Design Space Explorer Parallel Compilation

[TimBot]

The design space explorer has options for the number of maximum parallel compilations and also the maximum number of CPUs.

Is the maximum number of CPUs the number of CPUs for ALL parallel compilations combined or is it the number of CPUs for each parallel compilation?

[RichardTanSY_Intel]

The number of CPUs is the number of core processor detected on your local machine, that you can use, to perform the number of parallel compilation that you set.

Example. no. of CPU - 4; no. of parallel compilation - 2 : The DSE tool will use 2 core processor to run 4 compilations parallel.

And if you set no. of seed to 10, it will run 4 compilation first, then another 4, and lastly 2.

Take note that* Performing more than one compilation requires significant computing power, memory resources, and additional Quartus Prime licenses.

You may checkout this youtube video for further information.

https://www.youtube.com/watch?v=1cc74E3zaeI

Best Regards,

Richard Tan

p/s: If any answer from the community or Intel Support are helpful, please feel free to give best answer or rate 4/5 survey.

[TomCarpenter]

Hi Richard,

I'm using Quartus 22.1, and trying to get DSE II and Quartus to match up (i.e. DSE gives me a fitter seed that works, after searching through several), but if I then apply that seed to the source project in Quartus and run the fitter, I get different results.

Having read another thread on this forum, they said that you have to make sure that the maximum number of processors used in DSE matches the maximum number set in Quartus as that has an impact on the fitting process.

In DSE, I'm using no of CPU = 24  and no parallel = 6. From your description, that should mean that each compile uses a maximum of 4 processors per seed. I had set the max of 4 in Quartus project, but compiling that, I still get different results.

I've opened one of the DSE generated project copies, and notice that the Fitter settings that DSE applied to the project are shown as:

So that seems to suggest that the no of CPU setting is per seed run and not total number.

Would you be able to double check whether this is correct?

I'll try running again with DSE tomorrow with it set to no CPU 4 and see if they start to line up.

[TomCarpenter]

Ok, found the problem. The `set_global_assignment -name NUM_PARALLEL_PROCESSORS 4` was missing from the QSF files that DSE was testing on. Because I added it after generating the QAR that DSE was using, the two were producing different results - I'd assumed it would be added but obviously that was a bad assumption.

Running DSE again with the assignment produces matching results between Quartus and DSE which is what I need.

[RichardTanSY_Intel]

Thank you for acknowledge the solution provided. 

 I’m glad that your question has been addressed, I now transition this thread to community support. 

Best Regards,

Richard Tan

p/s: If any answer from the community or Intel Support are helpful, please feel free to give best answer or rate 4/5 survey.

[sstrell]

Why do you need to match in Quartus?  If DSE is producing a result that works for you, can't you just use that generated project?

[TomCarpenter]

We version control our project, but only at the source file level because the netlists are simply too huge (several GB) to be in a git repo. Couple that with an internal need for reproducibility (assuming same version of same Quartus toolset), we need to have a single project that can be stored rather than having multiple DSE generated projects.

While yes I could set DSE running again to get the required result, that takes a day to run and find the working seed (densely packed Arria 10 SX660). I'd rather then just be able to store and compile that one seed in Quartus the future (~4hrs) if I need to reproduce the netlist.

But anyway, it's moot now, I found the issue to be that I'd added the processor limit to the Quartus file after DSE generated its archive, so the two weren't running on the same `NUM_PARALLEL_PROCESSORS` setting.