Download the latest CUDA toolkit or not?

PixInsight Addicts 5 replies283 views
Jeffery Richards avatar

So just got an email from NVIDIA about the release of a new CUDA toolkit. So what is the right answer…download and install or not? Would this enhance our RCAstro goodies or hinder them? I do have a new 50 series GPU and the toolkit mentions enhancements for the Blackwell cards but I don't know exactly how that impacts Russell's tensorflow custom package.

andrea tasselli avatar
Can't say for sure but if ain't broke don't fix it.
Robert Žibreg avatar
andrea tasselli:
Can't say for sure but if ain't broke don't fix it.

This.
Jeffery Richards avatar

Yeah, I’ll wait for guidance from Russell. 😁

Read noise Astrophotography avatar

Jeffery Richards · Aug 7, 2025, 04:30 PM

So just got an email from NVIDIA about the release of a new CUDA toolkit. So what is the right answer…download and install or not? Would this enhance our RCAstro goodies or hinder them? I do have a new 50 series GPU and the toolkit mentions enhancements for the Blackwell cards but I don't know exactly how that impacts Russell's tensorflow custom package.

It will not work because RC astro tools need CUDA 11.8 and the update is probably 12.x

If you try it you will give yourself headaches trying to fix it again.

Definitely don’t do it


If you do do it you will need to replace a load of files in the Pix bin folder and then you will need to wrap pix in something like this

@echo off

setlocal

set "PI_BIN=C:\Program Files\PixInsight\bin"

set "CUDA11_BIN=%ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin"

rem Prefer PI’s local DLLs first

set "PATH=%PI_BIN%;%PATH%"

rem Nuke CUDA 12.* from PATH for this process

for %%V in (0 1 2 3 4 5 6 7 8 9) do (

set "PATH=%PATH:%ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v12.%%V\bin;=%"

for %%M in (0 1 2 3 4 5 6 7 8 9) do (

set "PATH=%PATH:%ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v12.%%V%%M\bin;=%"

)

)

rem Add CUDA 11.8 bin after PI (fallback for any dll not in PI\bin)

if exist "%CUDA11_BIN%\cudart64_110.dll" set "PATH=%PI_BIN%;%CUDA11_BIN%;%PATH%"

rem TensorFlow logging + behaviour

set "TF_CPP_MIN_LOG_LEVEL=0"

set "TF_FORCE_GPU_ALLOW_GROWTH=TRUE"

set "CUDA_VISIBLE_DEVICES=0"

set "TF_LOG_DEVICE_PLACEMENT=1"

cd /d "%PI_BIN%"

start "" "%PI_BIN%\PixInsight.exe"

endlocal

Peter avatar

Hi Jeffrey.

I’ve just upgraded my system with a RTX 5070 from a GTX 1080.

I used the following after reinstalling Visual Studio 2022.

cuda_11.8_522.06-Windows

cudnn_9.10.2_windows

Everything worked fine and I would say around 20% faster as I’ve got an extra 4Gb memory on card and more modern architecture.