HSA_OVERRIDE_GFX_VERSION environment variable on linux appimage

[DimitriGilbert]

Which version of LM Studio?
LM Studio 0.3.9-6

Which operating system?
Fedora 41

What is the bug?
I wanted to try to run the ROCm backend on my ryzen 5 8600G (760M I think).
In order to get llama.cpp to recognize your gpu as a compatible ROCm one, you can set the HSA_OVERRIDE_GFX_VERSION env variable
I've configured my .desktop to override this en var when starting lmstudio (see later in report) but the detected hardware is still "RADV GFX1103_R1"

[Desktop Entry]
Categories=Development;
Comment[en_GB]=Use the chat UI or local server to experiment and develop with local LLMs.
Comment=Use the chat UI or local server to experiment and develop with local LLMs.
Exec=env HSA_OVERRIDE_GFX_VERSION=11.0.1 /home/myUser/.local/bin/lmstudio %U
GenericName[en_GB]=
GenericName=
Icon=/home/myUser/.local/share/zap/v2/icons/lmstudio..png
Keywords=developer;llm;
MimeType=
Name[en_GB]=LM Studio (AppImage)
Name=LM Studio (AppImage)
Path=
StartupNotify=true
StartupWMClass=LM Studio
Terminal=false
TerminalOptions=
TryExec=/home/myUser/.local/bin/lmstudio
Type=Application
X-AppImage-Integrate=
X-AppImage-Version=0.3.9
X-KDE-SubstituteUID=false
X-KDE-Username=
X-Zap-Id=lmstudio
category=Development;Utility;

Am I doing something wrong ? Maybe you do not support this work around ?
Mind you it was for testing purposes, the vulkan backend work just fine :)