Is this Intel i7 machine good for LTSpice?

[John Larkin]

On Sun, 02 Nov 2014 16:21:32 GMT, Jan Panteltje
<pNaonStpealmtje@yahoo.com> wrote:

On a sunny day (Sun, 02 Nov 2014 08:00:36 -0800) it happened John Larkin
jlarkin@highlandtechnology.com> wrote in
0nkc5aljhec5r36ptkoaqbt0a48ud2j5vo@4ax.com>:

On Sun, 02 Nov 2014 07:25:49 -0800, Joerg <news@analogconsultants.com
wrote:

Folks,

Need to spiff up my simulation speeds here. IIRC Mike Engelhardt stated
that the Intel i7 is a really good processor for LTSPice. According to
this it looks like the 4790 is the fastest of the bunch:

http://www.intel.com/content/www/us/en/processors/core/core-i7-processor.html

So, what do thee say, is the computer in the Costco link below a good
deal for LTSpice purposes?

http://www.costco.com/Dell-XPS-8700-Desktop-%7c-Intel-Core-i7-%7c-1GB-Graphics-%7c-Windows-7-Professional.product.100131208.html


It's also available without MS-Office Home & Student 2013 for $100 less
but I found that OpenOffice isn't 100% compatible in the Excel area so
that sounds like an ok deal. My hope is that it can drive two 27"
monitors but I guess I can always add in another graphics card if not.

Reason I am looking at these is that I absolutely positively do not want
any computer with Windows 8 in here and unfortunately that's what many
others come with.

I have spent too many hours this weekend tweaking the transient
response of a semi-hysteretic (we call it "hysterical") switchmode
constant-current source. There are about 8 interacting knobs to turn.
At 30 seconds per run, understanding the interactions is impossible.

I want sliders on each of the part values, and I want to see the
waveforms change as I move the sliders, like they were trimpots on a
breadboard and I was looking at a scope. I need maybe 500 times the
compute power that I have now.

Mike should code LT Spice to execute on a high-end video card.

Maybe building the real thing with some pots?

I am halfway through building a breadboard; I'll post pics. I'm after
extreme broadband high impedance output, which is hard to measure on a
breadboard; Spice lets me graph all sorts of currents and nodes, so
it's the best platform for development.


>But without some theory backing it up how would you know it always works?

I'll have to simulate, and then test, the thing over a range of loads.


>And with the theory you do not need the sliders.

I don't have sufficient theoretical skills to tune this circuit. I'm
not sure if anyone does.

I do not see the need for insane speeds, I have used LTspice more than often
the last few days, running on an old Duron 950, fast enough.
maybe you guys are doing something wrong?

At 30-50 seconds per run, iteration is slow. Worse, the time lag
wrecks my ability to acquire intuition about what's going on.

And it is always an approximation, build the real thing too,
needed tweaking with resistors in series, that is analog,
got some nice 25 turn Bourns trimpots from ebay.....

I'll have to tweak resistors and capacitors, and the cap values are
too big for variable capacitors. And, as noted, it would be hard to
instrument.

Here's the current output when the load voltage steps from about 0.5
to 3 volts.

>https://dl.dropboxusercontent.com/u/53724080/Circuits/Current_Sources/Hysterical_A1.jpg

I want it as flat as possible.

The fast ripple is the basic 1.5 MHz switcher frequency. The various
whoopie-doos are from loop dynamics and the chain of progressively
smaller, bias-tee-like damped inductors between the switcher and the
load. The constant-current hysterical switcher is, natively, about 4
or so orders of magnitude too slow for my application.

Everything interacts with everything else; it's like tuning a big LC
filter by hand, never a fun thing to do. Spice helps me acquire at
least some instincts for tuning. Maybe I can fix the cap values and
tune only resistors on the breadboard.

Rob, one of my guys, has a fierce Linux computer just for sims and
FPGA compiles, and he knows how to do automatic iterative parts value
tweaking in a loop around Spice. Maybe he can set up the problem and
run it for a couple of days or weeks.

I could probably step each of the six most important values, maybe 4
steps each, and pick the best waveform. That would be 4096 sims, about
60 hours of computing on my PC.




--

John Larkin Highland Technology, Inc

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[John Larkin]

On Sun, 02 Nov 2014 09:28:33 -0800, Joerg <news@analogconsultants.com>
wrote:

John Larkin wrote:
On Sun, 02 Nov 2014 07:25:49 -0800, Joerg <news@analogconsultants.com
wrote:

Folks,

Need to spiff up my simulation speeds here. IIRC Mike Engelhardt stated
that the Intel i7 is a really good processor for LTSPice. According to
this it looks like the 4790 is the fastest of the bunch:

http://www.intel.com/content/www/us/en/processors/core/core-i7-processor.html

So, what do thee say, is the computer in the Costco link below a good
deal for LTSpice purposes?

http://www.costco.com/Dell-XPS-8700-Desktop-%7c-Intel-Core-i7-%7c-1GB-Graphics-%7c-Windows-7-Professional.product.100131208.html

It's also available without MS-Office Home & Student 2013 for $100 less
but I found that OpenOffice isn't 100% compatible in the Excel area so
that sounds like an ok deal. My hope is that it can drive two 27"
monitors but I guess I can always add in another graphics card if not.

Reason I am looking at these is that I absolutely positively do not want
any computer with Windows 8 in here and unfortunately that's what many
others come with.

I have spent too many hours this weekend tweaking the transient
response of a semi-hysteretic (we call it "hysterical") switchmode
constant-current source. There are about 8 interacting knobs to turn.
At 30 seconds per run, understanding the interactions is impossible.


That's exactly why I need all the speed I can get.


I want sliders on each of the part values, and I want to see the
waveforms change as I move the sliders, like they were trimpots on a
breadboard and I was looking at a scope. I need maybe 500 times the
compute power that I have now.


I use the .STEP command a lot, sometimes nested. Then I get multiple
sets of curve sets. But often I have to start it at night and see the
results the next morning. The nice thing in winter is that this
pre-heats the office.


Mike should code LT Spice to execute on a high-end video card.


There are so many variants of graphics cards that it would require tons
of work for Mike's team.

Just pick one of the Nvidia number cruncher monsters. There are C
compilers available.

You can buy an Nvidia "video" board with no video out; it's just a 200
or something core compute engine.

Maybe the SuperSpice guy is interested. There's money there.


--

John Larkin Highland Technology, Inc

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[Spehro Pefhany]

On Sun, 02 Nov 2014 07:25:49 -0800, the renowned Joerg
<news@analogconsultants.com> wrote:

Folks,

Need to spiff up my simulation speeds here. IIRC Mike Engelhardt stated
that the Intel i7 is a really good processor for LTSPice. According to
this it looks like the 4790 is the fastest of the bunch:

http://www.intel.com/content/www/us/en/processors/core/core-i7-processor.html

So, what do thee say, is the computer in the Costco link below a good
deal for LTSpice purposes?

http://www.costco.com/Dell-XPS-8700-Desktop-%7c-Intel-Core-i7-%7c-1GB-Graphics-%7c-Windows-7-Professional.product.100131208.html

It's also available without MS-Office Home & Student 2013 for $100 less
but I found that OpenOffice isn't 100% compatible in the Excel area so
that sounds like an ok deal. My hope is that it can drive two 27"
monitors but I guess I can always add in another graphics card if not.

Reason I am looking at these is that I absolutely positively do not want
any computer with Windows 8 in here and unfortunately that's what many
others come with.

Looks pretty decent for the money, 32G is a sensible amount of RAM-
the RAM speed might be a bit on the low side though. I only research
this stuff when I'm shopping for a new box, but that stands out.

You can get a 1T Samsung SSD for only about $450, which allows you to
leave the mechanical one it comes with (worth < $100) pristine and to
be returned with the computer if necessary (so no proprietary
information can escape).


Best regards,
Spehro Pefhany
--
"it's the network..." "The Journey is the reward"
speff@interlog.com Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog Info for designers: http://www.speff.com

 

[Tim Williams]

"John Larkin" <jlarkin@highlandtechnology.com> wrote in message
news:0nkc5aljhec5r36ptkoaqbt0a48ud2j5vo@4ax.com...

I want sliders on each of the part values, and I want to see the
waveforms change as I move the sliders, like they were trimpots on a
breadboard and I was looking at a scope. I need maybe 500 times the
compute power that I have now.

You can do that in Multisim, but it's single thread only, no faster than
any other simulator.

Tim

--
Seven Transistor Labs
Electrical Engineering Consultation
Website: http://seventransistorlabs.com

 

[Jeff Liebermann]

On Sun, 02 Nov 2014 09:53:42 -0800, Joerg <news@analogconsultants.com>
wrote:

>Any guess how long MS will support Win-7?

Jan 13, 2015
<http://windows.microsoft.com/en-us/windows/lifecycle>
Extended support means that MS will gladly continue supporting Windoze
7 if you throw money at them. MS is following the Apple policy of
killing everything that's more than 5 years old (except in California,
which is 7 years):
<http://support.apple.com/en-us/ht1752>

>When it comes to PCs I am lazy

That's why I'm still in business to seperate such customers from their
money. Lazy is good for supporting my decadent and lavish lifestyle.

I just want to plug it in and go. Re-installing all my stuff takes
enough time already.

It's fairly easy to migrate programs and data from most anything to
Windoze 7.
<http://www.dummies.com/how-to/content/how-to-use-easy-transfer-in-windows-7.html>
<http://windows.microsoft.com/en-us/windows7/products/features/windows-easy-transfer>
However, Easy Transfer disappeared in Windoze 8.1, so you're either on
your own or purchase a 3rd part program to do the dirty work.

Before buying anything, I suggest you try LTspice on the new machine.
This is VERY easy with LTspice which doesn't use the registry or
require admin rights. Just copy the files to a flash drive and it
should work.

I am quite sure Costco will not let me do this

It's easier to obtain forgiveness than permission.

Are you really going to spend $1,300+ for a new machine without even a
test drive? Costco has a very good return policy, but you still have
to haul it home, get it running, update, tweak, tune, and then try. If
unacceptable, you get to try and fit everything back into the original
box (which never seems to fit). This is not being lazy.





--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558

 

[Jan Panteltje]

On a sunny day (Sun, 02 Nov 2014 10:21:33 -0800) it happened John Larkin
<jlarkin@highlandtechnology.com> wrote in
<6erc5ahtpf0buuavb0fpbidbbeqi60v9r1@4ax.com>:

I'll have to tweak resistors and capacitors, and the cap values are
too big for variable capacitors. And, as noted, it would be hard to
instrument.

Here's the current output when the load voltage steps from about 0.5
to 3 volts.

https://dl.dropboxusercontent.com/u/53724080/Circuits/Current_Sources/Hysterical_A1.jpg

I want it as flat as possible.

It looks a bit under-compensated, as if it takes time for it to respond to a load change.
It can probably never be faster than the /\/\/ cycles, but should be possible bette rthan this.
Diff part..

The fast ripple is the basic 1.5 MHz switcher frequency. The various
whoopie-doos are from loop dynamics and the chain of progressively
smaller, bias-tee-like damped inductors between the switcher and the
load. The constant-current hysterical switcher is, natively, about 4
or so orders of magnitude too slow for my application.

Everything interacts with everything else; it's like tuning a big LC
filter by hand, never a fun thing to do. Spice helps me acquire at
least some instincts for tuning. Maybe I can fix the cap values and
tune only resistors on the breadboard.

Rob, one of my guys, has a fierce Linux computer just for sims and
FPGA compiles, and he knows how to do automatic iterative parts value
tweaking in a loop around Spice. Maybe he can set up the problem and
run it for a couple of days or weeks.

I could probably step each of the six most important values, maybe 4
steps each, and pick the best waveform. That would be 4096 sims, about
60 hours of computing on my PC.

I have never been a PID guy, really, I have a simlilar problem here with frequency stabilization.
been testing large part of the day, reading many papers, got things working,
got severely pissed with Analog Devices (they provide PLL calculation soft that refuses to run under Linux wine,
even seems encryped, takes hour to un-encrypt, then cannot find DLLs), OK,
then I decided to do it in all software and not buy their chip.
I think the software solution can be better than their chip, anyways, experiment is fun
I have coded it, but really need to watch some movie to prevent electronics overdose.

 

[Jim Thompson]

On Sun, 02 Nov 2014 09:28:33 -0800, Joerg <news@analogconsultants.com>
wrote:

[snip]

I use the .STEP command a lot, sometimes nested. Then I get multiple
sets of curve sets. But often I have to start it at night and see the
results the next morning. The nice thing in winter is that this
pre-heats the office.


[snip]

Here's how I handle a system that has multiple, interrelated
adjustments. I just .STEP CASE from 1 to 4...

..PARAM VDD={ROW1+ROW2+ROW3+ROW4}
+ B1V8={IF(CASE==3,1,0)+IF(CASE==4,1,0)}
+ S3V3={IF(CASE==2,1,0)+IF(CASE==4,1,0)}
+ CASE=4
+ PVDD=1 ; 0=TYP, -1=MIN, 1=MAX
ROW1={IF((CASE==1)&(PVDD==0),3.3,0)+IF((CASE==1)&(PVDD==-1),2.5,0)+IF((CASE==1)&(PVDD==1),3.6,0)}
ROW2={IF((CASE==2)&(PVDD==0),2.5,0)+IF((CASE==2)&(PVDD==-1),2.35,0)+IF((CASE==2)&(PVDD==1),2.75,0)}
ROW3={IF((CASE==3)&(PVDD==0),3.0,0)+IF((CASE==3)&(PVDD==-1),2.7,0)+IF((CASE==3)&(PVDD==1),3.3,0)}
ROW4={IF((CASE==4)&(PVDD==0),3.3,0)+IF((CASE==4)&(PVDD==-1),3.0,0)+IF((CASE==4)&(PVDD==1),3.6,0)}

...Jim Thompson
--
| James E.Thompson | mens |
| Analog Innovations | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| San Tan Valley, AZ 85142 Skype: skypeanalog | |
| Voice480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |

I love to cook with wine. Sometimes I even put it in the food.

 

[rickman]

On 11/2/2014 10:25 AM, Joerg wrote:

Folks,

Need to spiff up my simulation speeds here. IIRC Mike Engelhardt stated
that the Intel i7 is a really good processor for LTSPice. According to
this it looks like the 4790 is the fastest of the bunch:

http://www.intel.com/content/www/us/en/processors/core/core-i7-processor.html

So, what do thee say, is the computer in the Costco link below a good
deal for LTSpice purposes?

http://www.costco.com/Dell-XPS-8700-Desktop-%7c-Intel-Core-i7-%7c-1GB-Graphics-%7c-Windows-7-Professional.product.100131208.html

Hard to say. The devil is in the details. I have an i7 and I'm not
convinced it is much better if any than other CPUs for most tasks. One
problem with the quad core is that that each core still runs at the same
speed as a dual core or even slower due to the contention for memory
bandwidth. That is why I got a laptop with separate graphic memory.
But overall I don't see big speed improvements. I'd be willing to bet
you won't see a huge difference between this machine and one costing a
few hundred dollars less. BTW, do you really need a new monitor? I
expect you can save a couple hundred more by getting a unit without
monitor.

It's also available without MS-Office Home & Student 2013 for $100 less
but I found that OpenOffice isn't 100% compatible in the Excel area so
that sounds like an ok deal. My hope is that it can drive two 27"
monitors but I guess I can always add in another graphics card if not.

I use LibreOffice which is the same package from the developers who
jumped ship at Oracle to continue development of OpenOffice the way they
think is best. I don't see compatibility issues and 90% of what I use
office for is the spreadsheet.

Reason I am looking at these is that I absolutely positively do not want
any computer with Windows 8 in here and unfortunately that's what many
others come with.

Get something tailor made. You can get both the best machine and the
cheapest that way. After all, they are all made from the same parts.
It is just a question of who puts them together.

--

Rick

 

On Sun, 02 Nov 2014 07:25:49 -0800, Joerg <news@analogconsultants.com>
wrote:

Need to spiff up my simulation speeds here. IIRC Mike Engelhardt stated
that the Intel i7 is a really good processor for LTSPice. According to
this it looks like the 4790 is the fastest of the bunch:

http://www.intel.com/content/www/us/en/processors/core/core-i7-processor.html

So, what do thee say, is the computer in the Costco link below a good
deal for LTSpice purposes?

http://www.costco.com/Dell-XPS-8700-Desktop-%7c-Intel-Core-i7-%7c-1GB-Graphics-%7c-Windows-7-Professional.product.100131208.html

How well does LTSpice spread simulation cycles between multiple
processors and multiple cores (Hyperthreading etc.) ?

 

[rickman]

On 11/2/2014 12:53 PM, Joerg wrote:

Jeff Liebermann wrote:
On Sun, 02 Nov 2014 07:25:49 -0800, Joerg <news@analogconsultants.com
wrote:

LTspice benchmark on various machines:
http://fetting.se/images/PC%20Speed%20Benchmark%20running%20LTspice%20circuits.pdf

Reason I am looking at these is that I absolutely positively do not want
any computer with Windows 8 in here and unfortunately that's what many
others come with.

Windoze 8.1 can be made semi-tolerable by putting the start menu back
in and making it look like Windoze 7.
http://www.classicshell.net
I've been installing it on all my customers Windoze 8.1 machines and
have had no complaints or problems. If you like wiggly icons on the
Windoze 8.1 start screen, you can do <Shift><Start>.


Too much risk. I've heard that running legacy software is tough in Win-8
but Win-7 can mostly do it. Not as good as XP.

What legacy software? I have Windows 8 and I'm not having problems
running anything I ran on my old Vista laptop.

When it comes to PCs I am lazy

I just want to plug it in and go. Re-installing all my stuff takes
enough time already.

I hear you. The big problem I had with setting up my Win 8 laptop was
that a lot of the freeware has become burdened with ads, toolbars and
other malware to the point I'm not willing to use it.

Before buying anything, I suggest you try LTspice on the new machine.
This is VERY easy with LTspice which doesn't use the registry or
require admin rights. Just copy the files to a flash drive and it
should work.

I am quite sure Costco will not let me do this

You can try finding the computer salesperson in the store. They are
limited by store policy of course, but I have met a few who were very
willing to help as best they could.

One catch. LTspice saves its preferences to:
C:\windows\scad3.ini
which has to be writeable. The fix is to use the
-ini <path
command line switch, which will:
Specify an .ini file to use other than %WINDIR%\scad3.ini
http://ltwiki.org/LTspiceHelp/LTspiceHelp/Command_Line_Switches.htm

I need to note this somewhere. Writing to the Windows directory is a
*very* bad idea. I can't tell you how many developers do all sorts of
things they aren't supposed to under windows. That is the actual cause
of many problems people have running older software under Windows. They
don't listen to the people providing them with the OS!

--

Rick

 

[Carl Ijames]

"Joerg" wrote in message news:cbn7m2Fla99U1@mid.individual.net...

Joerg wrote:

Carl Ijames wrote:
Don't know about computation speed, but this link says the video card
will
drive 3 monitors:
http://www.geforce.com/hardware/desktop-gpus/geforce-gt-720/specifications.
Looking at Dell's site I don't see any mention of expansion slots, and
looking at the one picture with the cover off I really can't see any
sockets
beyond the video card, so if any further expansion is important you need
to
ask Dell for clarification.


Looks like you are right:

http://www.dell.com/ed/business/p/xps-8700/pd
http://core0.staticworld.net/images/article/2013/07/1253541_sr-1160-100047019-orig.jpg
http://www.pcworld.com/article/2047487/dell-xps-8700-special-editions-review-a-little-less-performance-for-a-lot-less-cash.html

Quote "There's only one PCIe x16 slot, which means you won't be able to
add a second video card to take advantage of Nvidia's SLI technology".

No slots. There's one more card in the bottom, not sure what that is.
But if the video can drive three monitors it should be fine, I never
added any cards to my current PC either.

Only question is, how can one connect two regular OPC monitors to this?

http://www.geforce.com/hardware/desktop-gpus/geforce-gt-720/product-images

--
Regards, Joerg

http://www.analogconsultants.com/
==============================================================================

That's the gotcha, you need a different cable to each monitor. I'm no
expert, but I have an ATI Radeon HD5670 that supports 3 monitors and has the
same three connectors so I have my main monitor connected with the DVI-D
cable and my TV set connected with HDMI so I can watch video files on the TV
(I have a Hauppage tuner card and use my pc as dvr). I don't use the third
output. The installation guide shows pictures of adaptors from hdmi and
svga to dvd-d so if you used those you could use all dvd-I monitors, but
I've never looked for them so can't give any advice there. There might be a
lower max resolution on the svga output. The video driver lets me choose
which monitor is which and how to arrange them., so that is painless.

-----
Regards,
Carl Ijames

 

[DecadentLinuxUserNumeroUn]

On Sun, 02 Nov 2014 07:25:49 -0800, Joerg <news@analogconsultants.com>
Gave us:

Need to spiff up my simulation speeds here. IIRC Mike Engelhardt stated
that the Intel i7 is a really good processor for LTSPice. According to
this it looks like the 4790 is the fastest of the bunch:

http://www.intel.com/content/www/us/en/processors/core/core-i7-processor.html

That actually looks like a really nice deal. You can't go wrong.
Mine cost me $2500 and took over a year to build as budgets tightened.

I'll bet my 3930K beats it though.

The newer fabs have higher GHz rates, but are not as fast as their
first series were, which the smaller fabs replaced..

I have 6x2 cores, and I don't even know if they do that any more
except on Xeons.

I scream past all the benchmarks. Two more cores really makes a
difference. I beat machines pegging faster raw "speeds" all the time.
Mainly because they only have 4x2 cores.

Not cheap. The i7-3930K was $695, and the X79 Mobo under it was $400.
The 32GB RAM was not cheap either for 2133MHz, And that was before the
2400MHz stuff appeared. I can still upgrade the GPU and the RAM and get
even faster.

Since I cannot afford to put $1000 into a Titan video card, I miss on
a few benchmarks with my $250 GTX650.

 

[DecadentLinuxUserNumeroUn]

On Sun, 02 Nov 2014 09:46:45 -0800, Jeff Liebermann <jeffl@cruzio.com>
Gave us:

The Dell XPS 8700 seems like a nice machine. However, if you want
performance, I suggest you look at an SSD drive for the OS.

Even better than those are the mSATA drives and now, the best... the
M.2 drives.

Not much bigger than a couple of air mail stamps (I date myself).

Way faster than the 2.5" form factor SSD "laptop drive" replacement
family.

Even an mSATA drive in a USB3 enclosure/interface plugged into a USB 2
port boots up faster. As you can see, I am impressed.
You would be as well.

I would look into M.2 drives as they are the hot, emerging storage
tech right now, and they DO make a difference.

Even with your SSD variety, machines scream in benchmarks.

And this is OLD stuff. Things are way faster now than even that.

http://www.youtube.com/watch?v=eULFf6F5Ri8

 

[Jan Panteltje]

On a sunny day (Sun, 2 Nov 2014 12:09:54 -0800 (PST)) it happened Lasse
Langwadt Christensen <langwadt@fonz.dk> wrote in
<dc45b7d4-b977-43ae-bb08-63635c4739b2@googlegroups.com>:

>VGA is not much use, but unless you want to watch something from Hollywood=

VGA is still very useful,
I have one HD monitor with an extra VGA to a PC at the other side of the room.
Its faster than ssh -Y and does not load the network.
It displays lots of technical stuff that I run remote via wireless keyboard and mouse on that PC.

 

[John Larkin]

On Sun, 02 Nov 2014 19:02:35 GMT, Jan Panteltje
<pNaonStpealmtje@yahoo.com> wrote:

On a sunny day (Sun, 02 Nov 2014 10:21:33 -0800) it happened John Larkin
jlarkin@highlandtechnology.com> wrote in
6erc5ahtpf0buuavb0fpbidbbeqi60v9r1@4ax.com>:

I'll have to tweak resistors and capacitors, and the cap values are
too big for variable capacitors. And, as noted, it would be hard to
instrument.

Here's the current output when the load voltage steps from about 0.5
to 3 volts.

https://dl.dropboxusercontent.com/u/53724080/Circuits/Current_Sources/Hysterical_A1.jpg

I want it as flat as possible.

It looks a bit under-compensated, as if it takes time for it to respond to a load change.
It can probably never be faster than the /\/\/ cycles, but should be possible bette rthan this.
Diff part..

I'm switching at 1.5 MHz and I want constant current for load steps in
the picosecond time domain.

The fast ripple is the basic 1.5 MHz switcher frequency. The various
whoopie-doos are from loop dynamics and the chain of progressively
smaller, bias-tee-like damped inductors between the switcher and the
load. The constant-current hysterical switcher is, natively, about 4
or so orders of magnitude too slow for my application.

Everything interacts with everything else; it's like tuning a big LC
filter by hand, never a fun thing to do. Spice helps me acquire at
least some instincts for tuning. Maybe I can fix the cap values and
tune only resistors on the breadboard.

Rob, one of my guys, has a fierce Linux computer just for sims and
FPGA compiles, and he knows how to do automatic iterative parts value
tweaking in a loop around Spice. Maybe he can set up the problem and
run it for a couple of days or weeks.

I could probably step each of the six most important values, maybe 4
steps each, and pick the best waveform. That would be 4096 sims, about
60 hours of computing on my PC.


I have never been a PID guy, really, I have a simlilar problem here with frequency stabilization.
been testing large part of the day, reading many papers, got things working,
got severely pissed with Analog Devices (they provide PLL calculation soft that refuses to run under Linux wine,
even seems encryped, takes hour to un-encrypt, then cannot find DLLs), OK,
then I decided to do it in all software and not buy their chip.
I think the software solution can be better than their chip, anyways, experiment is fun
I have coded it, but really need to watch some movie to prevent electronics overdose.

The switcher is hysteretic, so there's no PID loop. Being hysteretic
means there's no loop compensation to get just-right, and it also
means that the switcher is fast and has no memory of the past; every
switch cycle stands on its own.

I tried a true hysteretic switcher: current sense resistor, diff gain,
schmitt comparator, mosfet driver. That works, but the frequency
varies all over the place, which has bad side effects. We had a
brainstorm meeting and came up with the hysterical converter. A dflop
drives the synchronous switching fets. We clock it ON at 1.5 MHz and
let the sense current resistor/amp/comparator clear it OFF when the
current hits the setpoint. That results in fixed-frequency PWM, but
with no PID or other memory in the loop. (I'm sure the idea has been
invented many times before.)

By moving a couple of wires, that can be converted to a first-order
delta-sigma loop. It's fun, but has a lot more current ripple than the
clocked hysterical converter.

The current-sense resistor has to be small to keep power dissipation
down, so we need a ton of differential gain, and there will be volts
of 1.5 MHz triangular common-mode junk on the resistor. A MiniCircuits
balun seems (in sim) to cut the common-mode noise down by about 20:1.

Like all RF folks, their parts are grossly underspecified. If you want
to know the winding and leakage inductances of their parts, you have
to measure them. RF is more like plumbing than engineering.

What, me ramble?


--

John Larkin Highland Technology, Inc

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

 

[Jan Panteltje]

On a sunny day (Sun, 02 Nov 2014 12:39:52 -0800) it happened John Larkin
<jlarkin@highlandtechnology.com> wrote in
<6b4d5al4b453j1af426p57gpspuvb3gagj@4ax.com>:

On Sun, 02 Nov 2014 19:02:35 GMT, Jan Panteltje
pNaonStpealmtje@yahoo.com> wrote:

On a sunny day (Sun, 02 Nov 2014 10:21:33 -0800) it happened John Larkin
jlarkin@highlandtechnology.com> wrote in
6erc5ahtpf0buuavb0fpbidbbeqi60v9r1@4ax.com>:

I'll have to tweak resistors and capacitors, and the cap values are
too big for variable capacitors. And, as noted, it would be hard to
instrument.

Here's the current output when the load voltage steps from about 0.5
to 3 volts.

https://dl.dropboxusercontent.com/u/53724080/Circuits/Current_Sources/Hysterical_A1.jpg

I want it as flat as possible.

It looks a bit under-compensated, as if it takes time for it to respond to a load change.
It can probably never be faster than the /\/\/ cycles, but should be possible bette rthan this.
Diff part..

I'm switching at 1.5 MHz and I want constant current for load steps in
the picosecond time domain.

Yes, OK, then the switcher wil have to swicth off early, or you need to add some analog series regulator.

The fast ripple is the basic 1.5 MHz switcher frequency. The various
whoopie-doos are from loop dynamics and the chain of progressively
smaller, bias-tee-like damped inductors between the switcher and the
load. The constant-current hysterical switcher is, natively, about 4
or so orders of magnitude too slow for my application.

Everything interacts with everything else; it's like tuning a big LC
filter by hand, never a fun thing to do. Spice helps me acquire at
least some instincts for tuning. Maybe I can fix the cap values and
tune only resistors on the breadboard.

Rob, one of my guys, has a fierce Linux computer just for sims and
FPGA compiles, and he knows how to do automatic iterative parts value
tweaking in a loop around Spice. Maybe he can set up the problem and
run it for a couple of days or weeks.

I could probably step each of the six most important values, maybe 4
steps each, and pick the best waveform. That would be 4096 sims, about
60 hours of computing on my PC.


I have never been a PID guy, really, I have a simlilar problem here with frequency stabilization.
been testing large part of the day, reading many papers, got things working,
got severely pissed with Analog Devices (they provide PLL calculation soft that refuses to run under Linux wine,
even seems encryped, takes hour to un-encrypt, then cannot find DLLs), OK,
then I decided to do it in all software and not buy their chip.
I think the software solution can be better than their chip, anyways, experiment is fun
I have coded it, but really need to watch some movie to prevent electronics overdose.


The switcher is hysteretic, so there's no PID loop. Being hysteretic
means there's no loop compensation to get just-right, and it also
means that the switcher is fast and has no memory of the past; every
switch cycle stands on its own.

Yes, I have done that too...

I tried a true hysteretic switcher: current sense resistor, diff gain,
schmitt comparator, mosfet driver. That works, but the frequency
varies all over the place, which has bad side effects. We had a
brainstorm meeting and came up with the hysterical converter. A dflop
drives the synchronous switching fets. We clock it ON at 1.5 MHz and
let the sense current resistor/amp/comparator clear it OFF when the
current hits the setpoint. That results in fixed-frequency PWM, but
with no PID or other memory in the loop. (I'm sure the idea has been
invented many times before.)

This is what I do here:
http://panteltje.com/panteltje/pic/pwr_pic/
there is a current sense transformer, but also a current sense resistor that triggers a comparator
that switches of the MOSFET.
The problem is the energy in the inductors and capacitors will NOT suddenly disapear,
it takes time.
In the sixties I did that (much slower in those days) with an analog transistor series regulator after a thyristor switcher
where the thyristor switcher on average held the voltage over the series transistor as low as possible (just above the ripple)
that was for the telco here.
I dunno what you dissipation issues would be.

By moving a couple of wires, that can be converted to a first-order
delta-sigma loop. It's fun, but has a lot more current ripple than the
clocked hysterical converter.

The current-sense resistor has to be small to keep power dissipation
down, so we need a ton of differential gain, and there will be volts
of 1.5 MHz triangular common-mode junk on the resistor. A MiniCircuits
balun seems (in sim) to cut the common-mode noise down by about 20:1.

Like all RF folks, their parts are grossly underspecified. If you want
to know the winding and leakage inductances of their parts, you have
to measure them. RF is more like plumbing than engineering.

What, me ramble?

Me too, I am watching 'red' with whatshisname.

 

[Martin Riddle]

On Sun, 2 Nov 2014 15:02:04 -0500, "Carl Ijames"
<carl.ijamesXX@XXverizon.net> wrote:

"Joerg" wrote in message news:cbn7m2Fla99U1@mid.individual.net...

Joerg wrote:
Carl Ijames wrote:
Don't know about computation speed, but this link says the video card
will
drive 3 monitors:
http://www.geforce.com/hardware/desktop-gpus/geforce-gt-720/specifications.
Looking at Dell's site I don't see any mention of expansion slots, and
looking at the one picture with the cover off I really can't see any
sockets
beyond the video card, so if any further expansion is important you need
to
ask Dell for clarification.


Looks like you are right:

http://www.dell.com/ed/business/p/xps-8700/pd
http://core0.staticworld.net/images/article/2013/07/1253541_sr-1160-100047019-orig.jpg
http://www.pcworld.com/article/2047487/dell-xps-8700-special-editions-review-a-little-less-performance-for-a-lot-less-cash.html

Quote "There's only one PCIe x16 slot, which means you won't be able to
add a second video card to take advantage of Nvidia's SLI technology".

No slots. There's one more card in the bottom, not sure what that is.
But if the video can drive three monitors it should be fine, I never
added any cards to my current PC either.


Only question is, how can one connect two regular OPC monitors to this?

http://www.geforce.com/hardware/desktop-gpus/geforce-gt-720/product-images

DVI? You can get a Display port -> DVI adaptor. Dell sends them out
with their business laptops.

Cheers

 

[Jeff Liebermann]

On Sun, 02 Nov 2014 14:56:04 -0500, rickman <gnuarm@gmail.com> wrote:

On 11/2/2014 12:53 PM, Joerg wrote:
Too much risk. I've heard that running legacy software is tough in Win-8
but Win-7 can mostly do it. Not as good as XP.

What legacy software? I have Windows 8 and I'm not having problems
running anything I ran on my old Vista laptop.

I recently awarded myself a short vacation in honor my burning a huge
amount of time getting old software to run nicely on Windoze 8.1.
Specifically, the DOS versions of various fiduciary programs dating
1996 through 2002, which the customer insisted had to run even though
later versions worked just fine. The problem was that the tax rules
and tables all changed over the years and they wanted the original
versions. I ended up running them under DOSbox, which was originally
designed to run ancient games, but works equally well with ancient
business applications:
<http://www.dosbox.com/status.php?show_status=1>
I also tried them under VMware and VirtualBox, both of which worked
nicely, but DOSbox is easier and faster.

Another horror was Office 2003 on Windoze 8.1. It installs, updates,
loads, and looks like it might work, but eventually crashes. All I
really needed was Outlook 2003, but that would hang after polling for
mail a few times. I probably could have figured out the problem, but
convinced the customer that Mozilla Thunderbird would be a suitable
option.

Then, there's WordPerfect 12 which I think was introduced in 2002.
Amazingly, it worked 99%. However, the 1% was fatal. Windoze file
association would not start WP12 if I double clicked on a WPD file (or
any of the other WP files). It took a while to figure out that WP12
was trying to use an ancient ODBC version, which required that WP12
beg permission of the Windoze security abomination before it would
condescend to even supply an error message. Fixed by running WP12 as
administrator, which by passes most of the security mess.

I guess the moral here is to not try to run 12+ year old software on
Windoze 8.1. My mistake was assuming that since all the
aforementioned software ran just fine in Windoze 7, the new and
improved Windoze 8.1 couldn't possibly break something that already
worked so well.

One catch. LTspice saves its preferences to:
C:\windows\scad3.ini
which has to be writeable. The fix is to use the
-ini <path
command line switch, which will:
Specify an .ini file to use other than %WINDIR%\scad3.ini
http://ltwiki.org/LTspiceHelp/LTspiceHelp/Command_Line_Switches.htm

I need to note this somewhere. Writing to the Windows directory is a
*very* bad idea.

It was standard procedure in Windoze 3.1, where almost all
applications dropped pick_a_name.ini files in the C:\Windows\
directory. I do have to admit it was handy as the files were easy to
find and save. The new and improved versions of Windoze hide these
config files in either the registry, or bury them 5 directory layers
deep, where few can find them without specialized tools or inside
information.

I can't tell you how many developers do all sorts of
things they aren't supposed to under windows. That is the actual cause
of many problems people have running older software under Windows. They
don't listen to the people providing them with the OS!

LTspice (aka SwitcherCAD) is a rather old program, with many of the
traditions of Windoze 3.1 still present. If you don't like that, try
running some of the various NEC antenna modeling programs, that still
use the terms "card" and "deck" from the Hollerith punch card era. The
common mantra is the same everywhere... if it works, don't touch it.



Looking at the benchmarks at:
<http://fetting.se/images/PC%20Speed%20Benchmark%20running%20LTspice%20circuits.pdf>
my Dell Optiplex 755 clunker runs the 3 benchmarks at:
14.5 7.6 3.6
If I upgrade to the fastest machine on the list:
4.0 2.9 1.0
or roughly 3 times faster. Might be worth $1200+.

The database is at:
<https://groups.yahoo.com/neo/groups/LTspice/database/2/edit>
and shows no Windoze 8.1 benchmarks and no SSD, so those will remain
an unknown. The benchmark files and instructions are at:
<https://groups.yahoo.com/neo/groups/LTspice/files/%20Examples/Benchmark/>
If you run the benchmark, be sure to add it to the database.



--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558

 

[Jeff Liebermann]

On Sun, 02 Nov 2014 14:28:57 -0800, Jeff Liebermann <jeffl@cruzio.com>
wrote:

Looking at the benchmarks at:
http://fetting.se/images/PC%20Speed%20Benchmark%20running%20LTspice%20circuits.pdf
my Dell Optiplex 755 clunker runs the 3 benchmarks at:
14.5 7.6 3.6
If I upgrade to the fastest machine on the list:
4.0 2.9 1.0
or roughly 3 times faster. Might be worth $1200+.

I ran the benchmarks on my Dell Optiplex 960 (XP-SP3, E8500 3.2GHz,
4GB RAM, 1TB drive).
10.0 11.2 2.5
So, $1200 will get me about 2.5 times faster (ignoring some kind of
problem with the Mic2 test).

The database is at:
https://groups.yahoo.com/neo/groups/LTspice/database/2/edit
and shows no Windoze 8.1 benchmarks and no SSD, so those will remain
an unknown. The benchmark files and instructions are at:
https://groups.yahoo.com/neo/groups/LTspice/files/%20Examples/Benchmark/
If you run the benchmark, be sure to add it to the database.

--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558

 

[Jeff Liebermann]

On Sun, 02 Nov 2014 12:27:52 -0800, DecadentLinuxUserNumeroUno
<DLU1@DecadentLinuxUser.org> wrote:

Even better than those are the mSATA drives and now, the best... the
M.2 drives.
Not much bigger than a couple of air mail stamps (I date myself).
Way faster than the 2.5" form factor SSD "laptop drive" replacement
family.

The SSD drive I recommended comes in both SATA3 and mSATA
configurations:
<http://www.samsung.com/global/business/semiconductor/minisite/SSD/global/html/ssd840evo/overview_mSATA.html>
<http://www.samsung.com/global/business/semiconductor/minisite/SSD/global/html/ssd840evo/overview.html>
The specs look fairly close:
<http://www.samsung.com/global/business/semiconductor/minisite/SSD/global/html/ssd840evo/specifications.html>

I have an older mSATA drive in my Acer C720 running Linux. Very very
very very fast, but I haven't compared it with a SATA3 drive.

--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558