GPU Acceleration on AMD Graphics Cards for Tensorflow based processes [Deep Sky] Processing techniques · Philipp Weber · ... · 32 · 2517 · 2

Hey Guys,

thank you for the answers!
Since this is well above my knowledge I think I have to wait until someone makes the move and tries it for windows.

I might dig a little into this DirectML plugin and see if I can get something to run hehe

CS
Jakob

Yesterday I spent like 3hrs+ to find a solution to get GPU accelation within Pixisight to run on windows. In the end it didn't worked and I found a solution for myself

I tried to use the zluda process to get amd cards to work with cuda and I think (maybe) with a certain implementation or self made fix it could work with Pixinsight.
When I open Pixinsight with zluda and try to use any AI based plugin like noiseX, BlurX,... it tries to initialize the process and I can even see a roughly 10-15% jump in the gpu memory usage when I do so but after like 5-10sec the program crashes, the increase in memory usage can be a coincidence or maybe its a sign that it wants to work but some kind of translation layer or code is missing.

In the end I had the idea to just install my old gtx1070 and see if Pixinsight wants to use that gpu when I try to use the AI plugins... And for my surprise it worked out of the box without any software magic!


The gtx1070 will just idle in my PC and will be used for Pixinsight only in the future. The RX 7900XT is still the main card for everything else like gaming.  

Maybe that's a solution for someone like me that has an older gtx class card (which are still signifinantly faster than a mid/higher end modern processor. In my example the gtx1070 can complete a starX run in about 50sec-1min compared to 8min+ on a R7 5800x) and they want to upgrade to an AMD card because the raw price to performance is better compared to NVIDIA

Clear Skies
Jakob

Maybe slightly off-topic, but you can get an Intel Mac with an eGPU (always AMD). Runs all Xterminator SW straight out of the box, no drivers required.

This is awesome, this is what I’m going to do. Can you tell what graphic card did you buy and the results without and with egpu?

Hi @montas, I built my system a couple of years ago with an RX 580, which is x times faster than just the CPU. I don’t exactly how much faster because it is quite laborious to benchmark without GPU, but expect between 5 and 10 times faster, depending on the task at hand.

Today I would get myself an RX6800 and a Razer Core X. The RX6800 is quite cheap, especially second hand, and is at least twice as fast as the RX 580. Cards like the RX7900 are even faster but substantially more expensive. 

Hope this helps.

Byron Miller:
I got Tensorflow + ROCM 6.1.1 to work on Ubuntu, no zluda necessary. Put a build up for download since building is a royal pain right now - got a few PRs upstreamed to AMD to try and fix it

https://rockchucksummit.com/pixinsight-gpu-acceleration-for-amd/

Working on fedora 40 now

Very nice!