparisons, you have to try doing YOUR REAL WORK on the machine; even the best benchmark can be very, very deceptive.) -- Steve Dorner, U of Illinois Computing Services Office Internet: s-dorner@uiuc.edu UUCP: uunet!uiucuxc!uiuc.edu!s-dorner
hunt@boulder.Colorado.EDU (Lee Hunt) (10/09/90)
In article <1990Oct8.182201.7907@ux1.cso.uiuc.edu> dorner@pequod.cso.uiuc.edu (Steve Dorner) writes: >In article <Fyeiqvd2@cs.psu.edu> melling@cs.psu.edu (Michael D Mellinger) writes: >>etc. I think the IBM RS/6000 can do 4 instructions in one clock >>cycle(ideally). A CPU that runs at 50MHz and completes 4 instructions >>(on average) a cycle is going to give you 200 mips. > >No, because RISC instructions do less. You have to normalize to VAX 780 >mips if you want to make "real" comparisons. > >-- >Steve Dorner, U of Illinois Computing Services Office >Internet: s-dorner@uiuc.edu UUCP: uunet!uiucuxc!uiuc.edu!s-dorner As I understand it, "RISC" in the RISC/6000 is a misnomer -- the machine is only called a RISC machine because it processes at least 1 instruction per cycle. The instruction set is quite large (~180 opcodes, I think I recall). So, in this case the instructions are as "watered down" as in the "conventional" RISC machines (although I do believe that the more complex instructions must be run serially, 1 clock cycle each). | --Lee | | Dave Haynie is Bob! | | hunt@spot.colorado.edu ...!ncar!boulder!spot!hunt |
rca@cs.brown.edu (Ronald C.F. Antony) (10/12/90)
In article <0b3U_TS00VI8EgY2Jm@andrew.cmu.edu> dd26+@andrew.cmu.edu (Douglas F. DeJulio) writes: >rca@cs.brown.edu (Ronald C.F. Antony) writes: >> In article <1990Sep28.233054.2605@portia.Stanford.EDU> azure@portia.Stanford.EDU (Lai Heng Chua) writes: >> >Take Lisp for example. If you develop Lisp applications for the >> >NeXT that goes for $250, you don't want to have to have the >> >buyer pay $1500 for the Lisp. Who is going to buy your app.? >> >> Fortunately you can get a runtime license for Lisp and distribute it >> with your app. You then have to pay 2% of your price as license to >> Franz, inc. > >How would this work if you want to put a program in the public domain? >-- >DdJ Don't ask me :-) I would guess that 2% of 0$ are 0$. But there is a problem with the expression public domain. You can't put something into public domain if others have copyright on parts of it. I guess you would have to sell it for 0$ or get some lawyer write a license agreement that makes clear that not all of the program in PD. Or you could publish the source and put the source into PD and sell the compiled product for 0$. Maybe someone from Franz cares to comment on this one... Ronald ------------------------------------------------------------------------------ "The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man." Bernhard Shaw | rca@cs.brown.edu or antony@browncog.bitnet
moose@svc.portal.com (10/13/90)
In article <52954@brunix.UUCP> rca@cs.brown.edu (Ronald C.F. Antony) writes: >I would guess that 2% of 0$ are 0$. But there is a problem with the >expression public domain. You can't put something into public domain >if others have copyright on parts of it. I guess you would have to >sell it for 0$ or get some lawyer write a license agreement that makes >clear that not all of the program in PD. Or you could publish the >source and put the source into PD and sell the compiled product for >0$. Maybe someone from Franz cares to comment on this one... This can't be right. I'm a programmer not a lawyer, but here's what I know. On the Macintosh, under Think C (Formerly Symantic C, Formerly Light Speed C), anything you develop must have a copyright notice saying that parts of this program are copyright Think C. Even stuff in the public domain. The problem may be terminology. We have Public Domain, Freeware, and freely redistributable. I'm not sure what the legal definitions are, but I'd guess that at least one of these would allow a programmer to redistribute the LISP kernal at no charge. Then again, Franz may not give you the license. -- Michael Rutman | moose@svc.portal.com Cubist | makes me a NeXT programmer Software Ventures | That's in Berkeley smile, you're on standard disclaimer | <fill in with cute saying>