KV Cache Optimization: How to Extend Your LLM Context Window

The KV cache is what makes long-context LLM inference possible — and what limits it. Understanding and optimizing it is essential for anyone running LLMs locally with large contexts.

What Is the KV Cache?

During inference, the transformer computes Key and Value matrices for every token in the context. To avoid recomputing these for past tokens at each step, they're cached in VRAM — the KV cache. It grows linearly with context length and can consume more memory than the model itself at long contexts.

KV Cache Size Formula

KV_size = 2 × num_layers × num_kv_heads × head_dim × context_length × dtype_bytes

For Llama 3 8B at 128K context (fp16):
= 2 × 32 × 8 × 128 × 131072 × 2 = ~17 GB

This is why 128K context requires so much VRAM. LLMHardware.io's context length calculator helps you find the max context your hardware supports for any model.

Reducing KV Cache Memory

1. KV Cache Quantization

# llama.cpp: quantize KV cache to int8 (half the memory, minimal quality loss)
./llama-server -m model.gguf   --ctx-size 131072   --cache-type-k q8_0   --cache-type-v q8_0

Q8 KV cache cuts memory by 50% with negligible quality impact. Q4 cuts by 75% with slight degradation on very long contexts.

2. Grouped Query Attention (GQA)

Models using GQA (Llama 3, Mistral, Qwen2) share KV heads across multiple query heads, reducing KV cache size by the GQA ratio (typically 4–8x). When choosing models, prefer GQA architectures for long-context work.

3. Flash Attention

# vLLM (enabled by default)
python -m vllm.entrypoints.openai.api_server   --model Qwen/Qwen2.5-7B-Instruct   --enable-chunked-prefill   --max-num-batched-tokens 8192

Flash Attention reduces peak VRAM usage during prefill by computing attention in chunks without materializing the full attention matrix.

Practical Context Length vs Hardware

GPUMax Context (7B, Q4)Max Context (7B, Q8 KV)
RTX 3060 12GB32K64K
RTX 4090 24GB96K128K+
Mac Studio M4 Ultra 192GB512K+1M+

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