Phi-4 Mini (3.8B) at Q4_K_M needs 4.2 GB once weights, KV cache at 8K context and framework overhead are counted. 16GB RAM, no graphics card offers 0 GB, leaving you 4.2 GB short. You are 4.2 GB short in VRAM but have 12.8 GB of usable system RAM, so llama.cpp can hold the overflow layers in RAM instead of refusing to load.
You are 4.2 GB short in VRAM but have 12.8 GB of usable system RAM, so llama.cpp can hold the overflow layers in RAM instead of refusing to load.
Your machine is not exactly this one. Run this for your exact setup — the form opens pre-filled with 16GB RAM, no graphics card and Phi-4 Mini (3.8B).
Not as it stands. Phi-4 Mini (3.8B) needs 4.2 GB at Q4_K_M and this machine has 0 GB available, a shortfall of 4.2 GB. You are 4.2 GB short in VRAM but have 12.8 GB of usable system RAM, so llama.cpp can hold the overflow layers in RAM instead of refusing to load.
With no GPU to hold any of the model, more system RAM would let it load but it would run entirely on the CPU — too slow to use for anything interactive. The bottleneck here is the missing GPU, not the memory.
New GPU prices are far above MSRP because the memory on the board costs several times what it did in 2025. Relief is not expected before 2027-Q4. The current estimate for relief is 2027-Q4. If you can run this some other way meanwhile, waiting is defensible; if you cannot, the part still does the job today.
Yes, and it is the figure people forget. The 4.2 GB above already includes the KV cache at 8K; that cache grows roughly linearly with context, so doubling the window adds real gigabytes rather than a rounding error. If you plan to work with long documents on 16GB RAM, no graphics card, size for the context you will actually use, not the default.
Data behind this page last checked 2026-08-26.