Phi-4 Hardware: What 12GB and 16GB Cards Can Hold

Two graphics cards with memory bars for a 14B model

Microsoft's Phi-4 is a 14-billion-parameter model. Its model card describes a dense decoder-only Transformer with a 16K-token context window, released on December 12, 2024. Dense means every weight is read for every token, so the memory math is simple: at 4-bit it fits a 12 GB graphics card. Using the whole context window on that card is where it gets tight.

The weights: about 10.4 GB at Q4

llmhardware.io sizes a model as parameters in billions × bytes per parameter × 1.2, plus 2 GB for the system. At Q4_K_M a weight takes about half a byte, so Phi-4 comes to about 10.4 GB. At 8-bit it is about 18.8 GB, and at FP16 about 35.6 GB. Ollama lists the default phi4 download at 9.1 GB, which is the file alone, before the runtime and the context take their share.

That sorts the cards. An 8 GB card does not hold Phi-4 at Q4. A 12 GB card does, with a little room left. A 16 GB card such as the RTX 4060 Ti 16GB or RTX 5060 Ti 16GB holds it with space for a longer context, but not the 8-bit file, which wants a 24 GB card.

The context: about 3.4 GB more for the full 16K

The 1.2 factor in the formula covers a short context. A long one needs its own line. The KV cache is the memory a model uses to hold the conversation so far, and it grows with every token. Phi-4's published config lists 40 layers, 10 key-value heads and a hidden size of 5,120 split across 40 attention heads, so each head is 128 wide. At the default f16 precision that is 2 × 40 × 10 × 128 × 2 bytes, or about 205 KB per token. Fill all 16,384 tokens and the cache comes to about 3.4 GB.

Add that to the 10.4 GB and the full window needs about 13.8 GB. That fits a 16 GB card and does not fit a 12 GB one. Ollama's FAQ says a q8_0 cache uses about half the memory of f16, which cuts the cache to about 1.7 GB and puts the full window right at the edge of a 12 GB card, around 12.1 GB.

What Ollama does by default

Ollama's FAQ says it uses a 4,096-token context window unless you change it, so a plain ollama run phi4 stays inside 12 GB. To raise the window for every model, set OLLAMA_CONTEXT_LENGTH before you start the server. Inside a chat session, /set parameter num_ctx 16384 does it for that session. After either change, run ollama ps. If the PROCESSOR column shows less than 100% GPU, part of the model has moved to system RAM, and replies will slow down.

When Phi-4-mini is the better download

Phi-4-mini-instruct is a 3.8B model, and its model card lists a 128K-token context. By the same formula it needs about 4.3 GB at Q4, and Ollama's phi4-mini download is 2.5 GB. That makes it the Phi-4 option for an 8 GB card. Its long window can grow the cache past the size of the weights, so set the context to what your documents need instead of the maximum.

One more line from Microsoft's card: it describes Phi-4 as focused on English. If most of your prompts are in another language, compare it with a multilingual model of the same size before you settle on it.

The Phi-4 hardware requirements guide has the card-by-card table, and the VRAM calculator handles other context lengths.

Sources: Microsoft phi-4 and Phi-4-mini-instruct model cards and the phi-4 config on Hugging Face, Ollama library tags and FAQ. Memory figures are formula estimates, not benchmarks.

Comments

Post a Comment

Popular posts from this blog

AMD RX 9070 XT for Local LLMs: ROCm Performance Guide (2026)

The AMD RX 9070 XT launched in early 2026 at $599 with 16GB GDDR6 and 648 GB/s memory bandwidth. For local LLM users who want a mid-range GPU without NVIDIA prices, this changes the calculus significantly. Here is what it actually delivers. RX 9070 XT Specs vs Competition GPU VRAM Bandwidth Price LLM Score RX 9070 XT 16 GB 648 GB/s $599 ~60 tok/s RTX 4070 Ti Super 16 GB 672 GB/s $799 ~62 tok/s RTX 4080 Super 16 GB 736 GB/s $999 ~68 tok/s RTX 4090 24 GB 1,008 GB/s ~$1,100 ~90 tok/s At $599 with 16GB VRAM, the RX 9070 XT is the best value GPU for local LLMs in the mid-range tier in 2026. See the full comparison at LLMHardware.io's AMD GPU guide . ROCm Setup on Linux (2026) # Ubuntu 22.04/24.04 sudo apt install rocm-hip-libraries hip-base # or use the AMD installer: wget https://repo.radeon.com/amdgpu-install/6.3/ubuntu/jammy/amdgpu-install_6.3.60300-1_all.deb sudo dpkg -i amdgpu-install*.deb sudo amdgpu-install --usecase=rocm,lrt,openclsdk # Verify ROCm sees your GPU r...
Read more

llama.cpp flash attention: how to enable and benchmark

Image
Flash attention computes attention with far less memory traffic. The math is identical; the memory movement is not. In llama.cpp it matters most in two situations: long context windows and tight video memory (VRAM). With flash attention enabled, the key-value cache shrinks substantially. A 32k-token context window becomes realistic on a 16GB card, and the prompt-processing phase speeds up, so long documents get ingested faster. The catch is compatibility: not every model and quantization combination supports it, and on some setups the speedup is modest. If you run short chats on a strong GPU, you may barely notice. Where it shines is retrieval-augmented generation (RAG) and long-document work on mid-range cards. Enable it, compare your tokens per second, and keep it if your workload benefits. For the common case of pasting a long PDF and asking questions, it is close to free performance. See how context length changes VRAM needs .
Read more

SillyTavern with Local LLMs: Complete Setup for AI Companions

Image
SillyTavern is the most feature-rich frontend for character-based AI conversations. What began as an AI roleplay tool has evolved into a powerful platform with persona management, memory systems, extensive prompt engineering tools, and connections to every major local inference backend. Here's how to set it up with local models. Quick connect: SillyTavern connects to Ollama, KoboldCpp, llama.cpp server, and TabbyAPI (ExLlamaV2) out of the box. Any of the hardware setups at LLMHardware.io provides a solid backend for SillyTavern. Installation Install via git: git clone https://github.com/SillyTavern/SillyTavern && cd SillyTavern && npm install then node server.js . Access at localhost:8000. For persistent installation, use the included start.sh script. SillyTavern uses Node.js and runs on any platform. The companion app (STMP) enables mobile access to your SillyTavern instance via Tailscale or local network. Character Cards and the TavernAI Format Chara...
Read more