Beginner speed woes
 
Share:

10th Feb 2020 - APP 1.076 beta 3 is available... check the release notes !

APP 1.076 beta 3 download links (Windows, MacOS, Linux RPM&DEB) are here (must be logged in !)

2019 November: Complete LRGB Tutorial of NGC292, The Small Magellanic Cloud by Christian Sasse, Astronomer in Charge of iTelescope.net

2019 September: Astro Pixel Processor and iTelescope.net celebrate a new Partnership!

Beginner speed woes  

  RSS

(@milesy303)
White Dwarf Customer
Joined: 3 months ago
Posts: 10
December 11, 2019 02:42  

It’s been a long time since I used a message board, all the ones I used were taken over by Facebook haha. 

So I’ve installed fine trial into my laptop which is about 10 years old, 1Gb ram and maybe 1. Something amd processor. It was a windows 7 laptop so you can probably figure out the sort of spec it has.

are things supposed to take this long? I managed to stack my latest images in deep sky stacker in about several hours. Approx 197 lights totalling about 5 hours of exposure.

so far every single step I do in APP is taking that amount of time.

Today I started the saving registered frames and after about 6 hours I was playing in a terminal (I’m using Ubuntu on the machine now) and managed to accidentally cancel the save.

I was gutted at this so I had to start it again as there was no obvious way to restart it again and I left it and went out the house. I came back several hours later to find a message saying - do you want to overwrite - yes or no, and nothing had completed, and now I have no option to answer for all 197 lights - it would be handy to be able to say yes for all or no fall or all - in this particular instance - skip all - meaning I wouldn’t have to wait on every single light being processed again...

anyway I digress a little bit - the current saving process here is taking 3 minutes per image so I estimate that’ll take 9 hours to save the registered files, so I’m off to my bed to let it run again.

im scared how long the integration step will take - if I ever manage to get that far before my trial runs out.

so far excluding my accidental cancel - there’s been about 48 hours of processing time here - on 5 hours of andromeda - not 20 hours - 5 measly hours. 

Its not a 486 I’m running surely I can get better performance than this? Haha 

chris 

 

EE42357A F844 4AD1 A886 5467BA27673E

This topic was modified 2 months ago by Vincent Groenewold - Moderator

ReplyQuote
Topic Tags
(@vincent-mod)
Quasar Admin
Joined: 3 years ago
Posts: 1261
December 11, 2019 15:57  

So APP is using a very different and way more advanced calibration engine compared to DSS. But it requires quite a lot of calculating. Mabula is working on getting this faster by supporting GPU's in the near future, however it's not recommended to use very old computers. 1 GB of ram is very limited and many use at least 8, 16 being even better. A SSD instead of a regular harddrive will speed things up quite a bit as well. So unfortunately, the real issue in this case is the age of the computer I'm afraid. DSS might be able to still process it, but there's also a reason why results in APP are generally much better.

This post was modified 2 months ago 2 times by Vincent Groenewold - Moderator

ReplyQuote
(@milesy303)
White Dwarf Customer
Joined: 3 months ago
Posts: 10
December 11, 2019 16:10  

Thanks. I’ll see what the results look like. I’ll be getting a new computer next year but hopefully once my guiding stuff arrives I’ll be able to get longer subs and reduce this processing time. 

Looking at the way the tasks are performed this looks like something that could perhaps be done using a clustered system?

ive got a cluster of 6 raspberry pi’s sitting there idle. It would be fantastic if there was a way the software could be written to support executing commands in parallel where they could be farmed out. 

Chris 


ReplyQuote
(@vincent-mod)
Quasar Admin
Joined: 3 years ago
Posts: 1261
December 11, 2019 16:24  

Well, now there's an interesting idea. I don't think Mabula is targeting for running it on pi's (I love them, but they're not processing champs and APP does benefit from multicore CPU's, the new Threadripper cpu's are great for it), but I like the clustering idea, I'll ask him if that would be possible, sharing loads between fast computers.

This post was modified 2 months ago by Vincent Groenewold - Moderator

ReplyQuote
(@milesy303)
White Dwarf Customer
Joined: 3 months ago
Posts: 10
December 11, 2019 16:29  

Each of mine are probably as fast as this laptop it seems haha. 

I have some background in neural networks and my post grad was basically about sharing computation across a cluster where processing was something that could be done as discrete jobs on their own. So say normalise one image with the inputs it requires. 

Assuming the time to transfer the data is quicker than the time it takes to process it then there would be a net gain. 

Chris 


Share: