How-To: Speed up StarNet++ as PI Module on nVidia GPUs with CUDA on Windows

Hello darkarchon,

as the site https://darkskies.space/pixinsight-starnet-cuda/ refers to
does that mean that installing the latest versions of Cuda and cuDDN will not work ?
I've been trying the versions as indicated on a GForce GTX, without success.
I've only tried the latest versions on my new system with a GForce 2070 S, without success.
The latter one is a complete new system with latest drivers, clean win10 pro 64.

As you've already mentioned in this thread
" You probably did not download everything or don't set up the parameters correctly "
It's obvious that I've double checked a coupe times.

In case I still may have missed something during the installation,
Any hint on where it could be ?
Any other information to the way of succeeding this installation is of course welcome as well.

Any help would be very appreciated.

Johan.
 
Hello all,

I'll partly answer my own question.

Apart from changing to the versions as indicated on darkskies,
I have no idea what else has changed since I powered on my system yesterday. (shame on me ;-) )
Though kind of 'out of the blue' it now functions.
So unfortunately, I have no hints for the people out there who are still suffering to get this starnet++ speedup to run
apart from what I mentioned earlier,
I reinstalled according to versions as indicated on darkskies.

Stay healthy and clear skies.
Johan
 
yes it should work on linux but (you guessed it) i use osx. google dropped support for tensorflow-gpu some time ago. it might have something to do with apple locking out new nvidia cards after osx 10.13.

rob

and that is why you never buy an Apple product...
 
Just looking at the instructions, I see a potential error with:

step 3 - it says to copy the bin folder, but it highlights both bin and lib folders with movement arrows, suggesting that you should move both bin and lib, not just bin as per the written instruction.

I don't use starnet (yet), but thought I'd highlight this issue with the instructions.
 
Thank you sooooo much Stanley,
you make my life a little better;)

It works fine with a Geforce RTX 1660 Super on my System,
before i follow your tip i need >2h with Stride 16, now i have a result in 25 minutes.
That's great

CS
Chris
 
Amazing news. Can't wait to download PI optional update with GPU-enabled TensorFlow libraries for Linux and Windows! Hopefully the latest Ampere RTX chips (3070/80/90) will be supported.
 
Back
Top