Here is a summary of the net's responses to my questions about the
Weitek SPARC POWER uP. Responses from this list have only started to
arrive, but they are surprisingly positive -- I did not expect this chip
to boost Lisp performance as much as other programs, but apparently it
does!
Thanks to all for the help.
-- Chuck
Date: Fri, 13 May 1994 14:51:59 -0700
>From: "Greg R. Roelofs" <nas.nasa.gov at roelofs>
We bought one and put it in an SS2; as far as I know, no one but
Weitek is selling them because they can't meet demand (at least,
that was true last summer or fall).
CPU-intensive things which fit in the cache (i.e., compression/
decompression programs) are twice as fast (comparable to SS10/30);
big CPU-intensive things vary widely, but 50% faster was a rough
ballpark figure. I/O-intensive things don't show any real improve-
ment. It's about what you'd expect...well worth the price of ad-
mission.
Greg
Date: Fri, 13 May 1994 17:06:04 +0800
>From: <Eng.Sun.COM at Peter.Luk> (Peter Luk)
You should be ok running it on SunOS 4.x, Solaris 2.4 uncovers a Weitek
hardware bug, works fine on 2.3 though as far as I know.
-Pete
Date: Fri, 13 May 94 22:58:21 -0400
>From: Carl Mueller <cs.unc.edu at mueller>
In article <ptolemy-ethernet.arc.nasa.gov at 1994May13.211755.19720> you write:
>We're considering buying several Weitek SPARC PowerUP accelerator chips.
>What's the net.consensus on these chips? Are they as effective in real
>life as advertised? Are they completely compatible or are there
>occasional quirks?
Many people have asked this. The responses have all been very positive.
No quirks were reported. Performance increase was in the 1.5-2.0 range.
-Carl <cs.unc.edu) at (mueller>
Date: Sat, 14 May 94 08:50:44 PDT
>From: <zcon.com at szh> (Syed Zaeem Hosain)
They work like a charm. I get an effective speed-up that is much better
than a SS 10/20 and almost, but not quite as good as a SS 10/30.
Averages about 1.4 for most users and applications including the
effects of disk drive, ram, etc. (since is only the cpu that changes,
the other effects can impact overall improvement).
This is measured results, not just empirical observation.
>For the record, our applications use SPARCstation 2s and IPXs under
>SunOS 4.1.2 and 4.1.3, running Allegro Common Lisp.
I have sent out copies of a paper I wrote on what speed up I achieved
building the gcc compiler. A copy is available at ftp.uci.agh.edu.pl as
/pub/docs/weitek-powerup.ps.gz for European users - others can get one
>from me (uuencoded e-mail) if you ask for it. This is a PostScript
document, so be prepared to print it on a PostScript device or view
it using OpenWindows "pageview".
To be honest, I do not know what effect the chip will have on Allegro
Common Lisp. But it should be consistent with what others have achieved.
Z
P.S. I am posting this from an IPX equipped with the Weitek Power uP.
-------------------------------------------------------------------------
| Syed Zaeem Hosain P. O. Box 610097 (408) 441-7021 |
| Z Consulting Group San Jose, CA 95161 <zcon.com at szh> |
-------------------------------------------------------------------------
Date: Sat, 14 May 94 17:04:57 CST
>From: <maroon.tc.umn.edu" at "hartkda1> <maroon.tc.umn.edu at hartkda1>
To: <freud.arc.nasa.gov at chucko>
Subject: Weitek PowerUP
In about Feb. we needed to add two stations and could not wait for the new
boxes to come out. We added 2 SPARCII's with the Weitek chip in place.
Our package included 20in CRT, 515mb drive and 64mb memory. They were
loaded with SOLARIS 2.3.
They were added to a network of IPCs, SSIIs, S10/30s, and PCs. On these
boxes the most used applications are AutoCAD and a large third party
add-on. To date, we have had less problems with these two systems than the
others. Some may be because our best users are on these two boxes.
The comments on performace include; "I do not want to go back." "It took
37 hours on a DELL 486DX2-66 to do a smaller job than this 1.5 hour job
(heavy contouring)." "Should we buy 5's and 20's when these work and are
cheaper."
Date: Sat, 14 May 1994 21:24:55 -0400
>From: <turing.acs.ryerson.ca at dburke> (Darryl Burke - ACPS)
I've been using onw of thier "PowerUP" chips for about 4 weeks now, The
only "REAL" difference i noticed was during "heavy" CPU use, during
general "X" there wasd very little difference.. I installed the chip
into an IPX with 16 megs Memory. No probs with compatability. I would
only purchase one if you plan on running some "CPU intensitive "
programs
Darryl Burke
<turing.acs.ryerson.ca at dburke>
Date: Mon, 16 May 94 09:09:49 MDT
>From: <Ceram.COM at Craig.Warner> (craigw)
Chuck -
We use the Weitek PowerUp in house for our design group,
and have sold ~200 units to our customers.
We have had one damaged chip, which caused spurious errors. This
was replaced, and the customer purchased 9 more.
We have also had several chips which had a design/manufacturing flaw, which
has been correctly. Weitek replaced these chips under the warranty,
and they have worked fine since then. This is referred to as the
sendsig problem, which caused certain processes to die. This problem
has been fixed, and does not occur with the latest rev of the chips.
No customers have reported software incompatibility issues, and we
have not seen any in house.
We sell the chips through our email marketplace. Current pricing
is $1022 CPU Only, Net 30 Day Terms
$997 CPU Only, COD or prepay
$175 Optional install kit purchase.
We will loan an install kit at no charge for people who do not
wish to purchase them.
Regards,
Craig Warner
>From: Lee Ethetton <sugar.NeoSoft.COM at ethetton>
Subject: Re: Weitek SPARC PowerUP experiences?
Hi Chuck
> List price with installation kit is $1500, in single units. Ceram's
> $997 price is COD or prepaid only, without the installation kit.
> -- Chuck
>
I got a call from my Weitek "handler" today and the list price without the
install kit is going down to $995 and with the kit to $1,150. The new
5's must be putting a lot of pressure on them.
-- Lee
Date: Tue, 17 May 94 12:45:18 PDT
>From: <giants.mce.com at chris> (Chris Patterson)
Hello Mr. Wait,
We have the Weitek "Power-Up" chips in stock in large quantities.
Price is $1,150. If you have any interest I can be reached at:
Chris Patterson
Minicomputer Exchange, Inc.
610 N. Pastoria Avenue
Sunnyvale, CA 94086
(408) 733-4400 Phone
(408) 733-8009 Fax
<mce.com at chris> E-mail
We are one of the world's largest re-sellers of Sun Microsystems and Silicon Graphics equipment.
Would you have any interest in our latest price list?
Sincerely,
Chris Patterson
Date: Wed, 18 May 94 14:56:07 EST
>From: Andrew McRae <mega.com.au at andrew>
We had some of these chips fairly early (August/Sept 93), and
had some problems running them in IPXs and older SPARCstation 2s.
Weitek *very* promptly responded and said some of the older
motherboards had some suspect memory timings that were tripping
some problems in the PowerUP chip. A new production run had some
tighter tolerances enforced that solved the problem.
Since having the chips replaced, I can categorically state that
the Weitek chip has performed flawlessly, and have nothing
but high praise for the product, packaging and support.
> For the record, our applications use SPARCstation 2s and IPXs under
> SunOS 4.1.2 and 4.1.3, running Allegro Common Lisp.
I think you will see a major improvement, especially running that
application. We can't fault them in terms of compatibility.
I would make sure that you
do get some samples and try them out thoroughly in case your
particular machines suffer from the problems I mentioned.
> -- Chuck
Cheers,
Andrew McRae inet: <mega.com.au at andrew>
Megadata Pty Ltd, uucp: ..!uunet!mega.com.au!andrew
North Ryde 2113 Phone: +61 2 805 0899
NSW AUSTRALIA Fax: +61 2 887 4847
Date: Wed, 18 May 1994 10:37:41 +0000
>From: Pete Clinch <dundee.ac.uk at P.J.CLINCH>
In article <ptolemy-ethernet.arc.nasa.gov at 1994May13.211755.19720> you wrote:
: We're considering buying several Weitek SPARC PowerUP accelerator chips.
: What's the net.consensus on these chips? Are they as effective in real
: life as advertised? Are they completely compatible or are there
: occasional quirks?
We have one in a SPARC 2. Can't give you reliable performance gain
features with decimal points, but it's certainly quicker, the guy using
it hasn't complained and there have been no anomalies on the machine at
all since we installed it. Installation was painless as well.
Pete.
--
Peter Clinch University of Dundee
voice 44 382 60111 x 2050 Department of Medical Physics
fax 44 382 640177 Ninewells Hospital
email <dux.dundee.ac.uk at pjclinch> Dundee, DD1 9SY, Scotland, UK
>From: <elam.mdl.sandia.gov at aflundi>
Date: Wed, 18 May 1994 10:52:03 -0600
We replaced the CPUs in 10 IPX here and I know of
another group that replaced them in 15 SS2s. We
don't run Allegro Common Lisp, but the "upgrade"
has been 100% transparent (except for speed) over
the last ~year. In short, for us it has been
well worth it.
--alan
--
Alan Lundin <sandia.gov at aflundi>
Date: Wed, 18 May 1994 16:14:39 -0700
>From: <nas.nasa.gov at roelofs> (Greg R. Roelofs)
Whoever asked about these things a couple of weeks ago: I just got an
announcement from Weitek that the prices have dropped. They were $1500;
now they're $1200 for the full kit (includes special tool for extraction/
insertion) or $999 for just the chip. There are also volume discounts
of some sort available. These work in Sparc 2's and IPX's (not IPC's).
Contact numbers:
800 758-7000 (sales) 800 827-8708 (fax)
408 738-8400 (sales) 408 522-7525 (fax)
<weitek.com at cpu_upgrade>
Greg
Date: Tue, 24 May 94 16:15:54 EDT
>From: Michael Lucks <stsci.edu at lucks>
Chuck,
We tried a PowerUp earlier this year and were sufficiently pleased with the results
that we purchased a bunch more of them. Here is a summary of some (application
specific) performance tests I ran.
Date: Sat, 8 Jan 94 19:07:19 EST
Subject: Sparc accelerator works
Reply-To: <stsci.edu at lucks>
A Weitek POWER uP chip was installed yesterday on my Sparc 2 and it
appears to work as advertised. On my Spike benchmark (which loads 326
proposals and propagates constraints), it runs only about 3% slower
than a Sparc 10 in CPU activity. Overall (CPU + disk I/O), it runs
about 17% slower.
By comparison, a Sparc 2 runs about 30% slower than a Sparc 10 for CPU
activity, and 44% slower overall.
Here are the stats.
Machine Type CPU non-CPU Real
(mins) (mins) (mins)
loki SS10 53 8 61
hoku SS2/powerUp 55 18 73
mingus SS2 76 32 108
It looks like the chip turns a Sparc 2 into near CPU-equal of a Sparc
10 -- and I/O performance is about midway between the SS2 and SS10.
A good deal for $1,500.
-- Mike
Date: Tue, 24 May 94 16:24:08 EDT
>From: Glenn Miller <stsci.edu at miller>
The work like a charm with no problems. 100% compatible.
We have about a dozen Sparc 2s and IPXs with the chip and run some
heavy duty lisp applications with clim. We find them to be about 50%
faster. all in all well worth the $
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Glenn E. Miller <stsci.edu at miller>
Advance Planning Systems Branch 410-338-4738
Space Telescope Science Institute 410-338-1592 (fax)
3700 San Martin Dr.
Baltimore, MD 21218 USA
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Date: Tue, 24 May 94 13:20:39 PDT
>From: <vitsemi.com at chess> (Julie Reilly)
To: <freud.arc.nasa.gov at chucko>
Subject: Re: Weitek SPARC PowerUP experiences?
Weitek let us evaluate one for two weeks. We put it in a SPARC IPX
and saw significant speed-up in some CPU-intensive digital simulations.
Afterwards, we bought some. Have been running Sun OS 4.1.1, X11R4 and/or
Openwindows, and a couple of different CAD frameworks for a month now
with no problems.
>From: Daniel Cerys <BBN.COM at cerys>
Date: Tue, 24 May 94 16:45:49 EDT
We've been using the Weitek on 3 local IPX's for development, and
recommended their use to a number of our clients. You can expect a speed
increase of roughly 2X. Totally transparent; a lot cheaper than buying
Sparc10 upgrades.
Dan