Does size matter when it comes to processor speed?

I just heard an interesting comment that increasing CPU speed, say from
2.8 to 3.0 GHz will not really do much to speed up large assembly
performance. This came as a bit of a shock to me. Does anybody have any
comments or experience with upgrading CPU and assembly performance?

 

My experience is the same as Dale's, that rebuild times are inversely
proportional to clock speed, when all else is held equal. The person who
made the comment that going from 2.8 to 3.0 GHz won't do much was probably
talking about the perception of the speed increase. That's only a 7% change
in clock speed, and will only get you a 7% decrease in build time. I don't
think most people would notice a 7% change in speed. Many years ago I was
told by a code writer that to really get people's attention, you need to
speed things up by a factor of two.

Jerry Steiger
Tripod Data Systems
"take the garbage out, dear"

 

Still, if you're dealing with large assemblies you will never get
decent performance out of SolidWorks regardless of what you do to your
computer. SWX is a dog, plain and simple.

 

That is what I thought because that has been my experience. Still it came
from a "guru".

The other part of the problem and the one that I think is more interesting
is that this guy has been using a dual processor setup and running a lot of
ancilary programs. Someone made mention of the fact that it might be that
being dual processor, one of the processors might take ram away from the
other in ways that SW doesn't like. I think I have seen SW crash if it
didn't have enough contiguous memory. Given that SW power users are now
running up against the system limitations on ram on a regular basis I
wouldn't be surprised if this sort of problem cropped up with dual
processors running other applications where the second processor fragmented
the memory space.

 

Using your example.

Simple math:

(2.8) / (3.0) = 0.93
0.93 * 100 = 93%


Conclusion:
A 7% gain.


Do you really thing you'll notice the improvement?

jv

 

Conclusion:

which is an idealized maximum, because most likely your model and SW
do not fit in the processor cache, so access to data will still be
bottlenecked by your memory/bus bandwidth. I'd say you should be
happy if you get a 2% gain...
IMHO you get more bang for the buck with more & faster RAM, maybe a
newer mainboard.
Philippe Guglielmetti - www.e-systems.ch

 

Dale I don't think he was saying it takes more memory it just fragments the
memory more, and he was saying he thought SW likes memory to be in a strait
line instead of jumping all over the cards for needed memory.

Corey

 

I was the one that made the memory comment and I was wondering if maybe the
operating system might see 2 processors and allocate memory equally to both
of them. If that's so, then since SW can only run one processor, it only
has a max of half the physical RAM - not good.

WT

 

Dale,

Our MSI dual Opterons don't have seperate memory for each processor, they're
unified.

I haven't seen any NUMA boards for years.

Regards

Mark

 

The effective buss speed can be increased in many cases by
using a wider buss. More bits in parallel per clock cycle.
In some cases it can be "pipelined" too, IIRC, sending
data before the last send is completed, if latency along the
buss is an issue.

The speed of data on the buss should be the same, no matter
the distance (in a perfect world), being limited by the speed of
light *in the buss material* (speed of propagation of electric
fields).

The problem can be distance if the sending device of the data
must await a signal from the other end before sending more data.

I hope I got that all right <G>.

 

What usually makes a big difference is newer, better performing chipsets,
and faster memory. The computer as a whole "system" is what produces the
speed. Just sticking in a marginally faster processor won't do much, but
upgrading the chipset, the memory, and the disk drive to better performing
ones will certainly make a big difference. We upgraded Dell 360's to Dell
370's and the processor was only 15% faster, but the complete system yielded
Algor analysis time improvements of over 50% when tested.

Ken

 

I've heard that a gas engined bus was actually faster than a diesel engined
bus. Personally, I think if they installed a couple more lanes on the
highway, then multiple busses could run in parallel, thus transporting more
passengers. Of course, if a dedicated path was constructed between the
processor and the memory, a train could be used. It would be more power
efficient than any bus at transferring passengers between destinations, and
it would be much faster too, especially if a fast maglev system was
employed. Of course you would want to keep it at the other end of the case
away from your hard drives

Ken

 

This is one of the things I see running a 64bit AMD. The memory bandwidth
makes a big difference.

 

There are tools in MS Windoze to do this. Performance monitor will monitor
just about anything that can be measured and log it. I have found
procedures on the MS website on finding bottlenecks. But doing this is not
for the faint of heart.

 

So are we concluding that:

1. It is not currently possible to obtain a system that will show a
noticable/useful/cost effective improvement over a 2.8 to 3 GHz system?

2. That SW therefore has hit the wall in term of performance that can be
obtained with hardware?

3. That further improvements must come from the software side or taking
advantage of new technology like 64 bit or multi-processing?

 

My testing shows that SW does need memory in large blocks. It is the only
explanation on why my Patbench benchmark will die before exhausting all 3Gb
of ram available to it.

We have to be careful here though because SW has changed how it uses memory
from 2001+ to 2005.

It should be possible to use Performance Monitor to see which processor gets
the memory.

 

Higher speed = higher cost & more power/heat. Also
smaller features on the chip.
Information costs.
Most of the time (clock cycle to clock cycle) a bit in memory is
unused so why pay a lot to store it or access it if you don't
have to?

 

You could always try cooling the chips & stuff & then
overclocking them, if it's allowed.
Cooling/heat is a limit on clock speed.