Wow just downloaded it, pulled one sub out, ran gradient correction, bxt correct only, nxt then full bxt, Less than 90 seconds for all 4 process. Cheers for that, keep up the great work. Dave
Very well done video. Great, slow yet detailed explanation of the process. I wish I had this quick repository link back when I did the update manually.
Thank you! This is definitely much easier than going the old route. Plus this prevents the tensorflow file from being reverted during major PixInsight updates :)
@@davegage1949 hi I’m using a GeForce rtx 2060 on my laptop, and I was blown away, the post i sent was actually wrong, I did 5 processes, forgot to add star tx, so to get all 5 done in less than 90 seconds was great
@@Hidden.Light.Photography Not a exact comparison, but the other day I used 7 subs, put them in the image container, then used the process container to load up the processes, that took about 2 hrs to do all 7 subs, I’d be interested to see how it performs with deep snr, s that is very slow, nxt used to take about a minute old style but deep snr was about 5 minutes it works well but tended not to use it because of the time it took, so I will have a play with that next week, just upgraded my desktop a few months ago, and that’s got a really good graphics card in it so I’m looking forward to trying it out on there
Thank you! I apologize, trying to get to comments as fast as possible. Bouncing off of @thomasrider5852, the GPU must be on the list of CUDA enabled GPUs for this to work.
Thank you! You don’t necessarily need to uninstall it, but it is no longer used, so to save space on the disc, I would uninstall. Make sure to put the original PixInsight tensorflow file back in place and the repository will take care of the rest :)
@@Hidden.Light.Photography I'm still rocking an old 2080 Super and the acceleration is still super fast on that GPU. Not quite the same that you are getting but my times went from over 3 minutes to under 45sec.
I really wish I could, but unfortunately the GPU needs to be CUDA enabled for this to work :( I was using my imaging laptop to do my processing for a long time and worked multiple scenarios to try to figure out how to get that laptop to be able to run this. I ultimately broke down and bought my Alienware which had the proper hardware as this was the easiest way to do it. My best advice if you end up going this route is keep your eyes open and don’t jump on the first one. I ended up finally finding one on an incredible deal (previous model close out sale and paid a fraction of the normal price). It might be possible to convert your current computer to be compatible, but that is outside my realm of knowledge, so if that’s a route you want to go, I would find someone who builds computers as they would be the best resource for that. I hope this makes sense :)
I remember reading somewhere that Apple uses different algorithms and those processors have amazing performance when it comes to this. I just cannot confirm, unfortunately. Even though I am a huge Apple person, I don’t have a Mac with either of those processors :( If what I read in forums and reviews is correct, it runs incredibly :)
Great question! Yes and no. Yes as in there won’t be change in performance. No as in the change you would have is you won’t have to deal with the tensorflow file within PixInsight getting reverted during major PixInsight updates to where you need to go back in, rename the default tensorflow file and then move the CUDA tensorflow file back in. Also, as long of the repository is kept up, any updates that happen within CUDA should be handled through the repository with a simple update of the PixInsight repository list that we get vs doing the update through CUDA. If it were me and current system is working, I would leave it :)
I have an Intel Arc A770.. I also have a GTX1080... a few months ago the intel or Ngreedia drivers conflicted and BSOD'd my system.. :( I need to try GPU accelerated PI again..
I’m sorry to hear that that :( Most likely not the same thing, but I use an Alienware for my processing and back in June there was a major Windows update. Not everything was released to Alienware and I had some major issues to the point I had to redo my entire Alienware system. It might be possible that yours was on the intel side as the timeframe matches in a way. Another possibility is if it was Windows update related, it probably interfered with the background settings that are required for the old CUDA process. During my search to locate the instructions again to set up the CUDA end of things, I found this repository method which made this process very easy and accessible to anyone with compatible GPUs as everything in background with the old method is handled behind the scenes with no user intervention required. It might be worth a try with this method since you aren’t making any changes directly within Windows. I’m not exactly sure what happened in your case though so I can’t say for sure.
Wow just downloaded it, pulled one sub out, ran gradient correction, bxt correct only, nxt then full bxt,
Less than 90 seconds for all 4 process. Cheers for that, keep up the great work. Dave
It is absolutely incredible difference and thank you! Please let me know if there’s anything else you’d like to see :)
Nice one hope to do this next my pc when I get a minute
Thank you! Just make sure your GPU is on the list of CUDA enabled GPUs and you should be good to go :)
Very well done video. Great, slow yet detailed explanation of the process. I wish I had this quick repository link back when I did the update manually.
Thank you! I agree, the manual way was a bit tedious haha, but I’m very happy they came out with this repository :)
Wicked cool! Excited to try this.
It makes such a difference! Just make sure you have a compatible GPU :)
Great video for people who were unaware of this.
Thank you! This is definitely much easier than going the old route. Plus this prevents the tensorflow file from being reverted during major PixInsight updates :)
What Nvidia card are you using for this? I'm thinking about upgrading my graphics card.
@@davegage1949 hi I’m using a GeForce rtx 2060 on my laptop, and I was blown away, the post i sent was actually wrong, I did 5 processes, forgot to add star tx, so to get all 5 done in less than 90 seconds was great
This is an incredible method to boost the performance! I’m currently running a GeForce RTC 4070 :)
@Puffer001 that is amazing performance! Did you by chance calculate before time as a comparison?
@@Hidden.Light.Photography
Not a exact comparison, but the other day I used 7 subs, put them in the image container, then used the process container to load up the processes, that took about 2 hrs to do all 7 subs, I’d be interested to see how it performs with deep snr, s that is very slow, nxt used to take about a minute old style but deep snr was about 5 minutes it works well but tended not to use it because of the time it took, so I will have a play with that next week, just upgraded my desktop a few months ago, and that’s got a really good graphics card in it so I’m looking forward to trying it out on there
Ps just to let you know, you and seti Astro are putting some amazing videos out, they are helping me immensely. Cheers
This is a fantastic video. Clearly explained step by step. Are there any tutorials available for AMD GPUs?
AMD will not do this due to architecture differences between AMD and Nvidia
Thank you! I apologize, trying to get to comments as fast as possible. Bouncing off of @thomasrider5852, the GPU must be on the list of CUDA enabled GPUs for this to work.
Nice and informative video. I have the Cuda installed, do I have to uninstall it before updating with your solution?
Thank you! You don’t necessarily need to uninstall it, but it is no longer used, so to save space on the disc, I would uninstall. Make sure to put the original PixInsight tensorflow file back in place and the repository will take care of the rest :)
can’t wait to get a GPU and finally do this for myself! any recommendations?
It is incredible! I’m currently using the GeForce RTX 4070 :)
@@Hidden.Light.Photography I'm still rocking an old 2080 Super and the acceleration is still super fast on that GPU. Not quite the same that you are getting but my times went from over 3 minutes to under 45sec.
That is still very respectable improvement!
I have an issue that my Radeon 6800 isn't CUDA compatible. Can you fix this, Tony? Thanks.
I really wish I could, but unfortunately the GPU needs to be CUDA enabled for this to work :( I was using my imaging laptop to do my processing for a long time and worked multiple scenarios to try to figure out how to get that laptop to be able to run this. I ultimately broke down and bought my Alienware which had the proper hardware as this was the easiest way to do it. My best advice if you end up going this route is keep your eyes open and don’t jump on the first one. I ended up finally finding one on an incredible deal (previous model close out sale and paid a fraction of the normal price). It might be possible to convert your current computer to be compatible, but that is outside my realm of knowledge, so if that’s a route you want to go, I would find someone who builds computers as they would be the best resource for that. I hope this makes sense :)
I don't have Pixinsight but I wonder if Pixinsight has been modified for Apple Silicon, i.e, M1 through M4 processors.
I remember reading somewhere that Apple uses different algorithms and those processors have amazing performance when it comes to this. I just cannot confirm, unfortunately. Even though I am a huge Apple person, I don’t have a Mac with either of those processors :( If what I read in forums and reviews is correct, it runs incredibly :)
I imagine if we've already enabled GPU acceleration via the tensorflow file, this won't change anything...correct?
Great question! Yes and no. Yes as in there won’t be change in performance. No as in the change you would have is you won’t have to deal with the tensorflow file within PixInsight getting reverted during major PixInsight updates to where you need to go back in, rename the default tensorflow file and then move the CUDA tensorflow file back in. Also, as long of the repository is kept up, any updates that happen within CUDA should be handled through the repository with a simple update of the PixInsight repository list that we get vs doing the update through CUDA. If it were me and current system is working, I would leave it :)
I have an Intel Arc A770.. I also have a GTX1080... a few months ago the intel or Ngreedia drivers conflicted and BSOD'd my system.. :(
I need to try GPU accelerated PI again..
I’m sorry to hear that that :( Most likely not the same thing, but I use an Alienware for my processing and back in June there was a major Windows update. Not everything was released to Alienware and I had some major issues to the point I had to redo my entire Alienware system. It might be possible that yours was on the intel side as the timeframe matches in a way. Another possibility is if it was Windows update related, it probably interfered with the background settings that are required for the old CUDA process. During my search to locate the instructions again to set up the CUDA end of things, I found this repository method which made this process very easy and accessible to anyone with compatible GPUs as everything in background with the old method is handled behind the scenes with no user intervention required. It might be worth a try with this method since you aren’t making any changes directly within Windows. I’m not exactly sure what happened in your case though so I can’t say for sure.