STEP BY STEP! SPEED UP Starnet & Other AI Tools in PIXINSIGHT, 2023 edition
Vložit
- čas přidán 15. 02. 2023
- A huge thanks to Riku Talvio for his AMAZING tutorial: rikutalvio.blogspot.com/2023/...
Support me on Patreon! / cuivlazygeek
Support me by buying ANYTHING from Amazon after following the below link!!
amzn.to/3hTB5Ne
OPT Affiliate Link: bit.ly/2OIw6jH
RENEWED! High Point Scientific affiliate link: bit.ly/3lReu8R
Merch: / cuivthelazygeek
External shop link: cuivthelazygeek.myspreadshop....
Follow me!
My Instagram: / cuivlazygeek
My Facebook page: / cuivlazygeek
Order of the Lazy Geeks slogan: Christopher T
Merch Design: Radu Chelariu
#astrophotography
#astronomy - Věda a technologie
Merci, Cuiv! Best thing ever!
I've been using this for a while but this was the best tutorial I have seen! You are going to get a lot of subs with this one, well done! CS!
Thanks Dave!
I just set this up and it works great. Thank you Cuiv.
Thank you so much for a fantastic guide for setting up NVIDIA/CUDA. Before the addition of the updates I ran SXT on a Heart & Soul Nebula that took 22 minutes. After the install it took 1 min 28 sec. I closed the project and shut PI down. Restarted PI and reloaded the project. Ran SXT again and it tool 58 sec. I'm running a NVIDIA Quadro RTX 4000. Also the load on the CPU went from nearly 100% to less than 5% and CUDA had a high % usage. FYI, the temperature of the card went from 35C to 45C. Thank you again!!
Woohoo!! I'm so glad this helped!
took starnet from 15minutes to 30 seconds XD, BlurX from 10 minutes to 26 seconds!!!! thank the lord and thank you Quiv haha
Awesome glad it helped!!
Excellent tutorial on setting up CUDA. I have seen CUDA in the the past, but thought, this seems complicated for cutting down 12 minutes of my workflow. Then along come C3/2022 E3 ZTF and having to run STX on 90 images = 18 hours. Now, 23 seconds per image after CUDA installation = 30 minutes! I appreciate your content. CS!
Hahaha the comet also made me change my mind in a hurry!
Thanks finally got it working! Great tutorial keep up the good work
Holy mother of god! What a difference! Many thanks Cuiv. Subscriber earned.
Welcome!
Cuiv, by the time I pressed the button to start the process to the time I opened up the Task Manager performance tab my star extermination was already done. You're a hero to this community -- thank you for the excellent and straightforward tutorial!
I know this is an older video to be replying to but I finally bought a decent computer and followed your instructions and.... WOW! All the AI programs only take a fraction of the time it normally would b4. Thanks for the video
Glad I could help!
Love it! Great walk-thru Cuiv, thanks!
My pleasure!
Thanks Cuiv! This was a very helpful companion to Riku's online instructions. My graphics card was also not on the list (GeForce GTX 1650) but I took the plunge and it works great! Significant reduction in processing time! 👍👍
Great to hear!
Ahhh! Whatever we imagers do without you! Thanks enormously.
My pleasure Melvyn!
Thank you so much .Finally I got it working
Excellent! Glad it was helpful!
I did see this walkthru guide when it was published by the author a couple of weeks ago, but didn't give it a go (too scared). But after seeing your video and how easy it was to install, I gave it a go. And by gum, it works! High speed XTerminators now!!!!
Great video! Easy to follow and the results on my laptop were unbelievable!
Great to hear!
Thanks ! This is the first set of instructions on how to do this that didn't seem like Hieroglyphics to me. It worked.
Legend! Thank you Cuiv. I have to batch process 50x Comet files to remove stars with StarXterminator before alignment and stacking. This will save me soooooo much time. Thank you so much😍
Excellent!! Also will save some electricity hehe
thank you for shring the info, very needed!
Glad it was helpful!
Wow! Thanks, Cuiv. Can't wait to try this! (You've probably cut my coffee making/consumption when these processes are running by at least half! 😁) Later...just loaded it all and I cannot believe how fast those processes run!! Amazing. Thanks again.
Thanks Cuiv and Riku Talvio. What an improvement!!! 10 minutes down to 45 seconds.
Woohoo!! Glad it helped!
Great Cuiv!
That was brilliant. Thanks so much Cuiv. My already fast machine has gone into overdrive. Amazing.
Great video!... My StarXterminator process for a given image has gone from almost 5 minutes to less than 10 seconds... that is remarkable!!!
It makes a huge difference, glad to see this little hack is being used more!
A very helpful tutorial. It works a treat! Thanks very much! 👍👍👍👍👍👍
My pleasure!
Excited about this upgrade on a new/old XPS 8940 desktop with 64GB of ram and 12Gb Nvidia RTX 3060 GPU with 4Tb SSD for my scratch files in PixInsight…excellent step-by-step instructions for Windows 11 and the recent GraXpert video was the cure for an ugly gradient that was plaguing my Elephant Trunk too…
Awesome tutorial. I had to reboot my Windows 10 system before the GPU (1080ti) utilization would kick in. StarXterminator task is down to about 35 seconds from 10 minutes. I wish I had that last week when I was processing some comet frames.
Great tutorial! easy update with massive improvement 😀
Glad it helped!
You're a legend...great job
Thanks Guiv! You got CUDA to work for me.
Awesome!! Glad I could help!!
Superb video!!!!!!!!!!!!!!!!!!
Thanks a lot!
Wow, that really makes a difference. Thanks!!
You're welcome!
I did this a few days ago WOW what a change it speeds up things so much makes the work flow so much easier not so painful
It really is a big improvement:-) 😃
Amazingly clear tutorial! I'm going to explore your CZcams channel for more ;)
Very cool. Just did some StarX batch processing on Comet C/2022 E3(ZTF) subframes which took over 18 hours to run. This route should knock it down to less than 2 hours -great!
It's a massive difference, and your electricity bill will thank you :)
very interesting. 😎 Thank you.
My pleasure!
Thanks!!!!
No problem!
Thank you so much mate. In StarXTerminator, the results are insane. Down from 12min to 1min17sec for a full resolution image in 4k quality. (RGB-Image / GTX1060)
That's excellent to hear!
When I first did this I failed to add the "BIN" at the end of my enviro variables. Working now. Excellent!
Excellent!
My GeForce GT1030 is not in the list of supported NVIDIA cards but this process still took StarXterminator from 8 minutes per image to 2 minutes, so worth a try even if your card isn’t in the list. You can always revert back if needed.
Just wanted to say thanks for another great tutorial. Appreciate you and rikutalvio making this easy to follow. Can't believe how much faster these processes are. Why did I not do this long ago??
Great to hear! It's really such a huge, huge difference!
Excellent, so wish I had that two weeks ago with the comet frames. Thank you so much!! is there a known work around for it doing the same in photoshop?
I unfortunately have no idea about Photoshop...
@@CuivTheLazyGeek well thanks again anyway. Loving the articles at the moment. You are a key contributor to helping so many of us in this wonderful and equally frustrating hobby. Off to set up for Simeis147 now😜🤞Clear skies
Thanks Cuiv - My version had stopped working back in version 10...now working again. Apparently, all the Russ Crowman tools use GPU on Mac without any of this fussing about - might be time to change for me!
Another great video! Plan on going through this today. Does it interfere with any other graphically intense applications? I run my indoor golf simulator off the same laptop.
It would only interfere if you run both at the same time :p
This is great, thank you! Any word if Pixinsight will enable this feature for WBPP?
I use the Quadro GV100/Threadripper workstation I put together for doing quantum chemistry calculations in my Windows PI processing - as it's there anyway. The GV100 chews through all the different AI tools in no time at all as it's one of the things it was built for!😄 It should work with the Linux version of Pixinsight as well and maybe even better as PI is developed in Linux. All my heavy duty GPU work (non-astro) is done under Linux I just haven't tested it with PI.
Agree well done Cuiv
Thank you!
Thanks for the shoutout and the walkthrough!
Woohoo, the one and only! Thanks so much for putting the tutorial together!
@@CuivTheLazyGeek It was too annoying to see all the people getting lost trying to follow other tutorials, so I had to. 😅
Fantastic! it is like I got a new computer. Thanks for sharing
Glad it's useful!!
Finally, working tutorial. Thanks, Cuiv! For some reason it didn't work for me before. Probably tensorflow library or zlib was corrupt or something.
Glad it worked! All credit to the writer of the tutorial!
merci, cela fonctionne
Did this for my computer last week. StarX has gone for six minutes to thirty seconds. Amazing improvement in processing speed.
It truly changes things!
Quel gain de temps avec cuda, merci pour le tuto toujours aussi TOP !
De rien Joël, enjoy!
Holy moly. I have a RTX3080 and it took 7 sec for StarX to give me a starless image. 7 SEC!!! My system was no slouch before, but it normally took several min. The floating point processing vs a traditional processor is just remarkable. Thanks for this tip!
Woohoo!! Now THAT is speed!
@@CuivTheLazyGeek no kidding. I just thought about trying this with my C2022 E3 data and do the batch processing. I have 61 frames. When I did this the "old" way it took about 2.5 hours. I ran that job just now and it took less than 15 min. :o
YAY you rock - 40 sec for me.
Awesome!! So glad to see this method being used more!
Thanks, Cuiv ... as always ... great instructions. I do, however, have a question. Will all this work with the CUDA version 12.0.1?
I really don't know - trying to chase the latest version was really difficult last time I did it, and because 11.8 is working so well, I'm fine with it for now!
I do love talking about the potential excesses of AI tools in one video, and then about how to make them faster in the next :D
The AMAZING tutorial by Riku Talvio:
rikutalvio.blogspot.com/2023/02/pixinsight-cuda.html
Support me on Patreon! www.patreon.com/cuivlazygeek
Support me by buying ANYTHING from Amazon after following the below link!!
amzn.to/3hTB5Ne
OPT Affiliate Link: bit.ly/2OIw6jH
I did this last week and I couldn't believe the difference it made!! Definitely worth the effort if your GPU qualifies to run CUDA.
Great to hear!
Thank you for doing this! I tried to follow another tutorial that didn't work for Windows 11, so hopefully this does the trick.
Did it work?
@CuivTheLazyGeek I didn't get around to it yet. I'm confident it wil workl since I used your video before I upgraded windows. Can't wait to run star Xterminator batch on my comet subs
I configured this not too long ago on my workstation. Even with my lowly gtx1660 the difference was night and day. :)
Exactly! I always love seeing the impact even with relatively low cost GPU!
Wooow, thanks for this. What can I do now with all the spare time :D ? I can't even make coffee while waiting :D
Hahaha you now have to make coffee *before* processing!
Thank you for the great explanation and I am happy for you that you are successful with this change - as I am sure many others are. I myself have an NVIDIA GeForce GT 630, which also appears in the list. It's just that the changes don't seem to work for me. With the system variable with the value "true", I wonder whether I have to enter "wahr", since my system is installed in "German". Perhaps you have a tip on this. Many thank you and greetings
I think it should stay to "true" in all cases... The GT 630 is quite old, so I think a modern processor might actually not lose against it...
@@CuivTheLazyGeek Hi Cuiv, thank you for your quick reply. The good news first. With the setting "true", despite a German system environment, my GPU is now addressed and used by PixInsight. However, not to the extent expected. For example, with starX rather 0% with noiseX around 11%. I think I should invest in a newer graphics card next. It would be worth it. Despite the 20 hours I needed for starX of my comet images, I still got a very nice image of C/2022E3. Not least thanks to your brilliant video on this. Once again, many thanks.
This is an amazing improvement. What laptop are you rocking?
This is an older gaming laptop from a Japanese company called Thirdwave/Galleria
This was an amazing video and so clear, for some reason my GPU still isn't taking over and it still took a few mins to run :(
Thanks for bringing my attention to this, huge improvement with a 2070 Super vs i7 4790k.
Glad to help!
How would you compare that way to the Repository Link that is offered by Russel Croman himself?
It did makes a difference for me. Running BlurXterminator on the same file now takes 2mins 19secs instead of 2mins! I think it was working as "cuda" in task manager>performance>gpu was running at 99%. I suppose my GeForce GT 1030 is not as fast as my Ryzen 7 2700.
Ha! Inverse effect... Sorry about that...
Time for me to get a new graphics card. 🙂@@CuivTheLazyGeek
I now have a RTX 3050. BlurXterminator now takes 38s instead of 120s :-)@@CuivTheLazyGeek
At first I was disappointed because I'm on a Mac and this procedure doesn't apply, but then I ran StarXTerminator anyway and it took only 28 seconds on a 10 hour integration with thousands of stars. In addition the process didn't use CPU or GPU. I assume it used the neural engine of this M2 Pro. The entire operation was silent.
Great to know!
A++
Does this all need to be redone whenever PixInsight updates?
Hi Cuiv, what is your Gpu ? is it GTX 1660 Ti .....
Good evening Cuiv, Great tutorial!!!
Could we hope that all the code of pixinsight uses cuda GPU? Can you? you who know before the others, explain why it works so fast.
Unfortunately no, this is specific for AI, which benefits from GPU...
Hi Cuiv. I had this installed in the past. Version 11.5, but it did not really do much. I followed this tutorial, and I was pretty shocked! I have a fairly high end system, though 2 years old. 3950x AMD 16 core CPU and RTX 2070 Super. BlurXterminator would take about 30s to 1 minute, NoiseXterminatot about 30s and StatXterminator almost always took at least 1.5-2 minutes. I’m shocked at how much improvement this version does!! Literally 2-5 seconds and it’s done!! Nice one Cuiv! Many thanks.
That's awesome!! Very glad to see it helped!
Do you know if this works with the latest update 1.8.9-3? I don't want to update if it doesn't work
Quadro RTX 5000 - orders of magnitude faster for the rc-astro plugins! Well worth the effort.
Awesome!
Indeed it is magic - even I just have a puny Nvidia RTX 1660 super 6G card, with a XEON E3-1225V2..... the performance gain is just amazing.
Excellent !!
Very good vídeo.
Now, can you know if it possible in linux?
Thanks
It should in theory be possible in Linux, but the Nvidia situation there isn't great. Most people I know running Linux will choose an AMD GPU
Oh gods, now I have to buy a new computer with an NVIDIA GPU for processing images. It'll be worth it, though. (Currently my laptop takes an hour + to run StarXTerminator.)
Oh wow, an hour is too much, a new computer with an NVIDIA GPU (even the xx60 series like the 3060) would make SXT fly!
Now if only WBPP speeds up a bit I might just have time for a new hobby. Maybe I can start knitting 🙂
Thanks for yet another great video! I did this process (or something v similar) about six months ago and it massively speeded up starXT using my NVIDIA card. But can I just check what you said at the end there? Do I have to do all this again for each new version of CUDA?
You don't have to do it for each version of CUDA! But apparently this particular version is faster compared to previous versions, so it makes sense to upgrade (although not mandatory)
Bonjour Cuiv.
Impossible pour moi de télécharger le zlib DLL. Quand je clique droit sur le lien, Windows 11 ne me propose que de copier le lien ou enregistrer le lien sous…
Et si j’enregistre le lien sous…, il m’enregistre un fichier qui se nomme : Non confirmé 725008.crdownload mais pas le dossier zip.
Je n’ai pas l’option enregistrer sous avec le clic droit sur le lien. Uniquement l’option enregistrer le lien sous…
Merci si tu peux m’aider à récupérer ce dossier zip et merci pour tes super vidéo.
Amitiés de Nice.
JM
On dirait un anti-virus qui bloque le téléchargement. Essaye avec Firefox peut-être ?
@@CuivTheLazyGeek same problem for me too..i tried disabling antivirus and also unflagged unsafe downloads. still no luck. any fix?
Ok. J'ai utilisé le navigateur "Tor". L'option "save as link" pour télécharger le dossier. Et cette fois, ca marche! J'espère vous trouvez le succès avec cette solution.
will this work with my ADM Radeon GPU??
Unfortunately no :-( Nvidia only...
Damn bro☹️
what about amd?
Yeah I wish :-(
Interestingly in my case the speed improved but GPU utilisation shows 0% utilised. Why is that?
Some cards will show the GPU usage in a different category (I think CUDA is one?)
@@CuivTheLazyGeek what does that mean? sorry didn't get it. When I go to task manager-performance tab, I see GPU statistics. But GPU is not showing any dynamic changes when being used. That is my query.
anybody try it with Cuda 12.0?
Not yet!
@@CuivTheLazyGeek I did and it didn't work. I followed Guiv's workflow and it worked using Cuda 11.8 with all the other downloads for Cuda 11.8.
When can we use similar Tweaks with AMD - don't wanna buy NVIDIA
The devs of the plugins have to port the code to OpenCL. CUDA is Nvidia only.
Unfortunately not that I know of at this stage
Et maintenant plus dur, sur Linux 😁
Oui, euh, non lol
Or just buy an M2 MacBook 😂
Ha! Unfortunately I'm forced to use a MBP at work, and I hate the OS with a passion! The window management on multiple monitors makes me want to smash the damn thing
Great hack. This method works fine. Thanks Cuiv, now the Xterminator flying un my laptop with RTX3060. BR,
Woohoo! Really happy to see this!
Thanks!
After 4 tries with other guides that didn't work, this guide worked. I just timed my starXterminator and it ran in 20 seconds.
Excellent to hear, and thank you for the support!!