What is PC performance effective for improving the processing speed of LNC?

@mo-mo Before looking into PC specs, perhaps you can first try to investigate if you actually need LNC. Do you really get better results when enabling LNC? It only needs to be used if the individual subs vary a lot in illumination. You wrote that "it is not a wrong setting because 2nd or 4th was appropriate by Mabula's past comments" but were those comments made to you as a general statement or for a particular issue that you encountered? Disabling LNC, of course, will speed up the integration process a LOT. Can you please try and make sure that LNC is absolutely necessary?

Having said that, hyperthreading is not the solution. It is a virtualization technique that in my experience does not speed up resource intensive computations like the one required by LNC.

@wvreeven

The information I got about LNC is the following URL:

https://www.astropixelprocessor.com/community/faq/what-exactly-does-the-lnc-and-when-do-i-change-the-default-values-to-something-else/#post-908

It is the only information I could find in the forum.
If there is a manual for LNC, I want very much.
Perhaps I think that I can understand correctly about LNC by manual.

My wish is to do not make an incorrect choice to speed up APP processing.

Information about Hyperthreading will be useful.
thank you.

@mo-mo

In page of the URL that you posted, you can read "If you are combining data shot at 1 single night then LNC will still help, but the improvement is very little and visually hard to see, because locally the frames already match pretty well.".

Further down the same page you can read "For regular integrations (I mean not a mosaic), start with 1st degree and 1 iteration. Then improve from there if needed.". Note, IF NEEDED. So your statement "I think that it is not a wrong setting because 2nd or 4th was appropriate by Mabula's past comments." is invalid. @Mabula has not written that "2nd or 4th is appropriate". He did write something about 2nd or 4th degree LNC but clearly states that it depends on the data and that the user should try this out.

So my question remains: do you NEED LNC? Did you test with and without LNC and did you see that there is a clear difference? Because if not, then you are set.

Posted by: @mo-mo

Hello everyone.

I have integrated 150 Light, 20 Dark, 20 Flat and 20 Bias.
At this time, I understood that the most time is spent in the processing of LNC. The LNC setting was 4th Degree, and Interation was 3.

It was not a mosaic but a general process.
I think that it is not a wrong setting because 2nd or 4th was appropriate by Mabula's past comments.

What I want advice is an effective PC's grade up to increase the processing speed of the LNC.
Is it a CPU core number, Hyperthreading number, memory, HDD, or other hardware?

I have disabled HyperThreading.
Can I get high speed if I enable HyperThreading?

My PC spec

CPU Xeon E5 2650 8core
memory 32GB
graphics NVIDIA Quadro K2000 2GB

Hi @mo-mo,

Like @wvreeven says, for most datasets, LNC is not really required, so in fact you can always first make an integration without LNC enabled. I would recommend to turn on first Multi-Band Blending if you see stacking artefacts. In most datasets, just MBB 5-15 % will clean things up nicely already 😉 MBB takes up 1.5x the amount of memory and HDD space compared to integrating without it and is not cpu intensive.

If MBB does not clean up all the stacking artefacts, my recommendation is to always start with LNC degree 1, 3 iterations. If you then still see stacking artefacts, yes, you can then try a higher degree. And keep MBB enabled of course.

Now the LNC algorithm is what they call in math a linear regression. At the moment it can only be calculated with 1 cpu thread. It is not yet multi-threaded. So the only thing right now that would speed that up is a cpu with the best single core performance that you can find.

But... at some point I will make a new and better implementation of LNC and it will probably be a non-linear regression then and I will make it multi-threaded at that moment as well. We need to do big matrix operations in such a calculation and I have already made that multi-threaded.

Now, LNC with the current implementation will slow down fast with an increasing amount of frames because it needs a lot of memory. So high memory bandwidth and a lot of memory will help. But this will also work better I think in the new LNC algortihm that I am thinking about...

Hyperthreading enabled or not will not do anything I think because of the argument that Wouter posed, the extra threads that are there with hyperthreading are not real cpu threads but virtual threads borrowing resources of the real threads.

Mabula

@wvreeven

Thank you for clarifying what I was misunderstanding about LNC.

Let's see if I really need LNC. I understand that comparison is possible by setting no LNC.
If LNC is not needed, PC augmentation should be done from a different perspective.

As Mabula answered, it's a good idea to try MBB first and then use LNC.
I am delighted to have a better understanding of APP.

@mabula-admin

I'm glad I got rid of my misunderstanding about LNC.

Also, I was advised to try MBB first, so I will try it.
Regarding LNC, it was found that increasing the memory is more effective

than increasing the number of physical cores of the CPU.

I also found that enabling Hyper Threading, as I had half expected, was less effective.
thank you.

You are most welcome @mo-mo 😉

Mabula