#UU's MultiProcessor '060
44 messages in this thread
Stephen-
>As the owner of two Toaster systems, I'd much rather purchase a $2K card
>than shell out $6K for a new platform, which I'll have to learn all over
>again. Plus, I'd have to purchase all the software (beaucoup $$) and
spend >all that time getting to where I'm already at with LW. Perhaps I'm
>overestimating my learning curve, I don't know. I've never worked with
any >Power PC software, not to mention SGI and SoftImage et al.
Oh, sure, for existing Amigans $2k forestalls a big problem. But for
-new- Amigans? There's no reason for people to start out with an Amiga as
things stand now, and certainly not if things get worse. Even good
software isn't enough. I haven't used LightWave, but I tell you this –
Alias is a bitch and a half to use. It's not hard to understand, it just
has an interface that is from hell and makes everything take four times as
many steps as it should.
Anway, as far as the UU board, this is what I know. (Maybe other people
will fill in the gaps.)
Jim Drew, pres of UU, said that UU is working on a multiprocessor board
for Amigas that will use from 1 to 4 Motorola 060s. The base board, he
says, will cost ~$1k, with each additional 060 costing ~$300. He has not
said what clock this board will run at – i.e. 50Mhz or 66MHz, or whether
the board will be available in multiple versions for different clocks, or
how much faster clocks will cost. He said that the software required to
enable multiprocessing on the Amiga amounted to 300 bytes of code. Go
figure.
He also maintains that the 060 is more powerful than people think it
is, and that it does -not- run hot. Perhaps the fact that it is a 3.3v
chip explains the latter claim. He also said that Moto plans 100MHz and
200MHz versions of the chip. This seems unlikely to me in light of Moto's
history with the 0x0 series, but I suppose the 3.3voltedness of the 060
could at least help up to 100MHz. Additionally he claimed that UU's board
would -always- be ahead of any PPC or Pentium solutions. Since the 060
delivers 49SpecInt and ??SpecFP at 50MHz this could conceivably be true in
some ways if their multiprocessing software works correctly.
Pentiums are already available at 90MHz and will soon be available at
faster clocks I am sure. There is a rumoured 135MHz 601, and once the 601
and 604 are re-engineered (soon, according to Andrew Carol) they will be
even faster. I find it hard to believe that Moto will release any 060
faster than 66MHz any time soon, but who knows. At any rate, I think even
the 50MHz version won't be here until 3Q94. I expect UU's board will
follow closely, but Jim said nothing about availability.
To give Jim all credit due, while I have expressed skepticism here, he
-did- deliver handsomely with the Emplant. Admittedly it was released
incomplete, but it has developed into an absolutely splendid product. He
says he can deliver multiprocessing; I think that's bs and I'm eagerly
expecting it in the stores. <g> So who knows.
One last note. The 060 is severely handicapped by its 32bit path to
memory, a failing not shared by the Pentium or, I believe, the PPC. Jim
stated that UU's board would compensate for this with interleaving schemes.
Paul
Paul,
Pretty good summary of past messages and some facts, but…
>He says he can deliver multiprocessing; I think that's bs and I'm eagerly
>expecting it in the stores. <g> So who knows.
Do you really think that's bs? I don't. Jim has delivered a product that,
appearently (I don't have an Emplant), multi-tasks with the Amiga OS.
Since the Amiga OS is already pre-emptively multi-tasking it really
doesn't seem like that big a leap to me to change from scheduling
independant tasks on one CPU to scheduling them on several CPUs.
Robb
Robb-
Like I said I think it's bs but I also believe him. <g>
Paul
Paul,
BS and believe, hmm, OK.
Robb
Robb-
It's possible. I see no earthly reason to believe his extravagant
claims, but he surprised me with Emplant, so I have to give hime a bit of
credit. Suspension of disbelief. <g>
Paul
Paul,
I have a feeling he's gonna surprise us all, even perhaps above his
claims. We'll see.
Robb
Robb-
The thing that makes =NO= sense to me is the 300bytes claim. That
can't be possible to implement sophisticated smp, or even less
sophisticated asmp – like Paul H. said, what about shared memory?
Paul
Paul,
300 bytes…it'd have to be pretty tight code!
Robb
Hi again Paul,
Jim stated that the memory system was transparent to the OS, I believe.
So it would stand to reason that he is talking about 300 bytes of code
inserted into the OS, not all the code needed for the board to run.
Matt Straney, 3 hours late, without your luggage
from Grand Ledge — on AutoPilot ——–>
Matt-
Right. OK, so the 300bytes is the patch. I could see that. I wonder
how much -other- overhead there is.
Paul – Minus Pins!
Paul
Alot, to be sure if there are 4 copies of EXEC. Then again if each
processor does its own EXEC, then it wouldn't matter so much, although who
gets to talk to the `chipset' when, is a bit of a bender.
Matt Straney, 3 hours late, without your luggage
from Grand Ledge — on AutoPilot ——–>
> Since the Amiga OS is already pre-emptively multi-tasking it
> really doesn't seem like that big a leap to me to change from
> scheduling independant tasks on one CPU to scheduling them on
> several CPUs.
That will not help the average program. Sure, each program task could run on
it's own CPU, but if I want a _single_ calculation to run faster I'm screwed!
Unless that application had been designed to break into lots of tiny treads,
that _one_ flow of control must run on one CPU. I bet 95% of all Amiga
programs assume a single thread per program. Why, until now, assume otherwise?
Of course I could end up with my word processor on one CPU and my 3D program on
another, but neither will run any faster than a single chip could. Of course I
could run four instances of the 3D program at once! (What a waste!)
In effect my expensive quad CPU machine will run more programs at once, but
each program will not run any faster than a single dedicated '060 could run it.
This does not count custom programs which are specificly re-coded to use
threads to fix this.
Even though a PowerPC 604 chip is twice as expensive as an '060, it is three
times faster and saves the overhead of buying _mulitiple_ memory systems! Each
'060 needs glue chips, RAM, etc, etc. A single super speed CPU only needs one
of each. Four '060's have only 25% more raw CPU power than a single 604, yet
it will be hard to concentrate all that '060 power into a single, non custom
purpose.
A complete populated 604 mother board with similar total RAM would almost
certainly be cheaper than a populated quad '060 board, and the 604 would
generally be faster in practical use. i.e. making the average load of programs
run faster. There won't be any "wasted" cycles on one CPU which can't be
dedicated to the primary task on the "other" CPU.
This is not to say that multi CPU systems are not useful, it's just that the
software to take advantage of them is just not out there. If all programs knew
how to break their problems into smaller sub-tasks which could be done in
parallel (not an easy job), then we could all just pop into multi CPU machines!
That day is several years away. Oh well.
—– Andrew
Andrew,
Yep, I basically agree. I guess I'm coming from a point of view that
assumes with an MP system there is going to be thousands of tasks all
wanting a bit of the CPU cycles. If you've got one task, no, it won't
'easily' help having multiple CPUs, unless the program is coded is have
several independant tasks.
There are alot of programs that use multitasking (eg. AP which I'm using
now). I'd question your 95%.
I'm all for the PowerPC. We'll see.
But, no, the day of the several CPU machines has long past you by. We've
had 'em on the mainframes for (it seems like) ages.
Robb
> But, no, the day of the several CPU machines has long past you
> by. We've had 'em on the mainframes for (it seems like) ages.
My world is the military and we've had'em for years. It just hasn't "clicked"
yet in the PC world.
We need cheaper CPUs, an OS that can handle it, and program paradigms that
support parallel work. Many tasks just can't be broken into seperate tasks.
I agree a four '060 config in an Amiga would help raytracing folk a _lot_, but
most others would see a minimal speedup.
People somehow have it in their heads that a multi CPU system works by pooling
all the power and then dividing it up on the tasks! It don't work that way!
If the task can't be broken down, it only runs on one CPU!
This implies the "average" user might only get two CPU's worth of work out of
it. His "big" problem will hog 100% of one CPU and all the rest of his stuff
(OS, daemons, comm program, game) would probably run okay on a second CPU.
Of course you could come up with special cases otherwise, but todays current
code will still run fastest on one super CPU than one four medium CPU's. There
are a few exceptions of course. (But not many!)
——- Andrew
Well said!
Though, I do believe I could take advantage of a multi ASMP setup, and I
don't hardly ever do ray-tracing. 🙂
Bob Comer — Cruising the nets on Autopilot!
Cheyenne, Wyoming U.S.A.
When I first got into the Amiga I was doing alot of Neural Net stuff, so I
learned quick that multiprocessors might be critical. I began working with
Inmos transputers and parallel processing slaved to the Amiga (or PC). In that
environment there is a Unic clone for parallel processors called Helios. I went
to a training session where everybody mentioned what machine they were on. I
was last, and of course, the only Amiga guy. The guy teaching the workshop
laughed and said "Great machine. I wrote the EXEC for the Amiga nad we used
alot of the ideas in buiding Helios for parallel processing" !!! So, AmigaDOS
ought to be much more suitable by nature than the Mac or DOS (not withstanding
the obvious hardware differences).
Wouldn't happen to remember the guy's name that wrote the EXEC would you?
Just curious. I've heard that it evolved from a british design maybe
targeted at the Torch BBC computer originally.
Tom
Oklahoma City, OK
Wrong. Carl Sassenrath is the father of Exec.
AmigaDOS (i.e., dos.library, shell, many c: commands) originally came from
UK — Tim King, et al.
— Dale L. Larson, Intangible Assets Manufacturing
— INTERNET:dale@iam.com
I see, thanks for the information Dale. I seem to recall history of the
amiga file which I need to reread sometime.
Tom
Oklahoma City, OK
BINGO…I met the guy King I believe!
Dale,
What is he upto these days?
Matt Straney, 3 hours late, without your luggage
from Grand Ledge — on AutoPilot ——–>
Yeah, this guy was a Brit. Very proud of the EXEC, said it has never had to be
touched. I might have his name in some materials at home, I'll check
tomorrow…
Tim King worked on Tripos, an OS that was adapted to become AmigaDOS. Exec sits
below that, and was originally designed by Carl Sassenrath. And it has been
touched… in fact, when Syndesis was doing networking products, we found an
Exec-level bug in the area of message passing, believe it or not.
You might be interested to know that Sassenrath wrote a book on exec
called "Guru's Guide to the Commodore Amiga, Meditation #1 -Interrupts"
It was written in 1988, so how much of the hardware description is still
completely accurate I don't know, but the theory is very interesting.
-Karl Speaking from the Rubble.
Thank you Karl, I am sure that will be a real collector's item now. It is
possible that some budding genius will understand it and build the next
people's computer someday I hope.
The news still has me somewhat depressed at the moment I'm afraid.
Tom
Oklahoma City, OK
I don't think there'll ever really be a peoples' computer. The industry
is too big. There's really only a chance when things are in flux, mostly
during the birth of a new industry. If the nature of personal computing
changes significantly, which is certainly possible, that would be the time
for real progress. But as things stand computers will just keep evolving,
and if we're very lucky they'll lose most of the inherited garbage they're
currently carrying around.
Paul – Minus Pins!
Stephen,
The workshop you attended was taught by Carl Sassenrath?
-sja
No, apparantly Tim King, that name rings a bell, I'll check the file at home to
verify, but the name King is sounding very familiar on this.
Stephen,
I've never heard of Tim King. Sassenrath wrote exec.
In the early days… Commodore US was developing the Workbench (Intuition)
and the command line interface was contracted out to Metacommco in the UK.
The origianl Amiga designers _insisted_ on a command line interface.
Tim King was involved with this work as he was the major author of Tripos
which was used as a base for the CLI on the Amiga.
-mark=
Otto Pilot Engaged..
"I've never heard of Tim King. Sassenrath wrote exec."
And I treasure the only book I was able to get from his series "Guru's
Guide to the Commodore Amiga", Meditation #1-Interrupts. Wish I could get
the rest of the series.
// John (Speeding on the Information Superhighway)
\X/ Mercury@ins.infonet.net
Defintely more suitable than DOS or MAC!
Bob Comer — Cruising the nets on Autopilot!
Cheyenne, Wyoming U.S.A.
Was your class taught by Carl Sassenrath (sp?) He is supposed to be the
father of EXEC.
-Jonathan
// Amiga online via Autopilot. MAC Quadra 900 and
\\// 486 emulation. E-Mail: jonathanb@delrina.com
I believe Helios was the OS run on the vapor MP transputer board from
Germany back in 88, that made all the headline in the Amiga rags then
never heard another
peep out of it..
…Steve
====>>> Motorola 68030 Inside <<<====
Helios is an OS used on transputer's today! (And has been for at least 5
years…)
And, yes, that german transputer was supposed to have run Helios.
Bob Comer — Cruising the nets on Autopilot!
Cheyenne, Wyoming U.S.A.
Well I actually booted it on 4 transputers off the pc side of a bridgeboard
setup. It exists, weird as it seems!
By way of analogy, when 8088 PCs were the norm, a number of coprocessing
accelerator boards (usually 286, sometimes 386) emerged. These would use
the higher CPU for the real work, and relegated the 8088 to overseeing all
the I/O stuff, like disk operations, control of printers and other
devices, cache management, etc. Possibly an Amiga multiprocessing (OK,
coprocessing) scheme could do likewise and provide a tangibly faster
system with some software. The workhorse CPU running your fave art program
would pass mundane tasks off in batches to the other for disk storage or
other purposes. I'm no engineer, but I used to have a PC that did this
with pretty snappy results; strikes me that an Amiga could do it better.
Of course, I doubt a true MULTI-processor system would benefit much beyond
a CO-processor system at these kinds of tasks.
Paul,
Your comments about UU's board are most interesting, but your comment about
Emplant and its release makes me finally comment about something that I've
thought for a long time. To me, the most famous prolonged gestation period of
Amiga hardware was the now well-loved Toaster. Newtek had this thing in the
works for a very loooooooong time. I doubted that it was EVER going to show up.
But it did and the rest is history. For whatever reason it seemed that Amiga
users were more patient with that product than they were with Emplant so Newtek
was able to hold off release until it was a pretty stable product.
Bill
"Perhaps the fact that it is a 3.3v chip explains the latter claim."
I thought it was even lower than that, like 1v.
-Chris
Paul,
I heard years ago, when 2.04 came on the sceen that there was already
built into the system a way to have multi-processing, as opposed to
multi-tasking. This makes sense to me, though CBM doesn't look ahead, at
least its R&D staff does. I also read posting on bix(same time frame ~2
yrs ago) from people who claimed to be doing just that on 040 accellerated
amiga 3000's, using the 030 & 040, though not with any real success, No
Stability.
As for clock speed on the 68000 family, I am not so sure that the 060
is a direct member of the 68000 family. The way I understand it the 050
was junked because of realestate bottle-necks, and they used alot of
risc/PPC technology inside the 060 chip. That is to say it runs 680×0
code but the actual circuitry has little to do with the 68000, so it is
instruction compatible, but only pin-compatible with the 040(well not
perfectly). I think Moto has learned alot from the PPC developement, and
I'm sure IBM learned a thing or three, I don't want to imply that IBM is
lame or dim, just that they make chips/PCs/mainframes/etc., it doesn't
seem to me like they really needed Moto to get the deal done, but
obviously they knew they did. I think it may be a little easier to get
the clock up on this chip than the last one, because they know more now.
I do agree on the bus width, it is important, as I think that the
highest clocks will ever get is about ~330 mhz, well at least until
something drastic happens.
Matt Straney, 3 hours late, without your luggage
from Grand Ledge — on AutoPilot ——–>