Best Processor

I know this has been beat to death but Im gonna swing again.
What is the better processor today for Solidworks, Cosmos Works and
Motion. 1000 piece assemblies.
So many possibilities and so little time to do my homework, so I
figured I'd pick a few brains out there to save a bit of pain.

These postings usually get hijacked to discuss personal gripes. Please
keep it to relevant discussion.

 

Hard to say. I haven't seen anyone willing to benchmark the new crop
of Intel Duo processors. If you are comparing the older Pentium to AMD
Athlon or Opteron then the AMD items are better. I have tested a three
year old AMD against a newer Intel Duo (top of the line from Dell) and
find it to only be about 20% faster on some things and slower on
others. I have no explanation for this since the Intel Duo should be
multiple times faster than my AMD64 FX53. To only see 20% improvement
when comparing a three year old processor to a brand new dual core
just doesn't seem right.

TOP

 

Using a Intel Core 2 Duo E6700 (2.67GHz/1066/4MB) atm and it works
fine with 5k+ parts (its not really the processor that is the
bottleneck with large assemblies)

 

What is the bottelneck at the large assembly stage?
How about when doing FEA?

 

Ronni,

Have you run any benchmarks? I can say the same about my AMD64 FX53 or
the new Intel DUO. Fine is subjective. You should be able to find Ship
in a Bottle by searching the news group. Run that and let us know.
Thanks in advance.

TOP

 

We run a gigabit network. Until the IT guy got a hold of it, the
network wasn't the bottleneck. We could run large assemblies as fast
over the network as locally. From watching network bandwidth either
large assemblies or large assembly drawings will hit the network for a
very short intense burst and then settle down to regenerating whatever
needs rebuilding. And this takes processor time.

TOP

 

I presume the processor is not teh bottleneck on the following data I
have collected.

When I load a big assembly it uses less than 30% of my CPU while
loading the data into RAM.
After that is done my processor goes to 100%.

The loading time for a 5k assembly is often 5-10 minutes, where the
rebuild after loading it into RAM and then processing the data takes
20-30 secs.


But if you look at how it is to work with you get a better performance
with a Dual Core 2 processor compared to a Pentium 4 3.0 ghz which I
had before this.
Rotating, updating, section cuts, assembly cuts etc.

So of course you get a better performance working within the program
with a better processor but to me its not the bottleneck, since
waiting a few secs on an action is much less irritating than sitting
and waiting 10 minutes for a assembly or drawing to load.

But to put things in perspective from what I have heard:
from Pentium 4 3.0 ghs -> duo core 2 - 2,67 ghz is a ~30% upgrade.
where a duo core 2,67 -> duo core xeon 3,0 is a less than 5% upgrade.

so value for money, getting a 2,xx duo core is a good idea.


All these things has to been seen from a perspective working with big
assemblies where the complexitivity is low and it is the number of
parts and different configurations that needs to be handled.


@Top: I have looked for *"Ship in a bottle" but the links I have found
where dead ( and was dated 2003 - 2004) do you have a link for it?

 

That seems logical.
I am looking at Xi computers and had them spec a box that would run
SWX with Cosmos. I call the sales department and discussed the quote
that they were proposing. When I started discussing the "why he picked
specific components" it became apparent that what he was doing was
matching the bus speeds of the processor and memory (1066 FSB).
Eliminating the most critical bottleneck. The next step he was
proposing was using Raid 0 to increase the transfer rate when writing
to disk. I am still stuck behind a 10/100 network so I will have to
speed that up as well, but working local should be better.

 

This topic came up a while back in our User group and there were a lot
of comments where the quality of the graphic card seemed to have a
greater impact then just about everything else.

The other thing that the computer designers have typically "missed"
for CAD applications has been the back plane speed on the mother
boards. The CPU speeds seem to be increased but the back plane speeds
havn't always kept up. A few years ago when the back plane speeds
were improved but I don't know if this trend has kept pace with the
CPU's. It would be interesting to see a comparison between the newer
CPU Chip sets and the back plane speeds for the new computers compared
to some of the older boards/CPU's combinations.

I don't know but suspect that the better graphic cards are more
effecient and some how place a lighter load on the back plane speeds
which makes them appear to be better.

Just some thoughts,

EdT

 

If your network is the bottleneck then spend a few thousand or less on
fixing it and not the processor. I have run vault conversions on a
gigbit network and simply not seen those kinds of load times even with
older 3.4GHz Pentiums.

Pentiums are slow even by yesterday's standards. Intel Duos are going
to be faster, but are not that fast. Consider AMD processors and
plenty of RAM.

Ship in a Bottle is here:
http://www.mikejwilson.com/solidworks/files/ship_in_bottle.zip

You will find the definitive thread on Ship in a Bottle here:
http://groups.google.com/group/comp...a0299be4744/e882c5ea18d19185#e882c5ea18d19185

Don't be surprised if even fast machines can't get the kind of rebuild
times posted in this thread. That was with SW2001+.


TOP

 

Thanks TOP

tried 50 rebuilds, 3 tries with the following results:

28,24 secs
27,32 secs
27,18 secs

 

Now compare your times to this:

http://groups.google.com/group/comp...st&q=Ship+in+a+Bottle&rnum=3#0bc6c2df7487a584

The AMD 4800+ is available in the $250 range on the street and is
certainly not the fastest chip AMD makes.

TOP

 

That's one benchmark run without any standardization. Ronni - what
resolution were you running? What was your graphics quality setting?
This has a huge impact on SIAB results.

Benchmarks run in a controlled, standardized environment have been
published by several reputable sources which I've referenced several
times before. (www.mcadonline.com, www.cadcamnet.com, www.digit-life.com,
etc.) On most, not all, benchmarks, the Core2 Duo CPUs outperform
Athlon X2 CPUs running at the same clock speed. On the CPU-bound
SPECapc for SolidWorks 2005 benchmark the difference C2D advantage is
about 10% - and greater on most other benchmarks. The Athlon
outperforms the C2D on the SolidWorks STEP Import tests published by
MCAD Online and on some of the MATLAB tests published by digit-life.
In general, the Athlon architecture excels at "pure" floating point
operations that are not using SSE optimizations while the C2D excels
at SSE-optimized operations and integer performance. The Athlon may
be a better platform for intensive FEA and/or CAM applications that
rely on FP calculations, but there is little published to prove or
disprove this.

Until recently, Intel held a significant price/performance edge over
AMD at the mid-high end of the market, but the latest round of AMD
price cuts puts AMD in a much more competitive position - at least
until April 22 when new Intel pricing takes effect. Using the XI
Computer on-line configurator 4/11/07, in an otherwise identical
configuration, a system witn a 3.0GHz Athlon X2 6000+ is priced $160
less than a system with a 2.66GHz C2D E6700. AMD holds a similar
price/performance advantage when comparing the x2 5600+ with the
E6600. Below that, Intel again gains the upper hand in value as the
lower-end C2Ds are priced far lower than comparable X2s.

So, chalk two up for AMD - at least until Intel price reductions go
into effect on April 22.

On the other hand, the 6000+ is the top of line for AMD with
absolutely no room for overclocking. If performance is more important
than price, then Intel offers a 10% faster dual core CPU or a quad
core CPU at the same clock speed as the E6700. XI Computer sells pre-
configured Intel dual and quad core systems overclocked to 3.2-3.4GHz
- with full factory warranties at a still reasonable price point.
Obviously, you'll pay a hefty premium for such performance, but AMD
can't touch it at any price.