Pixinsight & StarXTerminator with AI 11 issues - Cured (was my computer bios settings not PI or StarXTerminator)

[teoria_del_big_bang]

Anybody else have any issues with latest AI version11 in PI ?

I have up to date PI on two computers, admittedly not gaming machines but not archaic by any means, but on my laptop I cancelled the process after about an hour it had only completed 10% and on my desktop I have tried 5 times now on various size images and always crashes , well I say it crashes everything becomes unresponsive, just freezes with no cursor movement, unable to switch between running processes and ctrl alt del also dows nothing.

Desktop:
Processor    AMD Ryzen 5 2400G with Radeon Vega Graphics       3.90 GHz
Installed RAM    32.0 GB (29.9 GB usable)
System type    64-bit operating system, x64-based processor
Pen and touch    No pen or touch input is available for this display
 

Laptop:

Processor Intel(R) Core(TM) i5-3210M CPU @ 2.50GHz   2.50 GHz
Installed RAM 16.0 GB (15.7 GB usable)
System type 64-bit operating system, x64-based processor
Pen and touch No pen or touch input is available for this display

EDIT: Now cured, my Bios settings had been tweaked to speed up processor which overheated when process was run, put Bios to default values and now okay.

Steve

Edited September 14, 2022 by teoria_del_big_bang

[scotty38]

I have seen lots of chat on the PI forum about the speed of the latest version, so much so that he created a lite version to speed things up. Have you seen it?

[teoria_del_big_bang]

2 minutes ago, scotty38 said:

I have seen lots of chat on the PI forum about the speed of the latest version, so much so that he created a lite version to speed things up. Have you seen it?

Yes tried that too.

I think the CPU or memory load may be why it is so slow on the laptop but if I leave it it does look like it would work, eventually, but on desktop I think there is a bigger issue, probably with my computer. I tried the older AI's and it also freezes, also sometimes does this on Starnet and Starnet++ so maybe my computer.
Not what I wanted. really wasn't

Steve

[scotty38]

My laptop is about the same spec as yours and Starnet has always been a bit tardy but I don't have Starxterminator to try unfortunately

[teoria_del_big_bang]

How much is a quantum computer these days ???

[tomato]

Just installed AI 11, runs about 2.5 times slower than AI 10 (AMD Ryzen 7 3700X 8-Core Processor  3.59 GHz, 32GB RAM), but no freeze ups or crashes and produced a much better result than the previous version. Did a M31 tiff file (6351 x 4138) in around 10 minutes.

 

[billhinge]

I  hadn't heard of this before so thanks for the heads up, gave it ago, hardly breaks a sweat, am I missing something?

Here are the Lite version and the Full version

StarXTerminator: Processing view: NBRGBCombination_DBE

Writing swap files...

644.561 MiB/s

Lite Version

RC-Astro StarXTerminator version 2.0.1, AI version 11 lite

Tile overlap = 20%

Removing stars: done

Generating stars image

27.068 s

StarXTerminator: Processing view: NBRGBCombination_DBE1

Writing swap files...

655.160 MiB/s

Full Fat Version

RC-Astro StarXTerminator version 2.0.1, AI version 11

Tile overlap = 20%

Removing stars: done

Generating stars image

21.114 s

I used the data here. IKO - M81 & M82 LRGB+Ha - Processing Competition

I didn't put a lot of effort into the test as I just wanted to try it out speed and quality wise.  The Full fat version is faster and better

 

 

Out of the box settings - seems OK to me Full Fat on the RHS

 

 

 

Apple Macbook M1 Max 64GB Ram - it does have a built in neural engine - not sure if this uses that though?

 

 

[Budgie1]

So, I also installed the update this evening and tried it out on my desktop.

The computer is an older one which I built a few years ago but works well. The specs are: AMD FX 6300 Six-Core 3.5 Ghz, 32GB RAM, 4GB graphics card, Windows 10 Home, 1TB SSD

Using StarXTerminator in PI on a non-linear image, Bin x2 (which my normal process) I got the following times from the versions using default settings:

• v10 - 1m19s
• v11 - 10m21s
• v11 Lite - 5m12s

Of the three AI versions I tried, I would say v10 still gave the best results, followed by v11 Lite, then v11 which left artefacts from one star. The PC ran fine throughout, with no lockups, Not Responding messages or anything and I was able to use my browser while the process was running.

[teoria_del_big_bang]

15 minutes ago, Budgie1 said:

So, I also installed the update this evening and tried it out on my desktop.

The computer is an older one which I built a few years ago but works well. The specs are: AMD FX 6300 Six-Core 3.5 Ghz, 32GB RAM, 4GB graphics card, Windows 10 Home, 1TB SSD

Using StarXTerminator in PI on a non-linear image, Bin x2 (which my normal process) I got the following times from the versions using default settings:

• v10 - 1m19s
• v11 - 10m21s
• v11 Lite - 5m12s

Of the three AI versions I tried, I would say v10 still gave the best results, followed by v11 Lite, then v11 which left artefacts from one star. The PC ran fine throughout, with no lockups, Not Responding messages or anything and I was able to use my browser while the process was running.

Thanks for the info, I will try tomorrow on laptop as it looks like there is something amiss with the desktop as it freezes using starnet as well.

Steve

[gorann]

Has anyone tried v11 on PS (if it is available)? Can one revert back to v10 if v11 is too slow or gives a poorer result?

[scotty38]

1 hour ago, gorann said:

Has anyone tried v11 on PS (if it is available)? Can one revert back to v10 if v11 is too slow or gives a poorer result?

I know on PI it's just a case of changing the URL to the new version while it's in testing so if PS works the same I guess you can too?

[billhinge]

12 hours ago, billhinge said:

I  hadn't heard of this before so thanks for the heads up, gave it ago, hardly breaks a sweat, am I missing something?

Here are the Lite version and the Full version

StarXTerminator: Processing view: NBRGBCombination_DBE

Writing swap files...

644.561 MiB/s

Lite Version

RC-Astro StarXTerminator version 2.0.1, AI version 11 lite

Tile overlap = 20%

Removing stars: done

Generating stars image

27.068 s

StarXTerminator: Processing view: NBRGBCombination_DBE1

Writing swap files...

655.160 MiB/s

Full Fat Version

RC-Astro StarXTerminator version 2.0.1, AI version 11

Tile overlap = 20%

Removing stars: done

Generating stars image

21.114 s

I used the data here. IKO - M81 & M82 LRGB+Ha - Processing Competition

I didn't put a lot of effort into the test as I just wanted to try it out speed and quality wise.  The Full fat version is faster and better

 

 

Out of the box settings - seems OK to me Full Fat on the RHS

 

 

 

Apple Macbook M1 Max 64GB Ram - it does have a built in neural engine - not sure if this uses that though?

 

 

Just noticed the blurb, it does seem that it uses the M1 Max hardware neural engine which would explain the speed (21 seconds)

GPU/Neural Engine recommended for fast performance, but not required. Not all GPUs are supported. 

 

[gorann]

44 minutes ago, scotty38 said:

I know on PI it's just a case of changing the URL to the new version while it's in testing so if PS works the same I guess you can too?

In PS you update StarXTerminator by clicking on "Download AI" when the Star XTerminator dialog box comes up. I assume clicking that button would update to v11 but I have no idea how to reverse that back to an earlier version. Maybe some PS user here knows more about this.

Cheers, Göran

[teoria_del_big_bang]

Hi @gorann if nobody gets back to you you could ask the question in StarXTerminator support (link below).

StarXTerminator Support

Steve

[teoria_del_big_bang]

Hopefully had some success, after trying all sorts all day.
What I did notice is that when it started to run any of the star removal processed (Starnet, Starnet2 or StarXterminator the CPU emperature shot up very rapidly and it always to freeze when it hit about 73 C. 
Now although I have never touched the Bios I went into Bios setup and selected load default values, now when I run the temperature only seems to peak about 63 degrees and the processes work without any freezes (SO FAR).

So I suspect that the manufacturer of my tower has tweaked things to make it seem a better machine that it is meant to be.
As I say it seems okay now but for what I thought a couple of years ago was a pretty fast machine (not gaming standards though) takes a while to do these things, using StarXterminator AI10 takes 35 seconds and A11 Lite  takes 21 minutes, not yet reied AI11.

Steve 

[Budgie1]

I went with liquid cooling of the CPU a few years ago, Steve. My desktop case also has extra fans in the front and base of the case to create a through draft of cool air from floor level. I did build it when I did a bit of gaming and noticed the temps going up, so took precautions.

I've just run a quick test and using v11 Lite my CPU was running all six cores at between 96 & 100% load. The CPU temp went from 36°C before I started, increased to an average of 43°C during the process and never went over 46°C.

[tomato]

I can hear the cooling fans kicking in when I run AI 11, I ran it simultaneously with  StarTools processing last night (not a good idea), there was a Not Responding message coming up in PI every couple of seconds but it still completed the task in around 10 minutes. 

[billhinge]

I tried the PS version last night, only gives the one AI version option. Takes about the same time about 20s, so quite quick.

With PI I got two images - galaxies + stars but the PS only seems to give galaxies? Did a good job though!

I'm quite impressed so I may try the noise reduction app as well

Anyone done a comparison with the old starnet (sp?) app in PI?

[teoria_del_big_bang]

36 minutes ago, billhinge said:

I tried the PS version last night, only gives the one AI version option. Takes about the same time about 20s, so quite quick.

With PI I got two images - galaxies + stars but the PS only seems to give galaxies? Did a good job though!

I'm quite impressed so I may try the noise reduction app as well

Anyone done a comparison with the old starnet (sp?) app in PI?

I have not done any thorough comparisons but on the image I was using to do the tests whilst my comuter kept failing when I got it working StarXTerminator did seem to do better than Starnet2.
There were two big stars that left awful glows around them when I originally processed the image that were greatly improved with StarXTerminator using either AI 11 or the lite version. That was using the 20% overlap I did not try the 50% overlap option but I guess that may further improve things but at the suggested 3X longer to run on my computer I will have a fair wait.

Generally so far I have always used both processes and then compared the images to see which is the best and go with that one but knowing that one is always significantly better would be great and save time in the future.

Steve

[billhinge]

Guess I shall have to volunteer to do a side by side then 😉 I'll try one of those images from the IKO contest

May give it a go tonight

[teoria_del_big_bang]

I too may do the same, if not now I have a full week at Kelling next week so lots of time without all the jobs around the house to do so should have plenty of time.
I guess you really need to test a variety of images, some with lots of nebulosity, some with very bright stars etc. 

Steve

Edited September 15, 2022 by teoria_del_big_bang

[billhinge]

I had a go using the competition data for this

 

I used the full fat AI with maximum processing options but it was still quick

Here are my effort,  if I had entered the competition - you can play with the colours yourself if you don't like mine

I was going to use starnet++ for comparison but it isn't available on the M1 mac because someone decided I shouldn't have it, duh!, however if you can find it on the internet then you can struggle to install  it manually. Unfortunately it is nowhere to be found. I looked but no luck. If they are making it so difficult I can't be bothered

 

Edited September 15, 2022 by billhinge