eho@clarity.Princeton.EDU (Eric Ho) (09/05/89)
Having seen Symbolics' recent ad about their co-processor board for the Suns I'm wondering if anyone at Symbolics is developing similiar stuff for the pizzabox (Sparcstation-1) ? It would be nice if Symbolics can find a way to use existing memory on the Suns rather than using their own tagged memory -- it all depends on how the Sun "sees" the co-processor board and how the board is used I guess. I mean, those who are opting for the co-processor boards might already have quite a bit of memory stuffed on their desktop and it is a bad idea I guess if that can't be used by the co-processor board. Without the 10 meg of RAM on the co-processor board I think that the price for the board (& the developement system) may drop quite a bit -- I mean, it is ridiculous to move (or buy) a 3 mips deskside 3/160 and buy a $21K co-processor board into your office when you already got a less than $9K 8~16meg 13 mips pizzabox with on your desk -- it just doesn't make sense (especially when the board is more expensive than a deskside unit let alone the pizzabox). -- Eric Ho Cognitive Science Lab., Princeton University voice = 609-987-2987 email = eho@confidence.princeton.edu 609-987-2819 (messages) eho@bogey.princeton.edu regards. -eric-
mike@ists.ists.ca (Mike Clarkson) (09/06/89)
In article <EHO.89Sep4161034@cognito.Princeton.EDU> eho@clarity.Princeton.EDU (Eric Ho) writes: >Having seen Symbolics' recent ad about their co-processor board for the Suns >I'm wondering if anyone at Symbolics is developing similiar stuff for the >pizzabox (Sparcstation-1) ? No chance, as the board is a VME bus board. I doubt you could use a SCSI bus for this kind of communication between the processors. Nice idea though... Mike. Mike Clarkson mike@ists.ists.ca Institute for Space and Terrestrial Science uunet!attcan!ists!mike York University, North York, Ontario, FORTRAN - just say no. CANADA M3J 1P3 +1 (416) 736-5611
miller@CS.ROCHESTER.EDU (Brad Miller) (09/06/89)
I suggest you look more carefully at the engineering of the UXL-400 and the Sparcstation-1. 1) The UXL needs a 40 bit wide memory; it does not access memory thru the VMEbus, but from it's own board, or extends it's mem bus to an adjacent board via ribbon cable. Using host memory would cut performance significantly. If that's what you want, buy a MacIvory and a Mac-II... it uses NuBus memory to the severe detriment of processor thruput. (it slows it down to 3620 speeds; the XL-400 is supposed to be about 5x a 3620 - get benchmarks from Symbolics). 2) The Sparcstation-1 has NO bus whatsoever. Where would you plug an embedded Ivory in? The serial port? :-) If you want a 13 mips machine, get a 4/330, instead of a 3/160. Why do you need such speed out of your SUN if you will mostly use the UXL anyway? Sure, a faster host will give better X and disk response, but it isn't that big a deal... The low end of their line is the MacIvory. The high end is the UXL, which is identical boardwise to their standalone lispm. (XL-400). Comparing apples to apples, you may, as an edu institution, be able to buy a UXL-400 AND a 4/330 for less than an XL400; I know we can. And the 4/330 can be used for other stuff; the UXL-400 is accessed via X so needent be accessed thru the host at all, but, say, a remote Sparcstation. Why not put UXL-400s into the existing 3/160s and then access them from your desktop Sparcstation? Leave the 160s in the computer room.
shiffman%basselope@Sun.COM (Hank Shiffman) (09/06/89)
In article <1989Sep6.000759.15274@cs.rochester.edu> miller@CS.ROCHESTER.EDU (Brad Miller) writes: >I suggest you look more carefully at the engineering of the UXL-400 and the >Sparcstation-1. > >2) The Sparcstation-1 has NO bus whatsoever. Where would you plug an >embedded Ivory in? The serial port? :-) > No bus? This is going to come as a *big* surprise to an awful lot of people. I wonder where they've been connecting their color framebuffers... The SPARCstation 1 has a perfectly good bus. It's called the S-bus. There are three connectors inside the box for S-bus cards. These cards need to be roughly 3" by 5", so fitting a Lispm onto one would be a pretty impressive feat of engineering. -- Hank Shiffman (415) 336-4658 Marketing Technical Specialist Software Engineering Technologies ...!sun!shiffman Sun Microsystems, Inc. shiffman@Sun.com
miller@CS.ROCHESTER.EDU (Brad Miller) (09/07/89)
In article <1989Sep6.000759.15274@cs.rochester.edu> miller@CS.ROCHESTER.EDU (Brad Miller) writes: >2) The Sparcstation-1 has NO bus whatsoever. Where would you plug an >embedded Ivory in? The serial port? :-) > No bus? This is going to come as a *big* surprise to an awful lot of people. I wonder where they've been connecting their color framebuffers... The SPARCstation 1 has a perfectly good bus. It's called the S-bus. There are three connectors inside the box for S-bus cards. These cards need to be roughly 3" by 5", so fitting a Lispm onto one would be a pretty impressive feat of engineering. Yep, I misspoke on this one; I just found out there is a proprietary bus for I/O devices. (Which of course one *could* treat the ivory as, but it would lose performance wise). I'm amazed; the physical dimensions of the Sparcstation don't look like it could handle anything of the sort... On it being proprietary; has Sun published it's specs yet? Anyway, as Hank points out, obviously existing technology would make a 3"x5" card hard to do, esp. since one would either have to have local 40bit memory, or a cache to allow the local processor to get 40bits doing 32bit fetches over the S-bus. I don't think it would fit but maybe next generation :-)
barmar@think.COM (Barry Margolin) (09/07/89)
Regarding a Symbolics co-processor that fits on the Sparcstation-1's S-bus: In article <1989Sep6.205843.23261@cs.rochester.edu> miller@CS.ROCHESTER.EDU (Brad Miller) writes: >obviously existing technology would make a 3"x5" card hard to do, esp. >since one would either have to have local 40bit memory, or a cache to allow >the local processor to get 40bits doing 32bit fetches over the S-bus. I >don't think it would fit but maybe next generation :-) It shouldn't be TOO long before it's possible. That's only about a third the size of a Macintosh II NuBus card, and Symbolics has an Ivory card that fits there. And they've already dealt with the problem of accessing memory in 48-bit chunks (don't forget the ECC) over a 32-bit bus in the MacIvory. Barry Margolin Thinking Machines Corp. barmar@think.com {uunet,harvard}!think!barmar
miller@CS.ROCHESTER.EDU (Brad Miller) (09/08/89)
It shouldn't be TOO long before it's possible. That's only about a
third the size of a Macintosh II NuBus card, and Symbolics has an
Ivory card that fits there. And they've already dealt with the
problem of accessing memory in 48-bit chunks (don't forget the ECC)
over a 32-bit bus in the MacIvory.
true, but it isn't the HOST's memory, but separate memory the host doesn't
use on the NuBus... I'd think accessing host memory properly (such that the
host leaves it alone except maybe handling vmem or some such?) would require
some sort of peer processor support that I don't know if the Sbus interface
allows... they only bill it as an i/o bus. Again, I haven't seen any specs
and I've been lead to beleive they aren't published.
Definitely getting a full ivory on a 3x5 card would be an acheivement,
though!