[comp.parallel] Need vision accellerator

wross@ius3.ius.cs.cmu.edu (Bill Ross) (01/29/91)

    We run a variety of computer vision tasks (fft's, neural nets,
    and other typical vision operations) on our Sun Sparcstations.
    In many cases, we find that the Sparcs are too slow to do what
    we want to get done.  In particular, the floating point is often
    inadequate.  It seems that there should be a spiffy S-bus card
    that can solve all our problems.  Does anyone have any suggestions?

    Here are some very rough guidelines:

	Easily programmable (transparent even?)
	Helpful for almost any vision task
	Fast I/O so that even simple (short) jobs can benefit
	Fits S-bus in Sun Sparcstation
	Cheap enough that I can put one in each of our machines

    The idea isn't to make any any single problem run super-fast,
    but to make it simple to speed up whatever problem you are
    working on (with minimum inconvenience in programmimg).  Am I
    correct that most of the cheaper array processors, parallel
    machines and DSP engines are likely to be too limited in
    application?  Is there any way to speed up the Sparc floating
    point performance by adding a piggy-back board with a big
    Weitek chip or something on it?

    Thanks for any and all ideas!

    Bill Ross
    wross@cs.cmu.edu

raja@bombay.cps.msu.edu (Narayan S. Raja) (01/31/91)

In article <12811@hubcap.clemson.edu>, (Bill Ross) writes:
 
<     We run a variety of computer vision tasks (fft's, neural nets,
<     and other typical vision operations) on our Sun Sparcstations.
<     In many cases, we find that the Sparcs are too slow to do what
<     we want to get done.  In particular, the floating point is often
<     inadequate.  It seems that there should be a spiffy S-bus card
<     that can solve all our problems.  Does anyone have any suggestions?


About other add-ons I have no idea, but Sun
itself apparently has an add-on called "taac".
Output of "man taac" is appended.

**Pl. summarize responses** to this newsgroup.

Thanks in advance,


Narayan Sriranga Raja.


------------------------------------------------------------------
TAAC(4S)         DEVICES AND NETWORK INTERFACES          TAAC(4S)

NAME
     taac - Sun applications accelerator

DESCRIPTION
     The taac interface supports the optional TAAC-1 Applications
     Accelerator.   This  add-on  device  is  composed of a very-
     long-instruction-word computation engine,  coupled  with  an
     8MB  memory  array.  This memory area can be used as a frame
     buffer or as storage for large data sets.
------------------------------------------------------------------

surfer@Erc.MsState.Edu (John West) (02/07/91)

If you are doing lots of FFT's and other image processing operations,
you may be interested in looking at some of the third party add on
external boxes for Sun workstations.  There are several which have
been around for a while.

For on-board graphics solutions, Sun recently introduced two new
graphics accelerators when they introduced the SPARCstation 2.  They
are the GS and the GT option.  They also introduced the VX/MVX
graphics subsystem at SIGGRAPH '90.  The quickie explanation of these
products follows:

GS: 24bit framebuffer, 16bit Zbuffer, this card uses all 3 S bus slots
on a SS2. It provides some 2D and 3D graphics acceleration.

GT: 24bit (double)framebuffer, 24bit Zbuffer, this card resides in an
external chasis.  It provides impressive 3D graphics acceleration.

VX/MVX: Sun's i860 graphics subsystem! You need a SPARCstation 330 for
this option (requires VME slots).  The VX/MVX combo provides you with
a *hot* framebuffer card (VX) and a 4 PE i860 accelerator card(MVX) for your
*cpu hogging* graphics applications.

For more information, reuest the appropriate `Technical White Papers'
from you local Sun rep.  If you wish to chit-chat about these
accelerators, send me personal mail to the address shown below.

I have no affiliation with Sun Microsystems. I am a Sun (& SGI) user.

--
hangin' ten:
-John West-      MSUNSFERC4CFS          surfer@erc.msstate.edu
Engineering Research Center for Computational Field Simulation
Mississippi State University ***** National Science Foundation