nather@utastro.UUCP (Ed Nather) (05/21/84)
[]
I played with an ATT 3B2 a couple of days ago, running system V Unix,
so I guess the secret is out. A "back door" copy was sent to an OEM
who has signed up a a local dealer, apparently one of the first in the
US, and plans to provide business applications software for it and its
larger brothers. He got a 3b2 with 1 MB of memory, ~30 MB disk, and
the WE 32000 chip(s) -- it's several separate chips on a common carrier.
The hardware is spectacular. You can hold the 1MB card in the palm of
your hand. Most of the room is taken up by the two disks -- a 5-1/4"
floppy and a (CDC) hard disk with the same size footprint -- and the power
supply. The computer is almost an afterthought. The basic motherboard
is less than 1/2 the size of an IBM PC's, with 4 I/O slots hovering
above it (only one was used on the machine I saw, 3 were empty) and it
used modular plugs (what else?) to connect terminals. The Unit I saw,
with an external terminal added, would run about $15,000 list. (Is there
a net price? Does AT&T have educational discounts? Hello?)
The only benchmark I had time to run was the "sieve" from Byte Magazine,
which ran 100 iterations in 36 seconds, or 3.6 seconds for 10. Our Vax
11/780 runs the same benchmark in 1.3 seconds if the variables are
declared "register" and in 2.5 seconds if they are declared "int", for
10 iterations. This places the 3b2 [for integer operations only!] at
about 0.5 Vaces. I didn't have access to the source code for the bench-
mark so I don't know how the variables were declared.
While I was very impressed with the hardware, I was very unimpressed with
the software. They *did* have "vi" but termcaps were not yet set, they
didn't have "more" or "pg" and the "ls" command ran things off the screen
(fast) in a single column display (!). Apparently someone at AT&T decided
to chop Unix up into clumps of a few programs each, and sell them separately.
You can get the "C" compiler in one package -- but the assembler and loader
are in a different one! [If you use "C" you don't *need* assembler, right,
guys?] Nroff was conspicuous by its absence, as was troff and any other
word-processing software.
This may all be growing pains at AT&T, who never had to sell anything ever,
and they just don't know how people use their computers. I hope their
learning curve is very steep.
But that little chunk of harware is a *honey*! I want one.
--
Ed Nather
ihnp4!{ut-sally,kpno}!utastro!nather
Astronomy Dept., U. of Texas, Austintwh@mb2c.UUCP (Tim Hitchcock) (05/21/84)
Sure, let's start with how AT&T ships their 3B's with a number of bad boards, then tell's you to wait a couple weeks (or months) to get new ones. We have brought up two 3B20's at MI Bell, and it happened both times.
guy@rlgvax.UUCP (05/24/84)
> I played with an ATT 3B2 a couple of days ago, running system V Unix, > so I guess the secret is out. It's definitely been announced publicly (full-page ads and everything), so it certainly is out. > The only benchmark I had time to run was the "sieve" from Byte Magazine, > which ran 100 iterations in 36 seconds, or 3.6 seconds for 10. Our Vax > 11/780 runs the same benchmark in 1.3 seconds if the variables are > declared "register" and in 2.5 seconds if they are declared "int", for > 10 iterations. This places the 3b2 [for integer operations only!] at > about 0.5 Vaces. I didn't have access to the source code for the bench- > mark so I don't know how the variables were declared. I suspect the benchmark was written to make full use of the 3B's capabilities, and it has a number of registers (not quite as many as the VAX-11 - somewhere around 11 32-bit general purpose registers, if I remember correctly), so I suspect they were declared "register" - so it's about 1/3 11/780 under those circumstances. I'd be curious to see the Bell B1-B7 benchmarks on that machine. > While I was very impressed with the hardware, I was very unimpressed with > the software. They *did* have "vi" but termcaps were not yet set, they > didn't have "more" or "pg" and the "ls" command ran things off the screen > (fast) in a single column display (!). Hmmm... I'd heard System V Release 2 had "pg" and a Berkeley-style "ls". Either I heard wrong or they didn't have 5R2 on that 3B. > Apparently someone at AT&T decided to chop Unix up into clumps of a few > programs each, and sell them separately. You can get the "C" compiler in > one package -- but the assembler and loader are in a different one! [If you > use "C" you don't *need* assembler, right, guys?] *Choke* Geepers, don't "pcc" (which is the basis of almost all the UNIX C compilers out there) and most other UNIX C compilers produce assembly code which is then run through the assembler? Sounds like you can't even produce object code, much less executable programs, with that package! *D*U*M*B*, guys... > Nroff was conspicuous by its absence, as was troff and any other > word-processing software. > This may all be growing pains at AT&T, who never had to sell anything ever, > and they just don't know how people use their computers. I hope their > learning curve is very steep. Fortune packaged/packages their UNIX systems similarly; I suspect it may be aimed at the end user who buys a box and a bunch of applications software (including, possibly, a what-you-see-is-what-you-get word processor which will probably not only be considerably more user-friendly than "vi" and "nroff", but a lot faster than those two beasts to boot!) and won't need all that stuff. On the other hand, if they don't have any fully-packaged systems to sell to people who want a computer, not a turnkey box, they'll probably get beaten up. Guy Harris {seismo,ihnp4,allegra}!rlgvax!guy
hart@cp1.UUCP (05/24/84)
Nroff and etc are available in a package called documentation workbench.
The local reps have not circulated a price list yet, we are holding our
breath. Hopefully the stuff will be priced reasonably.
--
======================================================================
signed: Rod Hart (wa3mez)
Chesapeake & Potomac Tel. Co.
Bell Atlantic Inc.
Silver Spring, Md.
gamma!cp1!hart - umcp-cs!cp1!hart - aplvax!cp1!hart
======================================================================nather@utastro.UUCP (05/25/84)
[]
Guy's insight is correct -- some checking showed that the "sieve"
benchmark on the 3b2 that took 3.6 seconds *did* have the variables
declared "register" -- without that nicety the same benchmark takes
6.3 seconds, about the same ratio as on the Vax 11/780. This puts
the 3b2 as 0.4 Vax 11/780's.
I'm told by the AT&T salespeople that the clock rate on all 3b2's
shipped after 1 July 84 will run at 10 MHz, while the current ones
have a clock rate of 7.2 MHz. If this is correct it will raise the
3b2 to 0.6 Vaxen (for fixed point operations).
Curiously, both 3b2 units I have seen have TWO clock crystals on the
main motherboard, one labeled 28.8 MHz and the other 20.0 MHz. Maybe
you can choose whatever clock speed you want by deciding how you divide
down the clock. (28.8/4 = 7.2 MHz, 20.0/2 = 10 MHz) I'd like to try
one at a full 20 MHz and see if it would run!
The demo unit shown at the University of Texas yesterday had a C compiler
but lacked as, ld and the C libraries, so we couldn't benchmark it.
The minimal 3b2 (512K memory, 10 MB disk, 720K floppy, no terminal) sells
for $8500, with "core" Unix system 5 bundled in. "Core" Unix, as now
defined, includes the compiler but not the assembler, loader, or libraries.
The salespeople were sufficiently embarrassed by this (as well as by the cost
and performance of the teletype terminal shown with it) that these things
are likely to change quickly.
Another oddity: they have "vi" running, but we were unable to locate any
termcap or equivalent -- on startup, the software asked which (of 2!)
terminals were connected, and "vi" displayed glitches indicating it got
the details wrong. "vi" without "termcap" is like "cc" without "as".
They have a lot to learn.
But the hardware is a real ... uh ...gem.
--
Ed Nather
ihnp4!{ut-sally,kpno}!utastro!nather
Astronomy Dept., U. of Texas, Austinrej@cornell.UUCP (Ralph Johnson) (05/25/84)
I don't understand why people seem to think that the 3B is such a wonderful machine. "Nice", OK, but why "wonderful"? I have seen a number of 68000 based machines that are just as powerful and are cheaper. I assume that AT&T can provide better service (eventually) and that they could provide better versions of Unix (if they wanted to) but the 3B really seems a lot like many other systems that have been available for a while. Since a 10MZ 68000 is about 60% of a 11/780, a 16MZ is 96%! :-) (Please, no flames if you didn't get the joke.)