Protip if you want to squeeze most out of your VRAM if you have a CPU with iGPU
Posted by Th3Sim0n@reddit | LocalLLaMA | View on Reddit | 11 comments
If you want to reclaim couple hundred MB of VRAM, enable iGPU in bios and plug in the display cable to motherboard, that way iGPU handles the system and frees up the memory of dedicated GPU entirely.
This is especially useful for those of you who run Windows or non-server Linux with GUI.
Hope that helps!
StarshipCherry@reddit
Eh I lose 800mb from the virtual dummy EDID plugged in (DP-1) for Sunshine control and gaming...
Blizado@reddit
I only have 2mb VRAM used on my "AI" server with Sunshine and Moonlight running. I use a HDMI dongle on my Intel iGPU.
StarshipCherry@reddit
Are you running on bare-metal? My system is on proxmox and i haven't passed through my iGPU. I must be rendering a 4k HDR 90fps display constantly... 730mb
WouterC@reddit
Good advice
typical-predditor@reddit
For some reason my computer won't let me do this. I'd love to use the hdmi port on the MB for an extra monitor but it won't let me.
Th3Sim0n@reddit (OP)
You're asking for hybrid graphics mode where both gpus serve the system to display desktop components, this can be tricky. My bios has a "hybrid graphics" option, try that maybe?
Real_Chard5666@reddit
Generally the igpu is in the cpu, AMD 9000x/3d series CPUs
Blizado@reddit
Good advice. I run my "AI" server with a HDMI dongle on the iGPU, only 2MB used on my RTX 5080.
FranticBronchitis@reddit
sad iGPU/dGPU conflict noises
GoodSamaritan333@reddit
Good thing is that normally, CPUs with iGPU offer the same numper of pcie lanes as their counterparts without iGPU
GoodSamaritan333@reddit
Unfortunately my main gpu doesn't have an igpu, so I bought a cheap radeon, so it doesn't interfere with my nvidia drivers/stuff, on windows. On linux, I plug a notebook and main PC on a switch, so I can unplug video cables and use ssh from the notebook to control the llama.server from remotely.