AstroBin

AstroBin

AstroBin

Finally got starnet in Pixinsight to use my GPU card for acceleration -- 5X faster than CPU [Deep Sky] Processing techniques · Andy Wray · ... · 5 · 1118 · 0

andymw 11.96
Andy Wray
...
· 
·  5 likes
·  Share link
I've struggled to get starnet in Pixinsight to use my RTX 2060 to accelerate its performance.  I eventually stumbled across the following link which had versions of the various libraries that worked with the latest Nvidia driver for my card: https://towardsdatascience.com/how-to-finally-install-tensorflow-gpu-on-windows-10-63527910f255

Just thought I'd share as there are so many old links on this topic.
Like
×

Sorry :-(

Editing a post is only allowed within 24 hours after creating it.
×

Share this forum post

Copy the URL below to share a direct link to this post.

Close
mxpwr 7.29
D. Jung
...
· 
·  1 like
·  Share link
Does it work with StarXTerminator, too?
Like
×

Sorry :-(

Editing a post is only allowed within 24 hours after creating it.
×

Share this forum post

Copy the URL below to share a direct link to this post.

Close
Alan_Brunelle
Alan Brunelle
...
· 
·  Share link
It is really nice.  I got at least a 5X acceleration.  

Then I moved to StarXterminator.  Also worked very fast.  Not sure it was using graphics acceleration.  But something has changed and now it crashes the computer when working on larger files.  But with no images to process, it really doesn't matter for me right now.
Like
×

Sorry :-(

Editing a post is only allowed within 24 hours after creating it.
×

Share this forum post

Copy the URL below to share a direct link to this post.

Close
Alan_Brunelle
Alan Brunelle
...
· 
·  Share link
D. Jung:
Does it work with StarXTerminator, too?


I read on the StarXTerminator page that it had once used graphics acceleration, but he killed that process because it was buggy.  And maybe it will come back, once the bugs are fixed.  Being new to StarXTerminator, I am not sure what I experienced.  Was it actually using the graphics acceleration when I was getting great performance?  And then not currently, which crashes on larger files?  I am guessing that I never actually got the graphics accelerated version.  But it was and is still pretty fast, when it works.  So if he makes a faster version that would be great.  I will have to see if I can solve my issues by dropping my CPU overclock to see if it was just stressing the processor.  That should not be the issue if it was graphics accelerated.  Sorry for my lack of definitive answers, but I look forward to hearing someone give some definitive answers on this.
Like
×

Sorry :-(

Editing a post is only allowed within 24 hours after creating it.
×

Share this forum post

Copy the URL below to share a direct link to this post.

Close
Djibi 2.41
Jean-Baptiste
...
· 
·  Share link
Thanks @Andy Wray
On my side I tried but I could not get the GPU acceleration up to now.

When you download starnet V2, it comes with a tensorflow.dll that is located in the PixInsight/bin directory, did you keep it or did you replace it with another file and which one ? 


I tried various versions : 
For tensor flow : TF from starnet v2, TF 2.3 TF 2.4 TF2.6 most often, PI crashs when I launch Starnetv2
Cuda 11.2 11.5 and 11.6
CUdnn for the associated Cuda version

Whatever I do, it is the CPU that works, never the GPU
I tried with the link you gave and also with the "usual" one :
https://darkskies.space/pixinsight-starnet-cuda/

This is frustrating because I had it work some time ago with Starnet V1 and I would like to enjoy the acceleration coming from my 3070
Edited ...
Like
×

Sorry :-(

Editing a post is only allowed within 24 hours after creating it.
×

Share this forum post

Copy the URL below to share a direct link to this post.

Close
Djibi 2.41
Jean-Baptiste
...
· 
·  3 likes
·  Share link
I got it,
I leave a trace here for the others : 
- it seems that when I updated the PATRH environment variables, the changes were not saved, it may be safer after a change to edit again the PATH variable and double check that the changes were take into account
- as of March 4th 2022, the most recent article which gives the most recent set of compatible ressource between tensorflow version (2.7) / Cuda version (11.6) / Cudann version (8.2 for Cuda 11.4 but also for Cuda 11.6) is this one : 
https://www.williamliphotos.com/starnet-cuda

works great now :-)
Like
×

Sorry :-(

Editing a post is only allowed within 24 hours after creating it.
×

Share this forum post

Copy the URL below to share a direct link to this post.

Close
 
Register or login to create to post a reply.
×

Warning!

This operation will reload the current page. If you have any unsaved information in a form, it will be lost.

You can also dismiss this window, and your operation will be applied at the next page navigation.