I am a bit confused with all of the high performance simulation options in ADE-XL/GXL.
I am using IC 22.214.171.1246 with MMSIM 10.11.200
On ADE I have:
1. "Options->Job Setup->Max Jobs"
2. "Options->Run Options-> Run In"
3. "High-Performance Simulation->Mode, Multithreading options, Affinity"
4. "Options->Analog->Multithreading Options"
What is the up to date way to run my simulations as fast as possible?
The main reason I'm asking is that I've had some problems when setting things such as APS + Max Jobs=8. For example, I've had these settings and sent a small simulation with a few corners and after waiting 10 minutes it still hadn't started. In general, it seems to take minutes until simulations start running sometimes (and then they run very fast).
Can anyone help me on this?
In reply to tkhan:
I can run in either local or distributed, and I have plenty of licenses, as I am a student in an academic institute and we get an enlarged package, so I don't mind using a lot of licenses.
In reply to aditeman:
This controls the number of parallel simulations that may be run at once in ADE XL. If you are using multiple tests in ADE XL, or corners, or sweeping variables or parameters, each one of these counts as a "job", and they can be run in parallel (on the local machine, on a remote machine, or distributed using a queueing system). Also if using Monte Carlo, it can break up the number of monte carlo points into chunks, based on the number of available parallel runs.
2. "Options->Run Options-> Run In"
This is not that useful - it's more to do if you hit the "Run" button again before existing jobs have finished - maybe having changed some setup.
This allows an individual simulation job to take advantage of multi-core technology and do things like the matrix solve in parallel - so it can speed up an individual simulation. Processor Affinity should be left alone (it allows you to get absolutely the best performance by telling the simulator which CPUs in a multi-core machine are to be used, but this is only ever useful if you have complete control over the machine you're running on, and has to be set carefully and is not normally worth the bother of setting)
This is pretty much obsolete nowadays - it was an earlier form of (limited) multithreadhing. It controls the same thing that's on the High Performance Options setting now - but that's the preferred location (the high performance options) to set multi-threadhing.
In reply to Andrew Beckett:
Do you have any idea what could be causing a long delay to initiate simulations? In other words, I press the "run" button, and once in a while I can wait a minute or longer (sometimes I stop it out of frusteration) before the simulations actually start. I have, for example, waited over 10 minutes for a simple simulation to start, then changed servers, and had the whole simulation from start to finish take 20 seconds. Our system administrator has tried to look into it, but as it happens quite sporadically, we have no clue what could be causing it. I do have a feeling, though, that it mainly happens when I try to run on multiple CPUs...
Hi Andrew and others,
I am running a long sim for a delta-sigma converter. Is there an optimum number of threads to use? I have six processors available, but I seem to remember that someone saying that at some point the speed increase is minimal if you add more processors -- I just don't remember what that point of diminishing returns was.
In reply to Rob Gregoire:
It's quite hard to be precise about this - APS supports up to 32 cores these days - and I have customers using that number (for very large circuits - e.g. several million transistors and even larger numbers of parasitics), so your mileage may vary. Generally speaking it's bigger circuits that will benefit from more threads. Doing some quick experiments with different number of threads would allow you to assess this for your size of design. You can check the utilization figures too that are reported in the log file (bearing in mind that you really wouldn't expect it to be able to fully utiliize all threads anyway).
Andrew, thanks for the answer. As always, I'm am very appreciative of the time you spend here helping.