LM Studio Hits AMD VRAM Ceiling

The useful takeaway is that LM Studio is not the thing deciding how much unified memory the 780M can expose; it mostly consumes whatever the platform and driver stack make available.

• –On AMD iGPUs, the reserved VRAM aperture is usually set in BIOS/UEFI under UMA frame buffer or GPU memory allocation settings.
• –LM Studio’s own docs emphasize GPU offload controls and note that some memory behavior is automatic; the “dedicated GPU memory only” option is currently CUDA-only, not an AMD fix.
• –If the BIOS is locked or only offers a small preset, the real limit may be the mini-PC vendor’s firmware rather than LM Studio.
• –For local LLMs, this is a hardware-capacity question disguised as an app question: bigger models want either more reserved iGPU memory or more aggressive offload-to-RAM tradeoffs.
• –In practice, 8 GB may still be usable for many quantized models, so the better optimization path may be model choice and offload settings rather than chasing a BIOS hack.