Price/Performance info on SGI vs Competition?

Price/Performance info on SGI vs Competition?

Post by E » Mon, 12 May 1997 04:00:00



Anybody know where I could find an objective price/performance
comparison?  I'm particularly interested in how SGI stacks up against
not only traditional competition like Sun, HP. IBM, etc., but also how
they match up with PPro/Pentium II.  Any good sources for such info?

Thanks very much for any advice!

  -Ev

 
 
 

Price/Performance info on SGI vs Competition?

Post by John McCalp » Tue, 13 May 1997 04:00:00



>Anybody know where I could find an objective price/performance
>comparison?  I'm particularly interested in how SGI stacks up against
>not only traditional competition like Sun, HP. IBM, etc., but also how
>they match up with PPro/Pentium II.  Any good sources for such info?

This is going to depend a lot on what you are doing with the
machines.  Can you be more specific?

In the scientific computing side, some info is available, such as
the price/performance tables from the NAS parallel benchmarks:

http://science.nas.nasa.gov/Software/NPB/Reports/NAS-96-018.fm.html#SPLU
--
--
John D. McCalpin, Ph.D.     Supercomputing Performance Analyst
Scalable Systems Group      http://reality.sgi.com/mccalpin/


 
 
 

Price/Performance info on SGI vs Competition?

Post by WO » Tue, 13 May 1997 04:00:00


This can't possibly answer the whole question, but just
for interest, here's what I came across at the Electric
Image website (just in case you didn't know, EI is a 3D
animation software package):

Hardware System *Rendering Speed

Silicon Graphics' OCTANE MXI R10,000, Single Processor,
195 MHz Rendering Time: 2 minutes

Silicon Graphics' O2 R10,000, Single Processor,
175 MHz Rendering Time: 3 minutes

Mac Clone 604e, 400 MHz Rendering Time:
5 minutes, 45 seconds

Mac Clone 604e, 225 MHz Rendering Time:
7 minutes, 46 seconds

*Render test times conducted on project from "Buried In Time"
?1996 Presto Studios, Inc. containing over 350,000 polygons,
197 texture maps and 14 shadow casting lights.

From what I understand, the 604e is faster than the pentium
pro chip (that's just what I've heard!), so it's your guess
as to how fast the PPs would be under this very specific
rendering test.

All of this leads me to my own question (I know I hate it when
replies get off topic, but I can't help myself): I was kind of
under the impression that SGI workstations in general weren't
very fast at rendering.  What explains the above results?  Is
the 'competition' just lying?  Are all those people who say
that PPs are waaayy faster than Rx000s plain wrong?


 
 
 

Price/Performance info on SGI vs Competition?

Post by Evil » Tue, 13 May 1997 04:00:00


typedef string bigLie;

bigLie    pcMarketingMachine;

pcMarketingMachine = "fool most of the people all of the time";

Trick is to exclude yourself from the above casted type.

Josh...

 
 
 

Price/Performance info on SGI vs Competition?

Post by Jeff Harrel » Tue, 13 May 1997 04:00:00



> Silicon Graphics' OCTANE MXI R10,000, Single Processor,
> 195 MHz Rendering Time: 2 minutes

> Silicon Graphics' O2 R10,000, Single Processor,
> 175 MHz Rendering Time: 3 minutes

> Mac Clone 604e, 400 MHz Rendering Time:
> 5 minutes, 45 seconds

> Mac Clone 604e, 225 MHz Rendering Time:
> 7 minutes, 46 seconds

Far be it from me to dig on EI's renderer (it's fast; I know it's fast),
but I'm wondering why the R10000 systems aren't a LOT faster than the
604e systems.

Quote:> All of this leads me to my own question (I know I hate it when
> replies get off topic, but I can't help myself): I was kind of
> under the impression that SGI workstations in general weren't
> very fast at rendering.  What explains the above results?  Is
> the 'competition' just lying?  Are all those people who say
> that PPs are waaayy faster than Rx000s plain wrong?

I can't speak about "in general," but the R10000 is a fast CPU. Since
rendering is generally a CPU-bound task, throwing more iron at the
problem will usually solve it.

But you wanna see fast, try rendering the same type of image with Alias
PowerRender on a big multiprocessor system. Wow...

--
Jeff Harrell, Systems Engineer
Digital Media Performance Labs, Inc.
   http://www.dmpl.com
   http://jeff.dmpl.com

 
 
 

Price/Performance info on SGI vs Competition?

Post by Ben Canno » Tue, 13 May 1997 04:00:00



> All of this leads me to my own question (I know I hate it when
> replies get off topic, but I can't help myself): I was kind of
> under the impression that SGI workstations in general weren't
> very fast at rendering.

They aren't! :)  At least, their rendering speed hasn't been great.
The graphics rendering engines are just so much faster on the SGI
that the CPU rendering can seem 'slow' :)

Quote:> What explains the above results?  Is
> the 'competition' just lying?

As always.

Quote:> Are all those people who say
> that PPs are waaayy faster than Rx000s plain wrong?

Depends on the value of "x" :)  for a R2000,
a PP will beat it :)

R4x00s are about twice as fast as the P-6 given
equal clock speeds (softimage)

And R10000s are, as you can see, an order of
magnitude or so faster than a PP.

--
| The wise man can look at a Grain of Sand, |
|  and envision a Universe. But the Stupid  |
|  man just rools around on the Beach for   |
|    a while, then stands up and yells,     |
|     "Hey, look!  I'm 'Vine Man'!"      |
-Ben.

 
 
 

Price/Performance info on SGI vs Competition?

Post by Ben Canno » Tue, 13 May 1997 04:00:00




> > Silicon Graphics' OCTANE MXI R10,000, Single Processor,
> > 195 MHz Rendering Time: 2 minutes

> > Silicon Graphics' O2 R10,000, Single Processor,
> > 175 MHz Rendering Time: 3 minutes
> > Mac Clone 604e, 225 MHz Rendering Time:
> > 7 minutes, 46 seconds

> Far be it from me to dig on EI's renderer (it's fast; I know it's fast),
> but I'm wondering why the R10000 systems aren't a LOT faster than the
> 604e systems.

I wonder who well-optomised for R10k the EI renderer
is (ie, is it just -mips2 or, gof forbid, -mips1
code? or did they do anything at all to make it fast?)
EI has spent a *lot* of time improving speed on the
mac (just like Adobe has with photoshop)  not only
millions of dollars were dumped here, but more
importantly, decades of research into how the
machine ticks at the core level.

Quote:> But you wanna see fast, try rendering the same type of image with Alias
> PowerRender on a big multiprocessor system. Wow...

I think it finishes before I actually save the SDL ! :)

--
| The wise man can look at a Grain of Sand, |
|  and envision a Universe. But the Stupid  |
|  man just rools around on the Beach for   |
|    a while, then stands up and yells,     |
|     "Hey, look!  I'm 'Vine Man'!"      |
-Ben.

 
 
 

Price/Performance info on SGI vs Competition?

Post by z.s. » Wed, 14 May 1997 04:00:00



>*Render test times conducted on project from "Buried In Time"
>?1996 Presto Studios, Inc. containing over 350,000 polygons,
>197 texture maps and 14 shadow casting lights.

I thought the Nintendo 64 did this in real time, at 30 fps   ;-)

--
  __                    //  MSDOS--Maybe Some Day an Operating System
 -/_)_ ( _ __)  __)___ //    http://jhunix.hcf.jhu.edu/~robodude
_/ \(_)_)_)_/(_/_/(-__/(___________________________________________________

 
 
 

Price/Performance info on SGI vs Competition?

Post by Michael B. Johns » Wed, 14 May 1997 04:00:00



>R4x00s are about twice as fast as the P-6 given
>equal clock speeds (softimage)

>And R10000s are, as you can see, an order of
>magnitude or so faster than a PP.

Ben, I appreciate your enthusiasm, but this information isn't only annoying,
it's really, really wrong.

In my personal experience, having access to all of the above, R10Ks are nice
chips in nice boxes, but so are the other two processors (modulo much of the
software running on them).  In my anecdotal experience (i.e. my friends and
I), for doing floating point intensive tasks (say, software rendering), a
200MHz Pentium Pro is about 80% the speed of a 195MHz R10K.  That's running
the same floating point intensive task atop a UNIX-based OS on each platform
- a friend running a renderer he wrote (and he's a smart guy) on my SGI and
his brand spankin' new 200MHz, <$3K PentiumPro box from Dell running Linux.  
Oh, and that PentiumPro kicks his old R4400's ass on the same app, MHz for
MHz.

Heck, the Pentium 133 running NEXTSTEP that I'm typing this on is only ~5-7
times slower for these sorts of things than my R10K Indigo^2, and that's
running prman, which I personally tend to think of as a pretty reasonable
real world benchmark for the kind of fpu and disk intensive activities that
*I* use computers for.  

SGIs are nice, and they're fast, especially for doing real time graphics, and
especially using their mid to higher end stuff, but saying things like "an
R10K is an order of magnitude faster than a PP" (whether PP stands for
PowerPC or PentiumPro), is just plain *wrong*, and undercuts the nice
rational arguments that can be made for SGI.

as always, these are personal opinions, and have no relation to my employer.

--

-->  Media Arts Technologist, Pixar Animation Studios (Eastern Office)
-->  alumnus, MIT Media Lab, Computer Graphics & Animation Group
-->  http://wave.www.media.mit.edu/people/wave/

 
 
 

Price/Performance info on SGI vs Competition?

Post by Bill Broadle » Wed, 14 May 1997 04:00:00



Quote:> Anybody know where I could find an objective price/performance
> comparison?  I'm particularly interested in how SGI stacks up against
> not only traditional competition like Sun, HP. IBM, etc., but also how
> they match up with PPro/Pentium II.  Any good sources for such info?
> Thanks very much for any advice!

Hmmm objective.... okay whats your measure of performance??

Standardized 3d benchmark tests?  (Arguably don't favor sgi)
        http://www.veryComputer.com/

Memory bandwidth?
        http://www.veryComputer.com/.*ia.edu/stream/standard/Bandwidth.html

Cpu int/fp performance/memory perf:
        http://www.veryComputer.com/

A different measure of cpu/memory systems:
        http://www.veryComputer.com/

Of course always best to use YOUR code as a benchmark, but barring that
the above can give you an idea of the relative performance of machines
for a specific task.

In general SGI's do well on memory bandwidth, and heavy 3-d graphics.  Of
course worlds fastest graphics aren't affordable to many.

Intel does very well on integer based applications for the money, I.e. specin95.

Alpha does very well on int as well as fp stuff (i.e. specint/specfp95)

Of course buying on these generalizations can get you in trouble.

Usually the best bet is to decide what software you need, eliminate
platforms it doesn't run on, eliminate what you can't afford,
then try to figure out some metric for measure the performance of your
application.

Best of luck.

--

Linux is great.         http://www.veryComputer.com/~bill                 PGP-ok

 
 
 

Price/Performance info on SGI vs Competition?

Post by Jeff Harrel » Thu, 15 May 1997 04:00:00




> > But you wanna see fast, try rendering the same type of image with Alias
> > PowerRender on a big multiprocessor system. Wow...

> I think it finishes before I actually save the SDL ! :)

Ah... real-time raycasting... <drool>

--
Jeff Harrell, Systems Engineer
Digital Media Performance Labs, Inc.
   http://www.dmpl.com
   http://jeff.dmpl.com

 
 
 

Price/Performance info on SGI vs Competition?

Post by Tony Sawye » Fri, 16 May 1997 04:00:00




> > Anybody know where I could find an objective price/performance
> > comparison?  I'm particularly interested in how SGI stacks up against
> > not only traditional competition like Sun, HP. IBM, etc., but also how
> > they match up with PPro/Pentium II.  Any good sources for such info?

> > Thanks very much for any advice!

> Hmmm objective.... okay whats your measure of performance??

> Standardized 3d benchmark tests?  (Arguably don't favor sgi)
>         http://www.veryComputer.com/

> Memory bandwidth?
>         http://www.veryComputer.com/.*ia.edu/stream/standard/Bandwidth.html

> Cpu int/fp performance/memory perf:
>         http://www.veryComputer.com/

> A different measure of cpu/memory systems:
>         http://www.veryComputer.com/

The company I work for, Gemini Technology, also has an "application"
benchmark which we use for comparisons of hardware at an application
level.  Although the baseline numbers for the benchmark are a little
out-of-date (SGI RE2) the comparisons are still valid and demonstrate
relative performance with the same application.  The application is
written using OpenGVS which uses OpenGL (or subsets for PRO-1000 and
3DFX) for rendering.  Keep in mind that these numbers use the EXACT same
application for all platforms and there may be cases where performance
could be improved (or quality improved without affecting performance) by
optimizing the application or database for a particular hardware
target.  These benchmarks are for the end-user who does not necessarily
want to do that and may wish to develop their application on/for
multiple platforms.  Check our home page http://www.veryComputer.com/

> --

> Linux is great.         http://www.veryComputer.com/~bill                   PGP-ok

--
Tony Sawyer

Gemini Technology Corporation
23792 Rockfield Blvd., Suite 160
Lake Forest, California 92630-2868

(714) 598-0961
(714) 598-0966 FAX



 
 
 

Price/Performance info on SGI vs Competition?

Post by Steve Deml » Fri, 16 May 1997 04:00:00



> I'm particularly interested in how SGI stacks up against
> not only traditional competition like Sun, HP. IBM, etc., but also how
> they match up with PPro/Pentium II.

The Apr. 22, 1997 'PC Magazine' compares graphics workstations by DEC, HP,
SGI, Sun, Compaq, and two Mac clone makers.

Their conclusion?  Figure out what software you need and what machines it
runs on, _then_ look at the hardware.

Steve
--

AHPCRC Graphics and Visualization Support             - (612) 626-8075 (voice)
Minnesota Supercomputer Center, Inc., Minneapolis, MN - (612) 626-1596 (fax)

 
 
 

Price/Performance info on SGI vs Competition?

Post by Ben Canno » Sat, 17 May 1997 04:00:00



> =

> =
> >*Render test times conducted on project from "Buried In Time"
> >=A91996 Presto Studios, Inc. containing over 350,000 polygons,
> >197 texture maps and 14 shadow casting lights.
> =
> I thought the Nintendo 64 did this in real time, at 30 fps   ;-)

LOL!  Maybe with 35 polygons, and 5 maps and 2 lights :)

350,000 polys is a *lot* of polys!!!

-- =

| The wise man can look at a Grain of Sand, |
|  and envision a Universe. But the Stupid  |
|  man just rools around on the Beach for   |
|    a while, then stands up and yells,     |
|     "Hey, look!  I'm 'Vine Man'!"      |
-Ben.

 
 
 

Price/Performance info on SGI vs Competition?

Post by Ben Canno » Sat, 17 May 1997 04:00:00




> >R4x00s are about twice as fast as the P-6 given
> >equal clock speeds (softimage)

> >And R10000s are, as you can see, an order of
> >magnitude or so faster than a PP.

> Ben, I appreciate your enthusiasm, but this information isn't only annoying,
> it's really, really wrong.

Um... thanks.....

Quote:> In my personal experience, having access to all of the above, R10Ks are nice
> chips in nice boxes, but so are the other two processors (modulo much of the
> software running on them).  In my anecdotal experience (i.e. my friends and
> I), for doing floating point intensive tasks (say, software rendering), a
> 200MHz Pentium Pro is about 80% the speed of a 195MHz R10K.  That's running
> the same floating point intensive task atop a UNIX-based OS on each platform
> - a friend running a renderer he wrote (and he's a smart guy) on my SGI and
> his brand spankin' new 200MHz, <$3K PentiumPro box from Dell running Linux.
> Oh, and that PentiumPro kicks his old R4400's ass on the same app, MHz for
> MHz.

Well, I pase my "order of magnitude" on expierence with Softimage
renderer (both SI's own, and the newer Mental Ray) running under
NT on the P-6 boxes. I maintain that r4400s are as fast as a P-6
mhz for mhz (with a 4meg cache, that is. I've far more expierence
with 4meg cache R4400s)

Who knows what kind of wild results a small, written-and-oirigionally
compiled for Linux/PC and proted to SGI renderer could yeild?

Did you compile it -O3 -n32 even?

Of course, the more responces anyone gets, the better idea of how
true any of us are being.   Anyone out there with more to offer?

--
| The wise man can look at a Grain of Sand, |
|  and envision a Universe. But the Stupid  |
|  man just rools around on the Beach for   |
|    a while, then stands up and yells,     |
|     "Hey, look!  I'm 'Vine Man'!"      |
-Ben.