riscy
[Top] [All Lists]

Re: What do we *really* want ?

To: riscy@pyramid.com
Subject: Re: What do we *really* want ?
From: Andreas Busse <andy@piggy.waldorf-gmbh.de>
Date: Sun, 4 Jul 93 15:55:09 GMT
Cc: andy@piggy.waldorf-gmbh.de
Reply-to: riscy@pyramid.com
Sender: riscy-request@pyramid.com
Hi Mipsers,

I've got some response (fortunally without flames and bombes)
to my "downgraded spec sheet". Thanks.


General
-------

> > Now we're on the way to build a complete workstation
> > with lots of features, on-board intelligent video, scsi,
> > ethernet, serial, sound and whatever.
> > That all sounds good, but I doubt that all this is really necessary.
> > 
> But by the time you reach the end of your list you've not only
> specified a complete workstation, you're calling it that yourself.

See keyboard controller.

CPU
---

> The only information I have on the R4200 is the announcement from
> Byte Magazine that was shared, and that sounded like complete
> Vaporware (tm) to me.  Impossible power specs and a promised
> delivery no sooner than six months in the future?  I wouldn't hold
> my breath waiting for that one.  On the other hand, I _would_ hold
> my breath waiting for the IDT3730.  That one would reduce the parts
> count and increase the probability of getting it working.

I'll try to find out what's about the R4200 and let you know.


RAM
---

> --- You wrote:
> * Ram. The more, the better. Someone suggested to use
>   those 36bit by 512k or 2M modules. That saves boardspace,
>   right, but they are nearly twice that expensive than four of
>   the good old x9 modules.
>   I would say, 64 Meg on board is *definitly* enough.
> --- end of quoted material ---
> I can get 60 ns 8MB 72-pin SIMMs for $260.  If you can find 30-pin 60 ns 16MB
> SIMMs for $260, that's great...  Or 4MB 60 ns SIMMs for $65.

What quantities ? What manufacturer ?

> I don't like to give as much board real estate to this as Neill wants,
> but you definitely need enough sockets to get the interleave that Neill
> wants.  32M (8 pieces, 4M X 9 bits) would probably satisfy me, so as
> long as I can leave some sockets empty I won't complain.

I totally agree.


Serial I/O
----------

> > * Serial I/O. For starting the Linux port, we will need
> >   two ports; one for a dumb ascii terminal, one for downloading
> >   code.

> I am not objecting to 2 serial ports in general, but there is no
> reason at all to do code downloads on a different serial line.

Of course, you might use the same port for debugging and download.
But it's much more convenient to have independent ports, right ?

> The boot rom should boot the processor and go into an endless loop
> providing a prompt, and reading commands. The three essential
> commands are "peek %d", "poke %d %d" and "go %d". 
> Downloading code goes through the poke command.

Yes, that might work, but it's slow. I would prefer S-record
download. Just hit "load" on the debugging terminal, and then
download the stuff over the other line.

> >   Two ports are still sufficient for a workstation:
> >   One for the mouse, the other one for the modem.
> >   A third port is an *option*. It could be used for a printer
> >   in case we don't have a parallel port.

> I think that you are right here: two ports is not a bad investment 
> after all. The first goes to a mouse, the second is quickly used, and
> more are optional. However if a 4 port chip is available, I wouldn't
> object .....

Is there a four port chip ?

> Agree, but I'd really be disappointed if I couldn't make one of
> those ports run at least 38400Kbits/sec.  My primary network
> connection would be a V.32bis/V.42bis modem, as this thing would
> reside in my spare bedroom.

Of course, 38.4k is *definitly* a must.


keyboard controller
-------------------

> > * Keyboard controller. This is a *must*.
> Disagree, if it's a card in the pc then hook up a cable from pc's serial
> port to card's serial port and voila you've got a terminal.

You got me a bit wrong. I wasn't happy with the pc card solution.
A workstation is much more what I want, but we should not try
to make a high-end thing.


SCSI / IDE
-----------

> > * SCSI. This is a *must*.
> >   I can't imagine a workstation without SCSI...

> Here is where you start calling it a workstation.  Actually, I
> agree.  SCSI is the _most_ important I/O to me.

Either it's a pc card or it's a workstation. Is there something
in between ? I don't like pc cards, so it must be a workstation :-)

> If you're buying a new disk drive, the difference is negligible.  I
> don't have any IDE disk drives lying around, maybe others do.  If
> you're buying a tape drive or a CD-ROM drive SCSI is certainly
> preferable, and I don't think we have room for the connectors for
> both.

I agree. Even my Linux box at home has never seen any IDE stuff.

> I can't understand why SCSI is preferred to IDE, at least not very
> definitely!  I know that one can connect a lot of different stuffs
> on the SCSI bus, but other than hard disk and tape drive, what are
> the items that will be of interest to majority people (or customers)?
> I'm now owning a total of 500MB IDE hard disk space, and I will be
> very happy if I can use it on the MIPS system.

Simple. There's much more than disk and tape drives you can
connect to SCSI. There are scanners, terminal servers, FPGA-
programmers, dsp cards and whatever. SCSI is much more than
just a mass storage interface. IDE is cheaper for the first
disk, perhaps still for the second one but finally SCSI is the
winner. 


ETHERNET
--------

> > * Ethernet. This is a *must*. We can't call the thing
> >   "workstation" if it doesn't have an ethernet port.

> I'd like to have it, but I could live without.  As I said, my
> primary network connection would be a modem.

My idea behind on-board ethernet is not to have because it's
essentially necessary but because it would cost around $30
on-board but at least $150 off-board in form of an ISA-card.
I would *definitly* have to buy another ethernet adapter.

video
-----

> > 
> > * Video on board. Don't laugh, but this is an *option*.

> OK I did laugh, you seem to contradict yourself: you regard SCSI&Ethernet as
> an important thing and video not? Nowadays my picture of a workstation is
> running xwindows with a big monitor  (Well at *LEAST* 1024*768 >70Hz) It
> seems that video keeps this group active. :-)

I have the same idea of a workstation. No video on board doesn't
mean no video at all.

> Decent performance is the issue, standard off the shelf (==cheap vga) 
> doesn't stack up in performance against workstation performance, and
> I'm certainly not comparing against top of the line!

Ok, you're right.

> > Yes, I know, it would be slower. The ray tracing guy
> > (sorry, I've forgotten your name) will kill me, but:
> > Please, go ahead and buy a SGI Indigo. We will never ever
> > reach the graphic performance of such a machine, not
> > with a 75MHz R4400 and not with tons of off-the-shelf
> > pseudo-intelligent video-controller stuff.
> >
> This is not what we (I guess) want: we want a reasonable performance for
> a reasonable price.

This is exactly what I want too.

> Well video is still important to me.

It's important to me too. I just have tried to find a solution
with low overhead and small design tasks.


> Fast graphics is a must for me. It is one of the reasons why I'd buy
> such a thing. I think that an onboard dumb frame buffer is the best
> solution. Using a conventional ISA VGA card will be slower (probably
> even with accelerator), and VESA local bus or some other interface for
> high performance graphics cards will be difficult. I think we should
> put our brains into a clean design, not into workarounds for intel
> braindamage.

> A dumb frame buffer on board, 8 bits per pixel, with no multiple
> scan frequencies (pick one when you buy the crystal).  That's high
> enough on enough people's wish lists to justify.  The parts count
> and board space don't seem too much to me.

Ok, looks like we come closer to a decision. I don't mind if
there's an on-board frame buffer. But it should be dumb, without
any "intelligence". What's about 8bit 768x1024 non-interlaced, 70 Hz ?
Most mid-class monitors can handle that. Please remember: Not
everybody has a $2000 Eizo Multisync !


SOUND
-----

> > * Sound I/O. This is an *option*.

> I don't care at all.  I wouldn't give much board space to it.

Ok. I wouldn't mind even if there's no beep !
Sound skipped ?


ISA BUS
-------

> > * ISA Bus... I would say: we *do* need exactly one slot.
> >   Every slot more is an *option*. Why ?

> I think we need some kind of I/O bus. ISA is nice for people who
> already have ISA cards or prefer to buy hardware instead of designing
> it. A simple proprietary bus may be better for the hardware gurus (and
> chaeper). I am not sure what I would prefer, but I am leaning more
> towards the ISA bus. I would keep the number of slots small, however
> (max. 3). 

A proprietary bus is *definitly* no solution.
If we need any kind of expansion bus, ISA is a good choice.

> The no DMA, no bus master, no 8/16 bit detection are good ideas
> worth discussing.  The original PC/XT bus was simple enough you
> could probably implement it with a couple of 22V10's and the
> IDT3730.  In that case, it would make more sense than an ad hoc
> expansion bus that has no market.
> Of course, I'm only guessing about the capabilities of the IDT3730.
> Sure would like to have a data book for that!

If we have all the high-bandwith stuff on board, eg. SCSI and
Ethernet, there's in fact no need for bus-master and dma stuff. 
What could be still needed ? If video is on the motherboard (seems that
most people want that), one might add: more serial lines,
a soundcard, a midi i/o card and the like. Nothing with high bandwith.
(I'm not sure wether soundcards need DMA or not).

I'll try get a 3730 databook.


Again, I'll summarize:

* 40 MHz 3051E/3081E, perhaps R4K2.
* 16 standard SIMM slots for memory sizes
  from 8 Meg (with 1Mx9 modules) to 64 Meg (4Mx9 modules)
  or
  8 72pin SIMM slots for memory sizes from 8 Meg (4 x 512k x 36bit)
  to 64 Meg (8 x 2M x 36bit).
* 2 serial ports.
* One parallel port or one extra serial port.
* SCSI controller
* Ethernet controller
* Keyboard controller
* Realtime clock
* on board 8bit color video, at least 1024x768, 70 Hz.
* 2 or 3 ISA slots, without DMA and busmaster stuff.

Cheers,
Andy



-------------------------------------------------------------------------------
Andreas Busse             | e-mail: andy@piggy.waldorf-gmbh.de (home)
Eislebenstrasse 5         |         andy@resi.waldorf-gmbh.de (office)
D-5300 Bonn 1             | Phone:  +49 (0)228-252687
-------------------------------------------------------------------------------


 

<Prev in Thread] Current Thread [Next in Thread>