Linux Fedora 37 no GPU utilizign with my Nvidia RTX 2070 Super

[bleak edge]

In the install i picked the Nvidia card but everytime i start Invoke AI it uses the CPU.

[humble peak]

You installed the nvidia drivers before running the installer?

[bleak edge]

Yes

[humble peak]

Hum, I can't remember now, but there is a command that you can run in python to see If It sees your gpu

But It seems by the first print that torch can't see your gpu

Try to reinstall torch

And the run the command inside python

Import torch

cuda.get_device_count()

Or

cuda.is_available()

The First need to return 1

The Second needs to return True

[bleak edge]

okay im very sorry but im a noob how exactly can i reinstall torch?

[humble peak]

Oh, ok, I'm on my phone so give me a minute

[bleak edge]

np

[humble peak]

https://invoke-ai.github.io/InvokeAI/installation/010_INSTALL_AUTOMATED/#installation

Use the process in troubleshoot

To "InvokeAI runs extremely slowly on Linux or Windows systems"

[bleak edge]

okay thanks i will try

[humble peak]

Good Luck my friend

[bleak edge]

this happens when i try run "pip install xformers"

[humble peak]

You need to run the command inside the option 5 of the invoke

Same with the pip install torch torchvision --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu117
pip install xformers

[bleak edge]

omg

im sorry

like i said im a noob

[humble peak]

No prob

[bleak edge]

btw my first month in linux

but ok i will do this

[bleak edge]

sorry but did you really mean option 5?

[humble peak]

Ops

They changed

Option 8

Console mode

[bleak edge]

so this happens when i run "pip install torch torchvision --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu117"

[humble peak]

For now let's see what the torch return when you run It with python

At least he installed the torch again

[bleak edge]

ok so what exactly should i do?

[humble peak]

In the same console

Type

Python

[humble peak]

And execute the same commands as before

[bleak edge]

okay

so thats the result

[humble peak]

Welp

Try to add torch."command" for the cuda. ...

[bleak edge]

so you mean torch."cuda.get_device_count()" ?

[humble peak]

Yes, without the quotation marks

[humble peak]

Hum

Let me see, I think that I wrote wrong

Try torch.cuda.is_available()

And torch.cuda.device_count()

[bleak edge]

so i guess this means it can detect my gpu?

[humble peak]

Yeah, but It is ignoring it

[bleak edge]

welp

[humble peak]

I think It can be two things

But let's try sometimes first

Type ctrl+d to exit the python

[bleak edge]

jep

now im in InvokeAI terminal

[humble peak]

Type

export HSA_OVERRIDE_GFX_VERSION=10.3.0

Then enter python

And try to see again If It gets your gpu

[humble peak]

Hum

What is the version of your nvidia drivers?

[bleak edge]

470.182.03

[humble peak]

The Cuda version is 11.7?

https://developer.nvidia.com/cuda-11-7-0-download-archive

[bleak edge]

so i found out my cuda version is 11.4

[humble peak]

Try to see If your gpu supports 11.7

I use AMD gpu

So I'm not used to how they work with drivers

[bleak edge]

do you know if it needs to be 11.7?

or can i install a newer version

like 12.1

[bleak edge]

Ok, i have to tell you im very thankfull that you tried to help me but i think i will only use this on windows.

[humble peak]

It is easier

[bleak edge]

sad but true

and jup it works ....

[humble peak]

Sorry but i really didn't try It with a nvidia gpu until now

[bleak edge]

its not your fault, you did very well

btw sorry for my broken english im from Germany