¿ªÔÆÌåÓý

Re: Benchmarking the new LTspice


 

Allan,

My home laptop is an older Lenovo quad core i7 and I also have an ancient dual core Pentium desktop machine that can't support Windows 7.? When I get home (after work and after my tennis match), I will run the benchmark on both and report back.

I am looking to buy a Core i7 Extreme 6/12 or 8/16 that runs at 4.2 GHz or so with lots of RAM, a solid-state drive and super quiet cooling.? I am hoping it will be more than twice as fast as my office computer.


---In LTspice@..., <allanvv@...> wrote :

I don't think you really need to go for a workstation PC. You wouldn't need an overpriced workstation GPU for basic CAD like for running SPICE or PCB design.

My recommendation is to buy a good desktop CPU with 6-8 cores, buy a gigantic $50 CPU fan (Noctua), and overclock your CPU. While some simulations can make good use of 4+ cores, single-threaded perf is still really important for a lot of applications and you'll get a lot more performance than any stock computer. With a good heatsink+fan, you can get great single-core GHz above the Turbo Boost frequency, and also good 8-core performance at the TB frequency without even coming close to thermal limits.

If I weren't traveling at the moment, I'd give you the performance of your benchmark at 6x4.4 GHz on a similar CPU generation.

Allan

On Mon, Aug 15, 2016 at 6:44 PM, analogspiceman@... [LTspice] <LTspice@...> wrote:


LTspice Yahoo Group readers,

I am shopping for a new home computer optimized to run LTspice and have been corresponding with Orbital Computers. ? Danny Payne, the owner there was willing to benchmark LTspice for me.? He downloaded the new version (LTspice XVII) and tried running some of the test jigs.? He chose the 3752-1 test jig, which was a very good choice for someone not familiar with LTspice because he chose a file that maxes out all eight of my work computer's threads.

However, it turns out that either the 3752-1 model or its test jig is not well written because it throws lots of Def Cons with the default simulation settings.? It runs much happier and faster with the Alternate Solver selected, Reltol relaxed to = 0.01 and Trtol set to 7 (click the Hammer icon to access the Control Panel and then click the Hammer tab to access the SPICE settings).

On my (4 core / 8 thread) Xeon E3-1240 v3 @ 3.40GHz with 8GB ram (NVIDIA Quadro K600) at work, the SPICE Error Log reports a run time of around 126 seconds (once the Alternate Solver has been selected and Reltol has been set to 0.01 and Trtol set to 7).? NOTE: Do not plot any traces when running this benchmark.? With the same simulation settings, the old LTspice IV runs this test jig in 137 seconds, which is about 8 percent slower.

I am going to send Danny Payne this message and have him run the benchmark with the same settings on an LTspice optimized computer platform and report his results (as well as the system configuration details).? It is interesting that he recommends Windows 7 for all CAD programs.

It would be fun to see what run times other group members report for this benchmark.





Join [email protected] to automatically receive all group messages.