Forum unknown
· Hot News & Rumors
#New Amiga's?
65 messages in this thread
A bit more information on the A2500UX… It's an A2000 with the
A2620 (68020 card) with the 4 Meg option (on the card), a 100 Meg
SCSI disk drive and AT&T UNIX ™ version 5.3 Present Amiga A2000
can be updatted to the A2500UX by simply adding the drive and
card. Cost of a new A2500UX is expected to be $7,500. (whoops…I
also believe it comes with the new hi-res monitor)
Don, I would be really interested as to how the A2500UX would fare against
the CSA '020 card. Do you know at what speed the '020 in the A2500 would
run? 14Mhz? Also is the 4 meg memory static or just normal fast ram. Just
wondering.
P.S. did you know that CSA have already got a 68030 piggy-back for the '020
board. I would really like to get my hands on one.
I have no test results to refer to…just 'intuitive' guessing on
my part as to how the 2 boards would stack up….about equal. Last I
talked to Hazy, the RAM on the 2620 card was going to be dynamic RAM…that
could have changed, but I doubt it.
p.s. Whoops…Hazy did say that the MMU would introduce 1 wait state just
because it's there….so if the CSA board doesn't have an MMU, then it
would be faster, but not useable in a UNIX environment. If it has
one…then my original esitmate would hold.
Don, there isn't an MMU on the CSA board. I have also got 1/2 a meg of
static RAM on mine. This really makes it shine. You can actually see the
difference between the same program loaded twice, one in static ram and one
in normal fast ram. There is about a 2:1 difference in speed. During my
usage of the CSA board, there isn't much improvement on programs loaded
into normal fast or chip ram, since there is a slight delay due to address
translation from 32-bit to the normal size. (I think) This makes the real
difference between the 68000 and the 68020 come out when accessing 32-bit
ram. Maybe I'll do some benchmarks and get concrete results.
I believe that as long as there are no wait states induced by the
memory itself…then static and dynamic RAM are the 'same speed' as far as
the CPU is concerned. I think your 2:1 is a result of the 32 vs 16 bit
wide memory bus. Then again, I could be wrong…I'm certainly not a
hardware expert.
That may be true on some CPUs (it is), but on the 68020, you can run the
CPU as fast as the chip will allow, and have it run a lot faster on ram
that has a short access time. You still need the minumum 4 clock cycles to
fetch, but at higher clock rates, the dynamic rams will tend to hold off
the CPU for a few more cucles.
or cycles, to9o. 🙂
And yes… a lot is gained by the wider bus.
That was my point, if you only need 100ns RAM due to clock rate, then
there is no difference between static and dynamic RAM with perhaps an
occasional refresh cycle getting in the way. But, if the RAM is creating
wait states due to the fact that it can't be 'ready' in those 4 cycles…then
yes, static, being significantly faster, should have an advantage over
dynamic RAM.
Don, I guess that the RAM on the CSA 1/2 meg board must be faster that the
normal 'fast' ram. BTW, the connection between the '020 board and the
static ram board is not through the system bus; there are four large
cables running between the two, and they carry the 32-bit bus. Also,
(excuse my ignorance) what use if made of the 32-bit width of each byte?
e.g. is a single char stored in a 32-bit location? That seeme . oops,
seems like overdoing it a bit! Khalid.
Khalid,
Yes….I understand that they have a bus cable going to the 32 bit
RAM. That's because the motherboard only provides 16 bits worth of data
lines. While the '020 can handle that…it requires 2 fetches to get 32bit
data then (just as the 68K does) so would slow things down a bit. That's
also the reason your 32 bit board is a bit faster than the normal stuff.
When it has to fetch a 32bit value…it can do it all at once rather than
having to get 2 16bit fetches.
Single byte values (such as a character array) are stored
consecuitive in memory…that is, 4 of them are returned with a 32 bit fetch.
Since it takes exactly the same amount of time to fetch 8 bits as it does 32
bits, there is no penalty for this. What gets returned depends on what is
being gotten…the rest is considered 'trash'.
At least that's the way I understand it…I'm not a CPU expert, but
reading the Motorola manual…that's what I understand it to say.
Khalid,
Yes….I understand that they have a bus cable going to the 32 bit
RAM. That's because the motherboard only provides 16 bits worth of data
lines. While the '020 can handle that…it requires 2 fetches to get 32bit
data then (just as the 68K does) so would slow things down a bit. That's
also the reason your 32 bit board is a bit faster than the normal stuff.
When it has to fetch a 32bit value…it can do it all at once rather than
having to get 2 16bit fetches.
Single byte values (such as a character array) are stored
consecuitive in memory…that is, 4 of them are returned with a 32 bit fetch.
Since it takes exactly the same amount of time to fetch 8 bits as it does 32
bits, there is no penalty for this. What gets returned depends on what is
being gotten…the rest is considered 'trash'.
At least that's the way I understand it…I'm not a CPU expert, but
reading the Motorola manual…that's what I understand it to say.
Don, I guess that the RAM on the CSA 1/2 meg board must be faster that the
normal 'fast' ram. BTW, the connection between the '020 board and the
static ram board is not through the system bus; there are four large
cables running between the two, and they carry the 32-bit bus. Also,
(excuse my ignorance) what use if made of the 32-bit width of each byte?
e.g. is a single char stored in a 32-bit location? That seeme . oops,
seems like overdoing it a bit! Khalid.
That was my point, if you only need 100ns RAM due to clock rate, then
there is no difference between static and dynamic RAM with perhaps an
occasional refresh cycle getting in the way. But, if the RAM is creating
wait states due to the fact that it can't be 'ready' in those 4 cycles…then
yes, static, being significantly faster, should have an advantage over
dynamic RAM.
That may be true on some CPUs (it is), but on the 68020, you can run the
CPU as fast as the chip will allow, and have it run a lot faster on ram
that has a short access time. You still need the minumum 4 clock cycles to
fetch, but at higher clock rates, the dynamic rams will tend to hold off
the CPU for a few more cucles.
or cycles, to9o. 🙂
And yes… a lot is gained by the wider bus.
I believe that as long as there are no wait states induced by the
memory itself…then static and dynamic RAM are the 'same speed' as far as
the CPU is concerned. I think your 2:1 is a result of the 32 vs 16 bit
wide memory bus. Then again, I could be wrong…I'm certainly not a
hardware expert.
Don, there isn't an MMU on the CSA board. I have also got 1/2 a meg of
static RAM on mine. This really makes it shine. You can actually see the
difference between the same program loaded twice, one in static ram and one
in normal fast ram. There is about a 2:1 difference in speed. During my
usage of the CSA board, there isn't much improvement on programs loaded
into normal fast or chip ram, since there is a slight delay due to address
translation from 32-bit to the normal size. (I think) This makes the real
difference between the 68000 and the 68020 come out when accessing 32-bit
ram. Maybe I'll do some benchmarks and get concrete results.
I have no test results to refer to…just 'intuitive' guessing on
my part as to how the 2 boards would stack up….about equal. Last I
talked to Hazy, the RAM on the 2620 card was going to be dynamic RAM…that
could have changed, but I doubt it.
p.s. Whoops…Hazy did say that the MMU would introduce 1 wait state just
because it's there….so if the CSA board doesn't have an MMU, then it
would be faster, but not useable in a UNIX environment. If it has
one…then my original esitmate would hold.
Don, I would be really interested as to how the A2500UX would fare against
the CSA '020 card. Do you know at what speed the '020 in the A2500 would
run? 14Mhz? Also is the 4 meg memory static or just normal fast ram. Just
wondering.
P.S. did you know that CSA have already got a 68030 piggy-back for the '020
board. I would really like to get my hands on one.
Don, approximatly how much would it cost to add a A2620, 100 meg SCSI drive
and UNIX to a 2000? How much is UNIX by itself and what is the minimum
ram and HD size needed to run UNIX effectively?
I have no idea on the UNIX, the A2620 was to have come in at under
$2000 with 2 Meg…so a guess would be in the $3000 for 4 Meg. A 100 Meg
SCSI drive probably goes for about $1500, the monitor for about $700 so
that'd leave about $300 for the cost of UNIX.
THAT IS A GUESS…NOT CBM figures!!
Not having seen this setup…I can only guess from our setups at work
that it would take a minimum of 50Meg HD and 2 Meg of RAM to even begin to
work, but it'd probably be slow. If it's not a complete full implementation
(similar to what's on the AT&T7300), then you could get away with 1 Meg and
20 Meg HD…but that is a very minimal system…very minimal.
Here's a portion of a usenet msg from Hazy — thought you'd be interested.
——————————————————
> – Is the windowing shell on the A2500UX (and A3000 I would suppose)
based on > X Windows or NEWS ? Commodore would be making a BIG mistake
if it were not.
The windowing system is a custom windowing system, but more of a low-level
window manager. For the UNIX user, it give you a way to start up new
virtual terminals in windows, each with their own shell and some other
attributes, such as terminal emulations. There are also function calls for
programmable graphics. I believe that things like X or NEWS would sit on
top of this underlying window manager when ported to Amiga UNIX.
> – Do the AT and UX have 68851 chips ? I would think the UX would have
to.
Yup. Current plans, as best they've told me, say that all A2620 boards
will have 68851s on them.
> – Will UNIX be available for the 2000 with the x020 board since it has a
> 68851 ? > – Was any hardware shown or is all of this VAPORWARE for the
moment ?
We showed the Amiga UNIX, a real AT&T V.3, running on an A2000 and A2620
board at the show. Even let hackers do stuff on it. Runs like a champ. I
really don't know how close the UNIX is to being complete, but it does work
real nicely now, you can edit and compile on it, its fast, etc. You really
won't believe how fast the text scroll is on the thing. Hardware, in the
form of the A2620, is nearly complete, and may be available before UNIX.
> – Is a 2000 upgradable to a 2500 ? (If so, how much ?)
A2000+A2620(2meg)+A2286+A2090A+40MegDrive+NewCustomChips+Monitor == A2500AT
A2000+A2620(4meg)+A2090A+100MegDrive+NewCustomChips+Monitor+UNIX == A2500UX
Or some such arrangement. The package deal monitor options are supposed to
be your choice of A2024 monitor (1008x800x2 monochrome) or A1950 monitor
(15kHz/31kHz "bisync" monitor, supposed to have really nice dot pitch,
etc)..
Yea…I saw that…neat, but a bit 'rich' for my blood. I wonder if
I can convince the folks at work to buy one as a workstation for me?
Yea…I saw that…neat, but a bit 'rich' for my blood. I wonder if
I can convince the folks at work to buy one as a workstation for me?
Here's a portion of a usenet msg from Hazy — thought you'd be interested.
——————————————————
> – Is the windowing shell on the A2500UX (and A3000 I would suppose)
based on > X Windows or NEWS ? Commodore would be making a BIG mistake
if it were not.
The windowing system is a custom windowing system, but more of a low-level
window manager. For the UNIX user, it give you a way to start up new
virtual terminals in windows, each with their own shell and some other
attributes, such as terminal emulations. There are also function calls for
programmable graphics. I believe that things like X or NEWS would sit on
top of this underlying window manager when ported to Amiga UNIX.
> – Do the AT and UX have 68851 chips ? I would think the UX would have
to.
Yup. Current plans, as best they've told me, say that all A2620 boards
will have 68851s on them.
> – Will UNIX be available for the 2000 with the x020 board since it has a
> 68851 ? > – Was any hardware shown or is all of this VAPORWARE for the
moment ?
We showed the Amiga UNIX, a real AT&T V.3, running on an A2000 and A2620
board at the show. Even let hackers do stuff on it. Runs like a champ. I
really don't know how close the UNIX is to being complete, but it does work
real nicely now, you can edit and compile on it, its fast, etc. You really
won't believe how fast the text scroll is on the thing. Hardware, in the
form of the A2620, is nearly complete, and may be available before UNIX.
> – Is a 2000 upgradable to a 2500 ? (If so, how much ?)
A2000+A2620(2meg)+A2286+A2090A+40MegDrive+NewCustomChips+Monitor == A2500AT
A2000+A2620(4meg)+A2090A+100MegDrive+NewCustomChips+Monitor+UNIX == A2500UX
Or some such arrangement. The package deal monitor options are supposed to
be your choice of A2024 monitor (1008x800x2 monochrome) or A1950 monitor
(15kHz/31kHz "bisync" monitor, supposed to have really nice dot pitch,
etc)..
If you need a 50 Meg hard disk and 2 Meg to run UNIX then someone's screwed
up. We were running it on a PDP-11/44 with 256 Kwords (512KB) and 2 10-meg
drives. Worked like a champ. Ah, the good old days before AT&T decided to
make UNIX "commercial" and knoecked it out of range of personal computers.
Shessh, I doubt the current unix distribution would even fit on a 50 meg
disk, although a usable system could do it. I wouldn't want to run unix
with less than a 50 meg disk for writing software, that's fer sure. sdb
You guys are spoiled. In my day we ran a multitasking multiuser operating
system in 4K of ROM and 12K of RAM. Phphphphpht…
Was that the 1950's or 60's? (grin)
The '70s, actually. In Forth on a Cosmac 1802 (a 4-bit micro).
Cosmac? You mean someone else remembers the Cosmac? I once wrote a Cosmac
assembler. In compiled CBASIC. On an IMSAI 8080 under CPM. It was very
painful. Let's hear it for "Standard Call and Return"…
When I was going to college, starting in 1973, I was in the first group in
the MIS concentration that did not learn how to set up wired control
panels, though we did learn how to use them! Of course, I' ve worked very
hard to forget!
When I was going to college, starting in 1973, I was in the first group in
the MIS concentration that did not learn how to set up wired control
panels, though we did learn how to use them! Of course, I' ve worked very
hard to forget!
Tha Cosmac Elf! My first 'kit form computer', right after my homebrew
Signetics 2650 system. Needless to say, I was spoiled by the architecture
of the 2650, and found the Cosmac "interesting".
Tha Cosmac Elf! My first 'kit form computer', right after my homebrew
Signetics 2650 system. Needless to say, I was spoiled by the architecture
of the 2650, and found the Cosmac "interesting".
To hell with SCRT. Have you looked at Forth? The 1802 is the only processor I
know of for which the Forth inner interpreter is *faster* than a subroutine
call!
As I recall, SCRT involved…
[about 8 instructions deleted because I can't recall them]
The Forth NEXT involved…
LDA IP; PLO WP; LDA IP; PHI WP; LDA WP; PLO PC; LDA WP; PHI PC; SEP PC
No, SCRT was more than that. As I recall it did something like:
GLO R6; STXD; GHI R6; STXD; GLO R3; PLO R6; GHI R3; PHI R6 LDA R6; PLO R3;
LDA R6; PHI R3; SEP 3
That's right. 4 more instructions to implement subroutine threaded code
than to implement ITC. DTC was even quicker. 5 instructions.
Just so you know, Multi-Forth on the Amiga uses a two-instruction NEXT. 22
clock cycles compared to a BSR/RTS's 34. The two instructions are MOVE.W
(A2)+,D3 and JMP d(A6,D3.W).
To hell with SCRT. Have you looked at Forth? The 1802 is the only processor I
know of for which the Forth inner interpreter is *faster* than a subroutine
call!
As I recall, SCRT involved…
[about 8 instructions deleted because I can't recall them]
The Forth NEXT involved…
LDA IP; PLO WP; LDA IP; PHI WP; LDA WP; PLO PC; LDA WP; PHI PC; SEP PC
No, SCRT was more than that. As I recall it did something like:
GLO R6; STXD; GHI R6; STXD; GLO R3; PLO R6; GHI R3; PHI R6 LDA R6; PLO R3;
LDA R6; PHI R3; SEP 3
That's right. 4 more instructions to implement subroutine threaded code
than to implement ITC. DTC was even quicker. 5 instructions.
The '70s, actually. In Forth on a Cosmac 1802 (a 4-bit micro).
phphphphpht???? Is that the sound of the hard disk at every memory
access???? (grin) sdb
phphphphpht???? Is that the sound of the hard disk at every memory
access???? (grin) sdb
You guys are spoiled. In my day we ran a multitasking multiuser operating
system in 4K of ROM and 12K of RAM. Phphphphpht…
Shessh, I doubt the current unix distribution would even fit on a 50 meg
disk, although a usable system could do it. I wouldn't want to run unix
with less than a 50 meg disk for writing software, that's fer sure. sdb
I'm speaking of a 'modern' implementation of UNIX. I realize that
older versions ran in a smaller amount of core and needed less space on the
HD for programs and swapping.
I'm thinking more in the line of a rather minimal version such as the
7300 version running in 1 Meg of core and with a 20 Meg HD…that works, but
not very well. Move it up to 2 Meg and give it a 40 Meg HD and you've at
least got a workable system (if you stay away from the windows–Ugh!!).
> I'm speaking of a 'modern' implementation of UNIX.
System V and BSD are not "Modern Implementations of UNIX". They're much
hacked descendents of Version 6 and Version 7 respectively. Modern
implementations of UNIX include such systems as Coherent, OS/9-68K, and so
on. There's no reason other than laziness on AT&T's part for UNIX not
being as small as AmigaDOS. They haven't even bothered to make all the
system programs call perror()!
I can only presume there was some reason for the drop from V6 or V7
back to V5. Considering what it does, a 400K kernel doesn't seem excessive
to me (under SVR3.1.1 on a 3B2-600).
What drop from V6 or V7 back to V5?
And System V really doesn't do much more than the Amiga Exec. The main
difference is the resource management, and V7 handles that… in under 64K.
They really need to drop back to square one and reimplement it from
scratch. I'm sure they could do everything SV does in a fraction of the
space. It doesn't even do windows.
Humm…I'm using SVR3 and I've got windows via layers. Not sure
what you're using. I also forget my UNIX(tm) history exactly, but seem to
recall that when AT&T started to let Universities license it…then System
6 and System 7 got spawned by the 'hackers' so to get it back under
control, AT&T decided that System V was the only 'official' version.
Extensions that were added to 6 and 7 were not necessairly included in
SV…thus a step backwards.
I've got SVR3, but I don't have a Blit. All the windowing smarts are in the
terminal, so that's cheating. I can use an Amiga for a terminal as well
:->.
There are two numbering schemes, at least.
The internal numbering was in editions. You had 4th Edition, 5th Edition
(the first one I saw… was still mostly in Assembly language), 6th
Edition, and so on. The latest Bell Labs version is 9th edition.
When the released a version to the Universities, they changed it from an
Edition to a Version. There were three of these… Version 5, Version 6,
and Version 7.
Around the time the 7th edition was being developed, a new group was being
formed: the Unix Support Group. Their first product was called Programmer's
Work Bench, and was based on Version 6. Later on this was renamed System I.
After the breakup, USG took the latest version of PWB — System III — and
released it as a product. Bell Labs never released 8th Edition. System III
was PWB with some Version 7 additions. System IV never saw the light of
day. System V was their next release. At this point they stopped
increasing the numbers, but if they had stuck with their numbering scheme
SVR3 would have been System VII.
I still find it hard to believe that they've never stepped back and done a
real redesign. They haven't even updated all the utilities to use perror(),
for gawd's sake.
Geeze…when did you get involved with UNIX? According to Ritchie
and Thompson (Bell System Technical Journal v57 #6) they rewrote the OS in
C in 1973 for the most part, with very little assembly left.
Geeze…when did you get involved with UNIX? According to Ritchie
and Thompson (Bell System Technical Journal v57 #6) they rewrote the OS in
C in 1973 for the most part, with very little assembly left.
I've got SVR3, but I don't have a Blit. All the windowing smarts are in the
terminal, so that's cheating. I can use an Amiga for a terminal as well
:->.
There are two numbering schemes, at least.
The internal numbering was in editions. You had 4th Edition, 5th Edition
(the first one I saw… was still mostly in Assembly language), 6th
Edition, and so on. The latest Bell Labs version is 9th edition.
When the released a version to the Universities, they changed it from an
Edition to a Version. There were three of these… Version 5, Version 6,
and Version 7.
Around the time the 7th edition was being developed, a new group was being
formed: the Unix Support Group. Their first product was called Programmer's
Work Bench, and was based on Version 6. Later on this was renamed System I.
After the breakup, USG took the latest version of PWB — System III — and
released it as a product. Bell Labs never released 8th Edition. System III
was PWB with some Version 7 additions. System IV never saw the light of
day. System V was their next release. At this point they stopped
increasing the numbers, but if they had stuck with their numbering scheme
SVR3 would have been System VII.
I still find it hard to believe that they've never stepped back and done a
real redesign. They haven't even updated all the utilities to use perror(),
for gawd's sake.
you forgot system 3 after version 7 came system 3 which at&t swore
blind would be the 'standard' eunuchs, of course a year or two later came
system 5.
I'm at work now…and have the books with the 'history' in them, and
Peter's basicaly correct.
I'm at work now…and have the books with the 'history' in them, and
Peter's basicaly correct.
you forgot system 3 after version 7 came system 3 which at&t swore
blind would be the 'standard' eunuchs, of course a year or two later came
system 5.
Humm…I'm using SVR3 and I've got windows via layers. Not sure
what you're using. I also forget my UNIX(tm) history exactly, but seem to
recall that when AT&T started to let Universities license it…then System
6 and System 7 got spawned by the 'hackers' so to get it back under
control, AT&T decided that System V was the only 'official' version.
Extensions that were added to 6 and 7 were not necessairly included in
SV…thus a step backwards.
What drop from V6 or V7 back to V5?
And System V really doesn't do much more than the Amiga Exec. The main
difference is the resource management, and V7 handles that… in under 64K.
They really need to drop back to square one and reimplement it from
scratch. I'm sure they could do everything SV does in a fraction of the
space. It doesn't even do windows.
I can only presume there was some reason for the drop from V6 or V7
back to V5. Considering what it does, a 400K kernel doesn't seem excessive
to me (under SVR3.1.1 on a 3B2-600).
> I'm speaking of a 'modern' implementation of UNIX.
System V and BSD are not "Modern Implementations of UNIX". They're much
hacked descendents of Version 6 and Version 7 respectively. Modern
implementations of UNIX include such systems as Coherent, OS/9-68K, and so
on. There's no reason other than laziness on AT&T's part for UNIX not
being as small as AmigaDOS. They haven't even bothered to make all the
system programs call perror()!
I'm speaking of a 'modern' implementation of UNIX. I realize that
older versions ran in a smaller amount of core and needed less space on the
HD for programs and swapping.
I'm thinking more in the line of a rather minimal version such as the
7300 version running in 1 Meg of core and with a 20 Meg HD…that works, but
not very well. Move it up to 2 Meg and give it a 40 Meg HD and you've at
least got a workable system (if you stay away from the windows–Ugh!!).
If you need a 50 Meg hard disk and 2 Meg to run UNIX then someone's screwed
up. We were running it on a PDP-11/44 with 256 Kwords (512KB) and 2 10-meg
drives. Worked like a champ. Ah, the good old days before AT&T decided to
make UNIX "commercial" and knoecked it out of range of personal computers.
I have no idea on the UNIX, the A2620 was to have come in at under
$2000 with 2 Meg…so a guess would be in the $3000 for 4 Meg. A 100 Meg
SCSI drive probably goes for about $1500, the monitor for about $700 so
that'd leave about $300 for the cost of UNIX.
THAT IS A GUESS…NOT CBM figures!!
Not having seen this setup…I can only guess from our setups at work
that it would take a minimum of 50Meg HD and 2 Meg of RAM to even begin to
work, but it'd probably be slow. If it's not a complete full implementation
(similar to what's on the AT&T7300), then you could get away with 1 Meg and
20 Meg HD…but that is a very minimal system…very minimal.
Don, approximatly how much would it cost to add a A2620, 100 meg SCSI drive
and UNIX to a 2000? How much is UNIX by itself and what is the minimum
ram and HD size needed to run UNIX effectively?