PC outperforming workstation when running EDA?

Hi all! This is probably not the right forum to post this question, but
I recently installed IC 5.1.41 on a IBM thinkpad T41 running Fedora
Core 4, and to my disbelief it seems to outperform the SUN Blade 1500
that I have always used!! The Analog Environment simulations of my
thesis circuit (using spectre) run about the same speed, and the Matlab
first-order model that I am also using for algorithmic validation
clearly runs faster in the laptop! (even 50% faster!).

I read somewhere on the web that nowadays the differences between
workstations and PCs were narrowing, but now I don't understand what's
the point of using expensive workstations when it seems that now you
can get comparable computational performance with PCs. For instance,
why virtually all EDA vendors compile for workstation platforms? What's
the benefit (besides reliability) of using workstations instead of
modern PCs?

BTW,
the laptop: Pentium M 1.6GHz, 512MB RAM, Fedora Core 4.
the workstation: Sparc 1.5GHz, 1GB RAM, Solaris 10.

Am I missing something here?

Thanks in advance for your comments.

Regards,

Jorge Luis.

 

Were you running IC 5.1.41 in 32-bit or 64-bit mode on the Sun Blade
1500? I very recently set up IC 5.1.41 on a Sun Blade 1500 (running in
32-bit mode) and encountered something similar. When accessing the
module from a Windows XP machine using Hummingbird Exceed to provide the
Xwindows interface, I remarked to a co-worker that it seemed to be
running faster on Windows remotely than it was running natively sitting
in front of it on the Sun machine. So, what's the difference?

A few days ago I encountered a tip about setting up Sun framebuffered
graphics cards. The tip makes mention of breaking out from the 8 color
default! Wow! I never paid attention, but it did occur to me that the
video mode and configuration could certainly explain the performance
difference.

Here's the link: http://tille.xalasys.com/training/solaris/resolution.php

I'd be interested to hear what others have to say on this topic of
Cadence performance on Sun machines vs. InX/86 machines. Anyone?

Thanks!

John

 

Aha! I knew I was not the only one noticing this...

With respect to the 64-bit or 32-bit question... well, throughout the
entire installation process I was never asked to make this selection,
so I have no idea. Is there a command that I could run in the console
to check what version am I running?

Thanks for the link on configuring the display on SUNs. I will review
it to see if by tweaking it I can improve the performance of Cadence
tools!

But the question still remains... are modern PCs comparable to SUNs
with respect to computational capacity? (not talking about realiability
or operating system estability).

Regards.

 

Jorge,

Calling something a "workstation" or "a PC" is a totally arbitrary
distinction meaning very nearly nothing.

There is just hardware, operating systems, and applications.
You try to make rational decisions based on what you know
and can test.

Pentium M is a nice CPU architecture. There's a reason that
new and future Intel processors will look a LOT more like Pentium
M than they do like P4.

No surprise that given an application with no special tuning
(Spectre) a 1.6GHz Pentium M and a 1.5GHz US-III run about
the same speed. That's what you would expect.

It is interesting that you see a significant difference on Matlab;
it looks like Mathworks has spent some serious effort tuning
for x86 hardware. Given their customer base (mostly x86
Windows), that would make sense. It would be fun to see
what a knowledgable Mathworks developer would have to
say about that.

I'm not aware of any organizations (except dinosaurs) that
are buying UltraSPARC hardware to run Cadence apps,
except for extremely large process size applications that
don't have good x86-64 coverage yet (can't think of any off
hand, but there might be some).

x86 and Opteron are the way to go. Sadly, stability/reliability
on Linux is not as good as Solaris, but you get a lot more
work done between seg faults. I'm still dreaming of
Solaris-x86...

-Jay-

 

That will make no speed difference. 64-bit lets you run bigger, not
faster.

See my post upthread, and search this newsgroup.

-Jay-

 

You're right... there is nothing mentioned about 64-bit when you go
through the installation. I stumbled on it in the, "Cadence Design
Framework II Configuration Guide." The best advice I can give you is:
get there and download the PDF version. Then you can quickly do a search
for "64 bit" and get right to it. It's a matter of setting an
environment variable, but there is more you need to read, such as,
always start 64-bit from the "wrapper", never directly.

http://sourcelink.cadence.com/docs/files/Release_Info/Docs/dfIIconfig/dfIIconfig5.0/dfIIconfig.pdf

Hope that helps!!

John

=======================

 

Jay, please see my previous posting to this newsgroup:
"Please see System Specs inside and comment. Thank You! "

This is a system spec I was given to post for feedback, prior to the
purchase order being approved. To summarize, it's a Sun Server running
Dual x Dual-Core 64-bit Opterons. Seems that would have the best of both
worlds--x86 performance and Sun reliability.

The one thing I should have clarified, 20 users is the Max (when a class
is held in the lab), not the expected norm. 4 to 8 users is a more
realistic expectation for the norm.

Feedback please, Jay? Thanks!!

Regards,

John

=================

 

Take a look at the spec benchmarks: http://www.spec.org/cpu2000/results/

There is no clear winner in terms of computational capacity. I wouldn't
even say that Sun machines are on average faster than PCs. There are many
measures of speed (floating point, integer, IO, cache access), and
different CPU architectures will perform differently on various tests. In
general I would expect a modern Sun/Sparc to be comparable in some ways to
a modern Sun/x86_64, a modern x86 PC, a modern AMD64 PC, etc. Same thing
with memory and most other performance/capacity measures. If you need >
4GB of memory, best to go with a 64-bit architecture. There are many other
considerations to make when choosing the right system, not just
computational capacity - if you're willing to pay enough any major
CPU/system vendor should be able to meet your capacity needs.

Also, with regards to workstation vs. PC vs. laptop: A PC can be used as a
workstation nowadays, and a laptop is a type of PC. The distinction
between PC and workstation has become fuzzy. I like to think of
workstations as having multiple simultaneous users and PCs as having one
user. Sure, a PC or even a laptop can be faster than a workstation at
running a single process, but smaller computers such as laptops don't
scale to larger jobs because they tend to have fewer processors, less
memory, etc, and in many cases having equivalent CPU power in a smaller
machine has higher cost.

Also, the CPU speed itself can be misleading
because the instruction sets differ across architectures and one machine
may have a lower CPU clock rate but higher bandwidth (and overall
instruction throughput) due to more pipeline stages, etc.

Frank

 

Thanks to everybody for the comments and observations; they are
certainly appreciated. Seems that I had better get accustomed to the
Cadence look on Fedora

Regards,

Jorge Luis.

 

You are not alone, my company has switched from sun to x86 also. We have
seen a major increase in speed on personal workstations, (PC). It not
exactly apples to apples but you can get much more of a machine for your
buck with an x86 rather then Sun!! On side there are some bugs but Cadence
seems committed to Linux.

 

I must ask how big ( memory footprint) the simulation were.

Classroom examples might typically require less than 32MEG of virtual
memory.
I remember a time when production simulations that were < 24Meg memory
footprints would smoke on a linux box,
but the Sun seemed to handle the larger jobs much better.

Note that these results are somewhat dated and I think that the memory
swapping has gotten much better in Linux hardware.

YMMV -- Gerry