UPDATE FROM MABULA
I have had a very rough 2 months unfortunately health wise. I was struck 3x in a row with bacterial infections. The second infection occurred  after a routine hospital checkup and I became very sick. I had to rest and take a lot of antibiotics. Once I was recovering and restarting work 1 month ago, I again became sick. The infection was not yet gone, so even more antibiotics and rest was needed. Needless to say, it took a lot of my energy and I needed a lot of rest.
Finally, the infection is really gone and my energy is coming back step-by-step now and I have started work again. I am terribly sorry to have kept you waiting for support. I will address all outstanding questions and e-mails step-by step and will be on the forum daily from today.
APP 2.0.0-beta47 will come soon as well, a lot of work was already completed before I became very ill, so the release is also nearly ready for you. Beta47 will be much faster actually. Many workflows will be more than 2x faster, mosaics can even be 10x faster than before because registration really received a major boost... all compared to beta46. Â Before I release it, I will make sure that everything is working properly and then I will release it.
Good evening, I would like to ask a question about the integration process.
I must say that compared to before with the two processors APP is much faster, but when it comes to the performing LNC calculation... I think it only uses 1 core since only 3% of the processors is shown by the Windows task manager.
Could it be made faster?
Thank you.
Â
Paolo.
Â
Â
🙂
Does anyone have an answer? 🙂Â
😑
I know that APP will only use one processor in a dual processor setup according to Mabula when I asked about it with my dual Xeon E5-2696v4 system. As far as the LNC calculation I am not sure if that will use the processor or if it uses anything from the GPU for calculations. I do know if you are running a multi session set do not have it do LNC again because that will take forever.
It doesn't appear to use just one processor....
Â
Looks like that must have been fixed in the past year then: https://www.astropixelprocessor.com/community/main-forum/dual-cpu-setup-app-is-only-using-one-of-them/#post-25198
I realised....
Hi @imnewhere and @paolo27,
Thank you Paolo for your question. The current LNC algorithm is a calculation (linear regression) that we can only perform at the moment using 1 cpu thread unfortunately. This calculation is not really subject to make it parallel. It is simply 1 big calculation with huge matrices... I am looking though to improve our algebra operations to see if we can make it run on multiple threads or perhaps rewrite the calculation in such a way that it can be made parallel. It is definitely a concern for us, we realize that the LNC is a big candidate for huge speed increase if we can make it work on multiple threads.
The issue of runnning with 2 CPUs and APP being able to use them, must be controlled by our development platform. Perhaps the latest APP versions with newer development platform can support it now. Paolo, you are running with the latest 2.0.0-beta28 right?
@imnewhere, can you still test on you 2 CPU system with beta28? It would be great if it confirms that it works on your system as well now.
Mabula
Hi @imnewhere and @paolo27,
Thank you Paolo for your question. The current LNC algorithm is a calculation (linear regression) that we can only perform at the moment using 1 cpu thread unfortunately. This calculation is not really subject to make it parallel. It is simply 1 big calculation with huge matrices... I am looking though to improve our algebra operations to see if we can make it run on multiple threads or perhaps rewrite the calculation in such a way that it can be made parallel. It is definitely a concern for us, we realize that the LNC is a big candidate for huge speed increase if we can make it work on multiple threads.
The issue of runnning with 2 CPUs and APP being able to use them, must be controlled by our development platform. Perhaps the latest APP versions with newer development platform can support it now. Paolo, you are running with the latest 2.0.0-beta28 right?
@imnewhere, can you still test on you 2 CPU system with beta28? It would be great if it confirms that it works on your system as well now.
Mabula
Â
The problem arose from the fact that by changing the two values with the arrows at 2 degrees and 4 iterations, the time had lengthened a lot compared to the previous system I had before. A Ryzen 3 4300 GE 4C/8T.
With the same single core, the Ryzen has a higher frequency, but the calculation time using a Xeon in that case had become very large.
I couldn't understand how such a difference was possible.
Â
Â
I must say that with the default values using Xeons it is a real monster APP! When the 32C/64T work together. @mabula-admin
Hi Paolo,
Yes, I can explain this. It has to do with single core/thread performance I think.
The LNC process is currently only using 1 core/thread. So the cpu that has better single thread performance will do LNC faster. In your case, the Ryzen is better in single core then. But the Xeon will be much faster ! for everything else in APP that can use multiple threads/cores 😉
We will try to have LNC run on multiple cores/threads for sure going forward !
Mabula
Â
A video
Now is better. Videos quickly become too large.
Â
Â
