speed of tran analysis in SPICE3

[vivian]

Hi,

SInce the stop time if my TRAN analysis is long, tens of micro
seconds, it takes
a long time. Any idea of decrease the time consuming on the simulation?
No difference if a big time step is used.

Thank you.

 

[Helmut Sennewald]

Hi,

SInce the stop time if my TRAN analysis is long, tens of micro
seconds, it takes
a long time. Any idea of decrease the time consuming on the simulation?
No difference if a big time step is used.

Thank you.

Hello Vivian,

maybe your PC is too slow.

A ".tran 100us" takes exactly 1.2345sec on my PC.

What exactly means SPICE3 in your case?

I recommend to try your circuit with some other SPICE-programs.
My secret tip is 'ecipsTL'. Just read this '' backward.

Best regards,
Helmut

 

[Jim Thompson]

Hello Vivian,

maybe your PC is too slow.

A ".tran 100us" takes exactly 1.2345sec on my PC.

Eh? How can that be? Simulation speed is circuit-content dependent.

What exactly means SPICE3 in your case?

I recommend to try your circuit with some other SPICE-programs.
My secret tip is 'ecipsTL'. Just read this '' backward.

Best regards,
Helmut

...Jim Thompson

 

[vivian]

Thank you for your reply.

SPICE3 is the spice program for unix.
WHat does 'ecipsTL' mean?

 

[vivian]

Hello, Jim

Is there anything I can do to increase the speed?
My circuit is a ring VCO.

Thank you.

 

[Jim Thompson]

Hello, Jim

Is there anything I can do to increase the speed?
My circuit is a ring VCO.

Thank you.

Which flavor of Spice?

Oscillators always need a small max timestep

Use a .IC statement, or a starting current pulse

...Jim Thompson

 

[Helmut Sennewald]

Thank you for your reply.

SPICE3 is the spice program for unix.
WHat does 'ecipsTL' mean?

Hello Vivian,

This 'ecipsTL' is LTspice. It's written for WIN-98,ME,2K,XP.
You can use it with WINE under Linux as well, but
99.99% of the users run LTspice under the native OS WIN-xx.
Please use it with WIN-xx to avoid any unnecessary problems.
It can be used with netlists or with schematics.

http://ltspice.linear.com/software/swcadiii.exe

Best regards,
Helmut

 

[Genome]

Hello Vivian,

This 'ecipsTL' is LTspice. It's written for WIN-98,ME,2K,XP.
You can use it with WINE under Linux as well, but
99.99% of the users run LTspice under the native OS WIN-xx.
Please use it with WIN-xx to avoid any unnecessary problems.
It can be used with netlists or with schematics.

http://ltspice.linear.com/software/swcadiii.exe

Best regards,
Helmut

You still get slapped wristies for......

<quote>
A ".tran 100us" takes exactly 1.2345sec on my PC.
</quote>

Fortunately your name sounds like you are German so, other than no sense of
humour, we might assume you have a different one.

I think the origin of Germans having no sense of humour arises because the
language is at least 84.607E3% longer than the rest of the stuff we grunt
with and the audience goes to sleep long before the punchline.

OK..... I'm an asshole.

DNA

 

[vivian]

Hello Jim,

I am using Spice3 under unix. Actually, I have used .IC statement in
my spice file.
Does that help?

 

[Jim Thompson]

Hello Jim,

I am using Spice3 under unix. Actually, I have used .IC statement in
my spice file.
Does that help?

The .IC statement usually improves start-up time dramatically, but
generally won't have any effect on the length of time the total
simulation takes.

...Jim Thompson

 

[vivian]

Thank you all for your kind help!

vivian

 

[Fred Bartoli]

You still get slapped wristies for......

<quote>
A ".tran 100us" takes exactly 1.2345sec on my PC.
</quote>

Fortunately your name sounds like you are German so, other than no sense of
humour, we might assume you have a different one.

I think the origin of Germans having no sense of humour arises because the
language is at least 84.607E3% longer than the rest of the stuff we grunt
with and the audience goes to sleep long before the punchline.

How can you think that people insisting on saying 21 "one and twenty" lack
sense of humor?

 

[vivian]

Thank you for your kind help.

 

[Brett]

Hmmm. Getting rid of any un-needed capacitances,or lowering their values may
help. Also, look into a faster machine with more memory? Reading the replies
from the (V-E-R-Y elite) S.E.C. social club probably won't help, because it
seems they are mostly a bunch of self-absorbed assholes.

 

[Helmut Sennewald]

Hello Brett,

The first message from Vivian contained no information about
the simulated circuit. How could somebody give any useful
advice except running it on the fastest PC?

After the second posting from Vivian we have got informed that
it is a ring oscillator. I expect now that it's an odd number
of inverters in a loop. There is still no detailed information
about the circuit, e.g. the number of stages. It makes a big
difference whether you use 3 or 31 inverters.
In the ladder case the simulation time will be 10 times longer.

After the third message we know it's any SPICE-3 simulator
under Unix. Something self compiled?
Who can give any advice for an unknown program?


---
Hello Vivian,

I have tried now with an 11 stage ring oscillator which oscillates
at about 1.8GHz. The extrapolated simulation speed on the fastest
edge P4 or AMD64 would be about 20ns per second of simulation time
with the high precision setting for the LTspice-simulator(default).
This speed can be increased by factors(2 to 5) when the accuarcy
requirement will be decreased. The important parameters in LTspice
are reltol and trtol.

It's still not specified how many micro seconds.
Let's assume he wants 40us.
This would require in my case 2000seconds when using the
high precision simulator settings. The amount of data may be
in the range of 20 millions data points per saved node voltage.

* Please remember that these are results for my circuit
* which is most probably not the circuit from Vivian.
* Maybe he has used less inverter stages which gives
* a shorter simulation time.

Conclusion:
Ring oscillators cause a lot of transitions where SPICE
automatically has to decrease the timestep to get useful results.
The number of circuit nodes also increases with more stages.
Overall the simulation time has been roughly proportional
to the number of inverter stages.

Best regards,
Helmut

LTspice is an unlimited free SPICE program with GUI from Linear Technology.
http://ltspice.linear.com/software/swcadiii.exe

There is an independent user group.
http://groups.yahoo.com/group/LTspice/

 

[Jim Thompson]

Hmmm. Getting rid of any un-needed capacitances,or lowering their values may
help. Also, look into a faster machine with more memory? Reading the replies
from the (V-E-R-Y elite) S.E.C. social club probably won't help, because it
seems they are mostly a bunch of self-absorbed assholes.

[snip]

I may be "self-absorbed", but I'm not clueless, as you appear to be.

...Jim Thompson

 

[Evgenii Rudnyi]

SInce the stop time if my TRAN analysis is long, tens of micro

seconds, it takes
a long time. Any idea of decrease the time consuming on the simulation?
No difference if a big time step is used.

In principle, there is a technique "model reduction" that allows us to
reduce the dimension of the original circuit. You can have a look at

R. W. Freund, Reduced-Order Modeling Techniques Based on Krylov
Subspaces and Their Use in Circuit Simulation
http://cm.bell-labs.com/cm/cs/doc/98/3-02.ps.gz

Unfortunately, in order to use this you need to have special software.

Best wishes,

Evgenii Rudnyi

 

[Jim Thompson]

Hello Brett,

The first message from Vivian contained no information about
the simulated circuit. How could somebody give any useful
advice except running it on the fastest PC?

After the second posting from Vivian we have got informed that
it is a ring oscillator. I expect now that it's an odd number
of inverters in a loop. There is still no detailed information
about the circuit, e.g. the number of stages. It makes a big
difference whether you use 3 or 31 inverters.
In the ladder case the simulation time will be 10 times longer.

After the third message we know it's any SPICE-3 simulator
under Unix. Something self compiled?
Who can give any advice for an unknown program?


---
Hello Vivian,

I have tried now with an 11 stage ring oscillator which oscillates
at about 1.8GHz. The extrapolated simulation speed on the fastest
edge P4 or AMD64 would be about 20ns per second of simulation time
with the high precision setting for the LTspice-simulator(default).
This speed can be increased by factors(2 to 5) when the accuarcy
requirement will be decreased. The important parameters in LTspice
are reltol and trtol.


It's still not specified how many micro seconds.
Let's assume he wants 40us.
This would require in my case 2000seconds when using the
high precision simulator settings. The amount of data may be
in the range of 20 millions data points per saved node voltage.

* Please remember that these are results for my circuit
* which is most probably not the circuit from Vivian.
* Maybe he has used less inverter stages which gives
* a shorter simulation time.

Conclusion:
Ring oscillators cause a lot of transitions where SPICE
automatically has to decrease the timestep to get useful results.
The number of circuit nodes also increases with more stages.
Overall the simulation time has been roughly proportional
to the number of inverter stages.

Best regards,
Helmut

LTspice is an unlimited free SPICE program with GUI from Linear Technology.
http://ltspice.linear.com/software/swcadiii.exe

There is an independent user group.
http://groups.yahoo.com/group/LTspice/

I just pulled out a 7-stage BIPOLAR ring oscillator from my files,
simulated for a total of 1us with max timestep set to 1ps (*).

Simulation time 915 seconds, 1.09ns per second of simulation,
oscillator frequency 2.7GHz.

(*) To get good resolution for the Spectral/Fourier that was also
performed.

With no max timestep restriction, 101.86 seconds of simulation time,
9.82ns per second of simulation, but seriously degraded spectral view.

A CMOS circuit would likely run faster, since there are far fewer
elements.

...Jim Thompson

 

[JosephKK]

Hmmm. Getting rid of any un-needed capacitances,or lowering their values
may help. Also, look into a faster machine with more memory? Reading the
replies from the (V-E-R-Y elite) S.E.C. social club probably won't help,
because it seems they are mostly a bunch of self-absorbed assholes.

Brett, you just elected yourself to the asshole status IMNSHO. Some of
these people are very helpful SPICE experts. They can also get distracted
talking to past and around each other and not answer OP. If you really
want them to help, post your netlist or if it is large put it on a web page
or some other publicly accessable place. And help you will normally get.

 

[Jim Thompson]

Brett, you just elected yourself to the asshole status IMNSHO. Some of
these people are very helpful SPICE experts. They can also get distracted
talking to past and around each other and not answer OP. If you really
want them to help, post your netlist or if it is large put it on a web page
or some other publicly accessable place. And help you will normally get.

And recommending getting RID of capacitance ranks Brett as the most
clueless one of the year... SHEEEESH ;-)

...Jim Thompson