Keyboard Shortcuts
ctrl + shift + ? :
Show all keyboard shortcuts
ctrl + g :
Navigate to a group
ctrl + shift + f :
Find
ctrl + / :
Quick actions
esc to dismiss
Likes
- LTspice
- Messages
Search
|
Re: Benchmarking the new LTspice
Mike claims that LTspice tests the hardware performance of each machine on which it runs and then decides for each simulation how many of the available cores and threads is worth the additional overhead.? It seems that Mike has learned how to reduce overhead so that running more cores has become very efficient for LTspice.
toggle quoted message
Show quoted text
The new Intel Xeon Phi Processor 7230 (16GB, 1.30 GHz) has 64 cores / 256 threads (and costs $4k!!).? GPUs have thousands of cores and are trending to becoming electronic simulation capable.? A lot of work has been going into understanding how to effectively use parallel processing (this includes making electronic brain-like neural networks - the positronic brain anyone?). ---In LTspice@..., <allanvv@...> wrote : Indeed, there are Xeons that have 20+ cores that you would be paying $10k+ for. However, I don't think you would be gaining much speed by splitting a simulation into even more threads, and in fact you would be losing it due to the overheads. So cores*clock speed might be the relevant factor, but upping the cores to 12 while keeping the cores*freq product the same makes your CPU cost way more for no performance gain. I originally did recommend getting a 6-8 core CPU but overclocking it regardless. You can actually set a different clock speed for 1 core use or 6 core use, since the limiting factor is TDP.
? |
|
Re: Benchmarking the new LTspice
Allan Wang
Indeed, there are Xeons that have 20+ cores that you would be paying $10k+ for. However, I don't think you would be gaining much speed by splitting a simulation into even more threads, and in fact you would be losing it due to the overheads. So cores*clock speed might be the relevant factor, but upping the cores to 12 while keeping the cores*freq product the same makes your CPU cost way more for no performance gain. I originally did recommend getting a 6-8 core CPU but overclocking it regardless. You can actually set a different clock speed for 1 core use or 6 core use, since the limiting factor is TDP. On Tue, Aug 16, 2016 at 2:30 PM, analogspiceman@... [LTspice] <LTspice@...> wrote:
|
|
Re: Benchmarking the new LTspice
Allan,
toggle quoted message
Show quoted text
What you wrote only applies to small simulations that generally run very fast anyway.? The typical large simulation that I run at work will max out all four cores and eight threads available to me, so for real work simulations, it is possible that by doubling the core and thread count, I might nearly double simulation speed.? It seems the trend for the design of newest processors has bifurcated into tradition designs (faster clock speeds and a only few cores) and those with dozen or even hundreds of cores.? It seems that Mike Engelhardt (LTspice's author) has become very proficient at writing code to take advantage of more cores. ---In LTspice@..., <allanvv@...> wrote : Your simulation speed will be directly proportional to the clock speed. The CPU speed is the bottleneck after all,? If your CPU turbo boosts to 3.8 GHz (one core), and you overclock to 4.4 GHz, then you should gain around 15%.
|
|
Re: Benchmarking the new LTspice
Allan Wang
Your simulation speed will be directly proportional to the clock speed. The CPU speed is the bottleneck after all,? If your CPU turbo boosts to 3.8 GHz (one core), and you overclock to 4.4 GHz, then you should gain around 15%. On Tue, Aug 16, 2016 at 5:03 AM, basier.philippe@... [LTspice] <LTspice@...> wrote:
|
|
Re: Calculation of the inductance of the amplitude of the first harmonic voltage and current amplitude
Hi.
toggle quoted message
Show quoted text
The fact that I am interested in the behavior of inductors and at the approach to saturation. Usually, lead ferrite losses at the induction B = 0.2Tesla. This is already affecting the core nonlinearity. I did modeling for N41 material. In the current transformer, according to my calculations, there is a decrease in gain with increasing current up to 100A by ~ 0.33%. This occurs due to nonlinearity. Bordodynov. 16.08.2016, 16:43, "Michael Peter Nekambuza Kiwanuka michael883575@... [LTspice]" <ltspice@...>: Hi, |
|
Re: DSBSC
Gunoi Nare wrote: ? ?"To my surprise I can not find a LTC multiplier." I would not use a real analog multiplier IC, if all I wanted was to simulate a double-sideband suppressed-carrier signal. (FYI, I am also not aware of a general purpose linear 4-quadrant multiplier made by LTC, similar to the ones Analog Devices makes.? The two companies' product lines do not totally overlap -- which is why ADI was interested in acquiring LTC.) Instead, use the multiplication function from a more primitive SPICE element. I think you have two main choices.? One is to use the Modulate or Modulate2, as Helmut suggested.? Yes, it does suppress the carrier, if the amplitude (AM) input is sinusoidal with zero offset.? The output amplitude follows the voltage of the AM input pin.? When that voltage goes negative, the output amplitude also goes negative (becomes inverted).? The net result is cancellation of the carrier component. The other choice, is to make a B behavioral source, with a formula that includes multiplication.? B-elements are very flexible, letting you make just about any kind of signal if you can express it in a formula. ?(You could instead use one of the traditional SPICE controlled sources = E, F, G, or H, but they are somewhat more difficult to use.) There is a third option, which would be to generate the two sideband signals directly, using two SINE sources and adding them together. Andy |
|
Re: Calculation of the inductance of the amplitude of the first harmonic voltage and current amplitude
John Woodgate
¿ªÔÆÌåÓýIn detailed analysis, the permeability of the core is a non-linear function of amplitude even at low amplitudes. 'Saturation' is at the H value at which the non-linearity becomes gross. ? With best wishes DESIGN IT IN! OOO ¨C Own Opinions Only J M Woodgate and Associates Rayleigh England ? Sylvae in aeternum manent. ? From: LTspice@... [mailto:LTspice@...]
Sent: Tuesday, August 16, 2016 2:42 PM To: ltspice@... Subject: RE: [LTspice] Calculation of the inductance of the amplitude of the first harmonic voltage and current amplitude ? ? Hi, ? I thought he Inductance of an inductor is independent of the amplitude of the driving signal assuming you don't saturate the core. ?Please correct me If I am wrong. ? Michael To: ltspice@... Vlad, |
|
Re: noiseless resistor
John Woodgate
¿ªÔÆÌåÓýThat number is Windows interpretation of the reciprocal of infinity. ? With best wishes DESIGN IT IN! OOO ¨C Own Opinions Only J M Woodgate and Associates Rayleigh England ? Sylvae in aeternum manent. ? From: LTspice@... [mailto:LTspice@...]
Sent: Tuesday, August 16, 2016 2:53 PM To: LTspice@... Subject: [LTspice] Re: noiseless resistor ? ? Hello Helmut, |
|
Re: noiseless resistor
Hello Helmut,
Thank you very much, that worked fine.? For some reason I was getting a runtime error when I'd done that before.? It must have been caused by some other error that is now fixed. It is interesting to see the noise density of a noise-free circuit? =? - 1E308 dB [!] Brett |
|
Re: Benchmarking the new LTspice
Hello
With Helmut's options Normal solver,? .options gmin=1e-10 abstol=1e-10 reltol=0.003No sdd .options gmin=1e-10 abstol=1e-10 reltol=0.003 .end ? WARNING: Node U3::44 is floating. ? WARNING: Less than two connections to node U3::44.? This node is used by Q:U3:1. Per .tran options, skipping operating point for transient analysis. Ignoring empty pin current: Ix(u1:2) Ignoring empty pin current: Ix(u1:4) Ignoring empty pin current: Ix(u1:17) Ignoring empty pin current: Ix(u1:19) Ignoring empty pin current: Ix(u2:15) Ignoring empty pin current: Ix(u2:16) Ignoring empty pin current: Ix(u2:23) Ignoring empty pin current: Ix(u2:25) Ignoring empty pin current: Ix(u2:27) Ignoring empty pin current: Ix(u2:29) Ignoring empty pin current: Ix(u2:31) Ignoring empty pin current: Ix(u2:33) Ignoring empty pin current: Ix(u2:35) Changing Tseed to 9.76562e-010 Heightened Def Con from 0.00462726 to 0.00462727 Heightened Def Con from 0.00469882 to 0.00469883 Heightened Def Con from 0.00495643 to 0.00495644 Heightened Def Con from 0.00508523 to 0.00508524 Heightened Def Con from 0.00511385 to 0.00511386 Heightened Def Con from 0.00525696 to 0.00525698 Heightened Def Con from 0.00528365 to 0.00528365 Heightened Def Con from 0.00528559 to 0.0052856 Heightened Def Con from 0.00559184 to 0.00559184 Heightened Def Con from 0.00562307 to 0.00562308 Heightened Def Con from 0.00801846 to 0.00801847 Heightened Def Con from 0.00814833 to 0.00814835 Heightened Def Con from 0.00838113 to 0.00838114 Heightened Def Con from 0.00858828 to 0.00858829 Heightened Def Con from 0.00878674 to 0.00878676 Heightened Def Con from 0.00885565 to 0.00885566 ? Date: Tue Aug 16 15:33:43 2016 Total elapsed time: 185.049 seconds. ? tnom = 27 temp = 27 method = modified trap totiter = 2840059 traniter = 2840059 tranpoints = 670882 accept = 479421 rejected = 191476 matrix size = 338 fillins = 124 solver = Normal Thread vector: 57.1/25.9[4] 32.9/15.8[4] 5.2/4.3[4] 1.8/2.0[1]? 2592/500 Matrix Compiler1: 17.74 KB object code size? 7.2/4.1/[2.0] Matrix Compiler2: 22.25 KB object code size? 3.8/5.8/[2.0] Regards PhB |
|
Re: Calculation of the inductance of the amplitude of the first harmonic voltage and current amplitude
¿ªÔÆÌåÓýHi, I thought he Inductance of an inductor is independent of the amplitude of the driving signal assuming you don't saturate the core. ?Please correct me If I am wrong. Michael To: ltspice@... From: LTspice@... Date: Tue, 16 Aug 2016 10:54:33 +0300 Subject: Re: [LTspice] Calculation of the inductance of the amplitude of the first harmonic voltage and current amplitude ?
Vlad,
I added one more circuit. And I asked the question. See my other letter! And see the file four_ways_to_determine_the_inductance.zip. Bordodynov. 16.08.2016, 10:35, "Vlad imbvlad@... [LTspice]" : >> I think not the average, and calculate the amplitude of the first harmonic. I made a parallel Fourier analysis. Look at the first harmonic and what I counted. The values are the same, with sufficient accuracy for practical purposes. > > Yes, the resultant is the amplitude, after multiplying the average -- the idt(...) term -- with 2. If, instead of idt(...)*2*f, you write idt(...)*f, you get the average, or half the amplitude, since you are calculating for the fundamental, only. At any rate, your calculations look correct, to me. > > Vlad > ______________________ > ltspicegoodies.ltwiki.org -- holding, among others: > a universal analog/digital filter, block-level models > for power electronics (and not only), math blocks > with a more stream-lined approach, some digital > ADC, DAC, (synchronous-)counter, JKflop, etc. > > |
|
Re: Benchmarking the new LTspice
Here is a summary of what performance has been reported so far with a guess at what a new system might do at the bottom.
Cores Clock Processor, Memory, Disk: Time (2/2) 3.00GHz Pentium D-930, 2GB, HD: 451s (my old home tower) (4/8) 1.73GHz Core i7-820QM, 4GB, HD: 246s (my old home laptop) (2/4) 2.70GHz Core i5-5200u, 4GB, SSD?: 145s (PhB's newish laptop)(4/8) 3.40GHz Xeon E3-1240 v3, 8GB, SSD: 126s (my office computer) (4/8) 3.50GHz Xeon E5-1620 v3, ?GB, SSD?: 91s (Helmut's office computer) (8/16) 4.20GHz Core i7-Extreme, 32GB, SSD: ~40s? (a new high performance tower) The last guess assumes that LTspice would use all of the cores. Most of the run times vary a little bit each run (10-20% or so), so it is fair to report the fastest, but my i7 laptop was acting very sluggish (>600s run times) until I killed and restarted Firefox (maybe that was it).? Whatever the problem was, it did not seem to show up in Task Manager. |
|
Re: noiseless resistor
Hello Brett,
The correct syntax is shown below.? With noise R1 N001 N002 {res} R2 7 3 1k Withot noise R1 N001 N002 {res} noiseless R2 73 1k noiseless Noise is only used in the .NOISE analysis. You can add "noiseless" in any line after the attribute-line. Ctrl right mouse click on the resistor. Enter noiseless in the line behind its value or in the following lines. Best regards, Helmut |
|
Re: noiseless resistor
Hi Brett.
toggle quoted message
Show quoted text
R1 N001 N002 {res} noiseless Or R1 N001 N002 R={res} Bordodynov. 16.08.2016, 15:37, "brett3nt@... [LTspice]" <ltspice@...>: When modeling circuit noise, I am using a resistor whose value is computed in a .param statement. |
|
Re: Benchmarking the new LTspice
Hello analogspiceman.
I tested on a E5 1620 v3 @3.5GHz. .Alternate solver ?options reltol=0.01 trtol=7? ?-> 96s I have seen sometimes wrong results in other circuits when using reltol=0.01 and trtol=7. Thus I never use or recommend this setting.?I recommend to only relax reltol to a maximum of 0.003 and let trtol as its default(1). Normal solver,? .options gmin=1e-10 abstol=1e-10 reltol=0.003? -> 150s It would be interesting if somebody with a i7 6700K could test the jigs circuit 3752-1 with the two settings above. By the way using Gear or cshunt=1e-16 significantly increased the simulation time. So it's not a good idea. Best regards, Helmut |
|
Re: Benchmarking the new LTspice
Gunoi Nare
We should use the Chinese supercomputer. I read is the fastest in the world and it is all made with Chinese hardware! No American intrusions :) G.
|
to navigate to use esc to dismiss