Forums

Sub circuit in Ltspice.

Started by Jamie October 6, 2012
TheQuickBrownFox wrote:
> > On Sun, 07 Oct 2012 20:04:18 -0400, Phil Hobbs > <pcdhSpamMeSenseless@electrooptical.net> wrote: > > >TheQuickBrownFox wrote: > >> > >> On Sun, 07 Oct 2012 16:09:14 -0400, Phil Hobbs > >> <pcdhSpamMeSenseless@electrooptical.net> wrote: > >> > >> > > >> >How much heavy duty floating point do you do in a day? Have you done > >> >any actual benchmarks for floating point performance between Intel and > >> >AMD? > >> > > >> > > >> >Cheers > >> > > >> >Phil Hobbs > >> > >> My video card does better performance than either. > >> > >> As CPUs go, the winner in that race would be, even after being four > >> years old, the Cell CPU. > >> > >> I am sorry, but a 20 ms spice sim calculation and a 17 ms calc of the > >> same sim is not enough to make me decide for the latter. > >> > >> I have a feeling that modern Intel CPUs beat AMD. > >> > >> THEN there is the chipset, and supporting Hdw thing too. > > > >Your video card can't address 32 GB of memory. What's its maximum > >sustained bandwidth to main memory? FDTD cycles through all of memory > >twice per time step. > > > >Cheers > > > >Phil Hobbs > > I'm sorry, but you do not now, nor will you ever have in the future, a > sim circuit where the array requires a 32 GB RAM space to be ran in. > > What a ridiculous claim.
I didn't say it was a circuit, I said it was FDTD. JFGI. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
On Sun, 07 Oct 2012 17:16:11 -0700, TheQuickBrownFox
<thequickbrownfox@overthelazydog.org> wrote:

>On Sun, 07 Oct 2012 20:04:18 -0400, Phil Hobbs ><pcdhSpamMeSenseless@electrooptical.net> wrote: > >>TheQuickBrownFox wrote: >>> >>> On Sun, 07 Oct 2012 16:09:14 -0400, Phil Hobbs >>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>> >>> > >>> >How much heavy duty floating point do you do in a day? Have you done >>> >any actual benchmarks for floating point performance between Intel and >>> >AMD? >>> > >>> > >>> >Cheers >>> > >>> >Phil Hobbs >>> >>> My video card does better performance than either. >>> >>> As CPUs go, the winner in that race would be, even after being four >>> years old, the Cell CPU. >>> >>> I am sorry, but a 20 ms spice sim calculation and a 17 ms calc of the >>> same sim is not enough to make me decide for the latter. >>> >>> I have a feeling that modern Intel CPUs beat AMD. >>> >>> THEN there is the chipset, and supporting Hdw thing too. >> >>Your video card can't address 32 GB of memory. What's its maximum >>sustained bandwidth to main memory? FDTD cycles through all of memory >>twice per time step. >> >>Cheers >> >>Phil Hobbs > > I'm sorry, but you do not now, nor will you ever have in the future, a >sim circuit where the array requires a 32 GB RAM space to be ran in. > > What a ridiculous claim.
And as far as bandwidth goes, the PCIe bus makes for fast, HUGE data transfers between video and main RAM as fast as it gets. They use video cards cores for protein modeling, and seti comm data parsing. I think they can handle a petty OR a complex circuit sim.
John Larkin wrote:
> > That's a lot of money for playing video games.
What else can dimmie do? No one would hire him.
Am 08.10.2012 00:40, schrieb Jim Thompson:
> On Mon, 08 Oct 2012 00:18:50 +0200, Gerhard Hoffmann <dk4xp@arcor.de> > > What operating system, Win7 ? > > How do you like it? I fear biting that bullet :-(
It came with Win7, I now run Win7 / xubuntu 12.04 dual boot. Win7 is mostly ignored. Xilinx/Sigasi FPGA stuff runs natively under Linux, most other CAD runs in the same XP virtual box that I use at home on the workstation. Just 'import' a 50 Gig file to the virtual box. No ado with licenses. It _is_ the same machine. The C: partition contains ONLY installed programs; D: of the XP can be seen as D: on win7 and as /d on Linux. The machine has a drawback: it is quite heavy and can generate some heat if one asks for it. regards, Gerhard
On Mon, 08 Oct 2012 08:46:14 +0200, Gerhard Hoffmann <dk4xp@arcor.de>
wrote:

>Am 08.10.2012 00:40, schrieb Jim Thompson: >> On Mon, 08 Oct 2012 00:18:50 +0200, Gerhard Hoffmann <dk4xp@arcor.de> >> >> What operating system, Win7 ? >> >> How do you like it? I fear biting that bullet :-( > >It came with Win7, I now run Win7 / xubuntu 12.04 dual boot. >Win7 is mostly ignored.
Yeah, as shitty as the last two Ubuntu releases have been, I am sure you are spending most of your time learning their 'new' 100% pathetic, "Unity" 'desktop'. Biggest mistake they ever made. Good luck with that. XUBUNTU is far better.
> >Xilinx/Sigasi FPGA stuff runs natively under Linux,
Big whoop. Under Linux the 'stuff' runs at whatever run time it takes to get it done. Under Windows 7, non-natively, it takes a whole 5 ms longer. Big deal.
> most other >CAD runs in the same XP virtual box that I use at home on >the workstation. Just 'import' a 50 Gig file to the >virtual box. No ado with licenses. It _is_ the same machine.
I laugh at some of the running around you dorks who "hate Windows" go out of your way to do. Absolutely pathetic.
>The C: partition contains ONLY installed programs; D: of the XP >can be seen as D: on win7 and as /d on Linux.
My machine has about 4TB of HD space, under about 15 partitions.
>The machine has a drawback: it is quite heavy and can generate >some heat if one asks for it.
Sounds like a pre-multi-core POS.
TheQuickBrownFox wrote:
> > On Sun, 07 Oct 2012 17:16:11 -0700, TheQuickBrownFox > <thequickbrownfox@overthelazydog.org> wrote: > > >On Sun, 07 Oct 2012 20:04:18 -0400, Phil Hobbs > ><pcdhSpamMeSenseless@electrooptical.net> wrote: > > > >>TheQuickBrownFox wrote: > >>> > >>> On Sun, 07 Oct 2012 16:09:14 -0400, Phil Hobbs > >>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > >>> > >>> > > >>> >How much heavy duty floating point do you do in a day? Have you done > >>> >any actual benchmarks for floating point performance between Intel and > >>> >AMD? > >>> > > >>> > > >>> >Cheers > >>> > > >>> >Phil Hobbs > >>> > >>> My video card does better performance than either. > >>> > >>> As CPUs go, the winner in that race would be, even after being four > >>> years old, the Cell CPU. > >>> > >>> I am sorry, but a 20 ms spice sim calculation and a 17 ms calc of the > >>> same sim is not enough to make me decide for the latter. > >>> > >>> I have a feeling that modern Intel CPUs beat AMD. > >>> > >>> THEN there is the chipset, and supporting Hdw thing too. > >> > >>Your video card can't address 32 GB of memory. What's its maximum > >>sustained bandwidth to main memory? FDTD cycles through all of memory > >>twice per time step. > >> > >>Cheers > >> > >>Phil Hobbs > > > > I'm sorry, but you do not now, nor will you ever have in the future, a > >sim circuit where the array requires a 32 GB RAM space to be ran in. > > > > What a ridiculous claim. > > And as far as bandwidth goes, the PCIe bus makes for fast, HUGE data > transfers between video and main RAM as fast as it gets. > > They use video cards cores for protein modeling, and seti comm data > parsing. I think they can handle a petty OR a complex circuit sim.
"As fast as it gets" isn't a number. The bottleneck in FDTD is main memory bandwidth, because you're cycling through all allocated memory many many times, so you get a lot of cache misses. Good caching will help with the main memory latency, because most accesses are to contiguous memory if you've coded it right. Did you look up FDTD? What do you think about it? Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
On Sun, 07 Oct 2012 18:56:37 -0400 Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote in Message id:
<507208A5.D8419D6@electrooptical.net>:

>Jim Thompson wrote: >> >> On Mon, 08 Oct 2012 00:18:50 +0200, Gerhard Hoffmann <dk4xp@arcor.de> >> wrote: >> >> >Am 07.10.2012 21:06, schrieb TheQuickBrownFox: >> > >> >> EVGA makes MODERN dual CPU motherboards. Supermicro is like Dell. It >> >> is two year old technology the moment you buy it. With this, even more >> >> than 2 years. >> > >> >I'm writing this in a hotel room in Delphi, Greece, on a 17 month old >> >Dell Precision M6600 17" laptop. Nobody else had 6 GBit/s SATA, USB-3, >> >16 or 32 GB RAM, Core I7 vpro and such a display at that time. >> >I have replaced the iron disk first with a OCZ Vortex3 SSD and now both >> >drives are 512 GB Samsung 830 SSDs. Plugged them in and they worked. >> >Both the OCZ and the Samsungs give about 500 MBytes/s transfer rate >> >in the laptop, about the same as a customer's CAD server monster RAID. >> >LT spice performance is somewhat better than twice that of the server. >> > >> >I have not yet tried the two SSDs as a RAID-0. >> >Plus, I'll be able to run CUDA code on the graphics card for >> >software defined radio. >> > >> >So, please don't tell us that Dell is >2 years back. >> > >> >regards, Gerhard >> >> What operating system, Win7 ? >> >> How do you like it? I fear biting that bullet :-( > >Works well for me, except that I have a few DOS programs that I still >use, which it won't run. (Freelance 4.0 is the biggest problem.) I >have an XP virtual machine on my Supermicro box to handle that sort of >stuff. > >Win7 also has an 'XP mode' feature, but it won't run on my standard >laptops.
Somebody mentioned this a while back, but in case anybody missed it. http://classicshell.sourceforge.net/index.html It gives Windows 7 (or 8) a more XP like feel.
On Sun, 07 Oct 2012 09:16:33 -0700, Joerg <invalid@invalid.invalid> =
wrote:

>Jim Thompson wrote: >> On Sun, 07 Oct 2012 07:29:30 -0700, Joerg <invalid@invalid.invalid> >> wrote: >>=20 >>> Jamie wrote: >>>> If I create a sub circuit in LTspice, will that circuit's content =
be
>>>> exported with the main file or do I need to include that along with =
the
>>>> file? >>>> I read the help file but it some what does not give me a fuzzy =
feeling..
>>>> >>> If you place the whole subcircuit text into your schematic as a SPICE >>> directive then it is included. That's how I always do it. Typically =
to
>>> the left or below so clients have a choice of printing it out along =
with
>>> the schematic or not. >>> >>> Much better than needing two files or having to load library parts. >>=20 >> What Joerg suggests is the equivalent of .INCLUDE in other Spice's. >> Slows down the loading of the input file. .LIB is faster and more >> efficient. >>=20 > >What I meant is .SUBCKT in the input file. I've never had any speed >issues, they load in a second or two. What does take forever is sims >with gapped transformers in there, leakage inductance and all that. And >then the office temp start to rise and rise and rise. > >I've got an Intel 1.6GHz dual core in there. Rumors have it that the >Intel i7 could be almost twice as fast. Have you heard and confirmations >in that direction?
Tell you what, let me get my racehorse running (3.6 GHz hex core, 16 GiB RAM, SSD) and we can do some timing tests. Not nearly as powerful as P.Hobbs fast box but should be plenty ok. ?-)
On Sun, 07 Oct 2012 16:09:14 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

>TheQuickBrownFox wrote: >>=20 >> On Sun, 07 Oct 2012 13:26:48 -0400, Phil Hobbs >> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>=20 >> >Jim Thompson wrote: >> >> >> >> On Sun, 07 Oct 2012 09:16:33 -0700, Joerg <invalid@invalid.invalid> >> >> wrote: >> >> >> >> >Jim Thompson wrote: >> >> >> On Sun, 07 Oct 2012 07:29:30 -0700, Joerg =
<invalid@invalid.invalid>
>> >> >> wrote: >> >> >> >> >> >>> Jamie wrote: >> >> >>>> If I create a sub circuit in LTspice, will that circuit's =
content be
>> >> >>>> exported with the main file or do I need to include that along=
with the
>> >> >>>> file? >> >> >>>> I read the help file but it some what does not give me a =
fuzzy feeling..
>> >> >>>> >> >> >>> If you place the whole subcircuit text into your schematic as a=
SPICE
>> >> >>> directive then it is included. That's how I always do it. =
Typically to
>> >> >>> the left or below so clients have a choice of printing it out =
along with
>> >> >>> the schematic or not. >> >> >>> >> >> >>> Much better than needing two files or having to load library =
parts.
>> >> >> >> >> >> What Joerg suggests is the equivalent of .INCLUDE in other =
Spice's.
>> >> >> Slows down the loading of the input file. .LIB is faster and =
more
>> >> >> efficient. >> >> >> >> >> > >> >> >What I meant is .SUBCKT in the input file. >> >> >> >> OK. I misread your intent. That's equivalent to a .LIB call... =
only
>> >> what you need is loaded. >> >> >> >> >I've never had any speed >> >> >issues, they load in a second or two. What does take forever is =
sims
>> >> >with gapped transformers in there, leakage inductance and all =
that. And
>> >> >then the office temp start to rise and rise and rise. >> >> >> >> Heats my office in winter ;-) >> >> >> >> > >> >> >I've got an Intel 1.6GHz dual core in there. >> >> >> >> That's what I have in my notebook (Thinkpad X61s): 1.60 GHz Intel >> >> Core2 Duo. It runs slower than my old Win2K machine (Analog3) with=
a
>> >> 2.20GHz AMD Athlon 64 and only 1G of RAM :-) >> >> >> >> >Rumors have it that the >> >> >Intel i7 could be almost twice as fast. Have you heard and =
confirmations
>> >> >in that direction? >> >> >> >> Since Analog1 (also 2.20GHz AMD Athlon 64, but WinXP Pro) crashed =
and
>> >> burned, I'm pondering what to replace it with. Mark/qrk is the >> >> resident expert on benchmarking. There's so much Intel malarkey =
out
>> >> there, it's difficult to know what would be best for a simulation >> >> machine. >> >> >> > >> >My main office machine is a nice Supermicro dual 8-core AMD Magny =
Cours
>> >with 32G of RAM and a nice RAID5 disc array. I bought it about a =
year
>> >ago for a bit under $4k. It runs CentOS 6.2 Linux, with kvm/qemu >> >virtual machines for XP/32 and Win7/64. >> > >> >LTspice flies. >> > >> >Cheers >> > >> >Phil Hobbs >>=20 >> It isn't "nice" if it doesn't have USB 3. also needs SATA 3, cause =
RAID
>> 5 on the old interface spec is no gain. >>=20 >> $4k??? Supermicro saw you coming. Or the guy you got to build it =
for
>> you. >>=20 >> I'd go with an EVGA dual XEOM MOBO and fill it with 6 or 12 way =
(core
>> pairs). Put all my money into the CPUs and MOBO. AMD mobos are all >> taking a hit these days, mainly because the idiots embraced and bought >> ATI. >>=20 >> Supermicro makes dual CPU mobos. Oh boy. They have always been =
hugely
>> overpriced and underfunctioned. >>=20 >> EVGA makes MODERN dual CPU motherboards. Supermicro is like Dell. =
It
>> is two year old technology the moment you buy it. With this, even =
more
>> than 2 years. >>=20 >> And the price difference is small enough that I stopped buying AMD 8 >> years ago. Intel Mobos and CPUs scream. > > > >How much heavy duty floating point do you do in a day? Have you done >any actual benchmarks for floating point performance between Intel and >AMD? > > >Cheers > >Phil Hobbs
=46or pretty much anyone here, that reduces to how much SPICE simulations= do you run. If it is a lot, AMD based machines may still have an advantage. Otherwise not much difference, most anything since a 1.4 GHz P4 is overkill unless wasted away by the OS and Desktop. ?-)
On Mon, 08 Oct 2012 10:30:57 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

>TheQuickBrownFox wrote: >>=20 >> On Sun, 07 Oct 2012 17:16:11 -0700, TheQuickBrownFox >> <thequickbrownfox@overthelazydog.org> wrote: >>=20 >> >On Sun, 07 Oct 2012 20:04:18 -0400, Phil Hobbs >> ><pcdhSpamMeSenseless@electrooptical.net> wrote: >> > >> >>TheQuickBrownFox wrote: >> >>> >> >>> On Sun, 07 Oct 2012 16:09:14 -0400, Phil Hobbs >> >>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >> >>> >> >>> > >> >>> >How much heavy duty floating point do you do in a day? Have you =
done
>> >>> >any actual benchmarks for floating point performance between =
Intel and
>> >>> >AMD? >> >>> > >> >>> > >> >>> >Cheers >> >>> > >> >>> >Phil Hobbs >> >>> >> >>> My video card does better performance than either. >> >>> >> >>> As CPUs go, the winner in that race would be, even after being =
four
>> >>> years old, the Cell CPU. >> >>> >> >>> I am sorry, but a 20 ms spice sim calculation and a 17 ms calc =
of the
>> >>> same sim is not enough to make me decide for the latter. >> >>> >> >>> I have a feeling that modern Intel CPUs beat AMD. >> >>> >> >>> THEN there is the chipset, and supporting Hdw thing too. >> >> >> >>Your video card can't address 32 GB of memory. What's its maximum >> >>sustained bandwidth to main memory? FDTD cycles through all of =
memory
>> >>twice per time step. >> >> >> >>Cheers >> >> >> >>Phil Hobbs >> > >> > I'm sorry, but you do not now, nor will you ever have in the =
future, a
>> >sim circuit where the array requires a 32 GB RAM space to be ran in. >> > >> > What a ridiculous claim. >>=20 >> And as far as bandwidth goes, the PCIe bus makes for fast, HUGE data >> transfers between video and main RAM as fast as it gets. >>=20 >> They use video cards cores for protein modeling, and seti comm data >> parsing. I think they can handle a petty OR a complex circuit sim. > >"As fast as it gets" isn't a number. > >The bottleneck in FDTD is main memory bandwidth, because you're cycling >through all allocated memory many many times, so you get a lot of cache >misses. Good caching will help with the main memory latency, because >most accesses are to contiguous memory if you've coded it right. > >Did you look up FDTD? What do you think about it? > >Cheers > >Phil Hobbs
I presume you are aware you are talking to AlwaysWrong. ?-)