# Quantization Calculator (www.quantizationcalculator.com) > The authoritative LLM Quantization, GGUF/EXL2/AWQ Memory, KV Cache, and Multi-GPU Inference Speed (Tokens/sec) Calculator for AI Engineers in the United States and United Kingdom. ## Core Calculation Formulas - **Model Weights Memory (GB)**: `(Parameters_Billions * Bits_Per_Weight) / 8` - **Grouped-Query Attention (GQA) KV Cache Memory (GB)**: `(2 * Layers * KV_Heads * Head_Dim * Context_Tokens * (KV_Cache_Bits / 8)) / 1024^3` - **Total Required VRAM (GB)**: `Model_Weights_GB + KV_Cache_GB + CUDA_Activation_Overhead_GB` - **Estimated Inference Speed (Tokens/sec)**: `(Memory_Bandwidth_GBs * 0.72 * MultiGPU_Scaling) / Active_Weights_GB` ## Programmatic Model Calculators - [Meta Llama 3.3 70B Instruct VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/llama-3-3-70b-instruct): 70.6B params (Dense), GGUF Q4_K_M requires ~44.3 GB VRAM. - [Meta Llama 3.1 8B Instruct VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/llama-3-1-8b-instruct): 8.03B params (Dense), GGUF Q4_K_M requires ~6.4 GB VRAM. - [Meta Llama 3.1 405B Instruct VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/llama-3-1-405b-instruct): 405B params (Dense), GGUF Q4_K_M requires ~247.0 GB VRAM. - [DeepSeek-R1 671B (MoE) VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/deepseek-r1-671b): 671B params (MoE), GGUF Q4_K_M requires ~408.3 GB VRAM. - [DeepSeek-R1 Distill Qwen 32B VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/deepseek-r1-distill-qwen-32b): 32.8B params (Dense), GGUF Q4_K_M requires ~21.4 GB VRAM. - [DeepSeek-R1 Distill Llama 70B VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/deepseek-r1-distill-llama-70b): 70.6B params (Dense), GGUF Q4_K_M requires ~44.3 GB VRAM. - [Qwen 2.5 72B Instruct VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/qwen-2-5-72b-instruct): 72.7B params (Dense), GGUF Q4_K_M requires ~45.6 GB VRAM. - [Qwen 2.5 Coder 32B Instruct VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/qwen-2-5-coder-32b-instruct): 32.5B params (Dense), GGUF Q4_K_M requires ~21.2 GB VRAM. - [Qwen 2.5 14B Instruct VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/qwen-2-5-14b-instruct): 14.7B params (Dense), GGUF Q4_K_M requires ~10.4 GB VRAM. - [Mistral Small 3 24B Instruct VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/mistral-small-3-24b-instruct): 24B params (Dense), GGUF Q4_K_M requires ~16.0 GB VRAM. - [Mistral Large 2 123B VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/mistral-large-2-123b): 123B params (Dense), GGUF Q4_K_M requires ~76.1 GB VRAM. - [Mixtral 8x7B Instruct (MoE) VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/mixtral-8x7b-instruct): 46.7B params (MoE), GGUF Q4_K_M requires ~29.8 GB VRAM. - [Mixtral 8x22B Instruct (MoE) VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/mixtral-8x22b-instruct): 141B params (MoE), GGUF Q4_K_M requires ~87.0 GB VRAM. - [Google Gemma 3 27B Instruct VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/gemma-3-27b-it): 27.2B params (Dense), GGUF Q4_K_M requires ~18.0 GB VRAM. - [Microsoft Phi-4 14B VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/phi-4-14b): 14.7B params (Dense), GGUF Q4_K_M requires ~10.4 GB VRAM. - [Cohere Command R+ 104B VRAM & Quantization Calculator](https://www.quantizationcalculator.com/model/command-r-plus-104b): 104B params (Dense), GGUF Q4_K_M requires ~64.5 GB VRAM. ## Hardware & GPU LLM Benchmark Pages - [NVIDIA GeForce RTX 5090 (32GB VRAM, 1792 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-5090-32gb): US MSRP $1999 / UK £1899. - [NVIDIA GeForce RTX 5080 (16GB VRAM, 960 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-5080-16gb): US MSRP $999 / UK £949. - [NVIDIA GeForce RTX 5070 Ti (16GB VRAM, 896 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-5070-ti-16gb): US MSRP $749 / UK £719. - [NVIDIA GeForce RTX 5070 (12GB VRAM, 672 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-5070-12gb): US MSRP $549 / UK £529. - [NVIDIA GeForce RTX 4090 (24GB VRAM, 1008 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-4090-24gb): US MSRP $1599 / UK £1549. - [NVIDIA GeForce RTX 4080 Super (16GB VRAM, 736 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-4080-super-16gb): US MSRP $999 / UK £959. - [NVIDIA RTX 4070 Ti Super (16GB VRAM, 672 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-4070-ti-super-16gb): US MSRP $799 / UK £769. - [NVIDIA GeForce RTX 4070 Super (12GB VRAM, 504 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-4070-super-12gb): US MSRP $599 / UK £579. - [NVIDIA RTX 4060 Ti 16GB (16GB VRAM, 288 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-4060-ti-16gb): US MSRP $449 / UK £429. - [NVIDIA GeForce RTX 4060 8GB (8GB VRAM, 272 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-4060-8gb): US MSRP $299 / UK £289. - [NVIDIA RTX 3090 / 3090 Ti (24GB VRAM, 936 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-3090-24gb): US MSRP $799 / UK £749. - [NVIDIA GeForce RTX 3080 12GB (12GB VRAM, 912 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-3080-12gb): US MSRP $499 / UK £469. - [NVIDIA GeForce RTX 3060 12GB (12GB VRAM, 360 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/rtx-3060-12gb): US MSRP $279 / UK £259. - [Apple M3 Ultra (Mac Studio) (192GB VRAM, 800 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/mac-studio-m3-ultra-192gb): US MSRP $5999 / UK £5799. - [Apple M4 Max 128GB (128GB VRAM, 546 GB/s) LLM Compatibility](https://www.quantizationcalculator.com/gpu/macbook-pro-m4-max-128gb): US MSRP $4499 / UK £4299. ## Trust, Legal & Documentation Pages - [About & Engineering Methodology](https://www.quantizationcalculator.com/about) - [Contact Engineering Support](https://www.quantizationcalculator.com/contact) - [Privacy Policy (GDPR & CCPA)](https://www.quantizationcalculator.com/privacy-policy) - [Terms of Service](https://www.quantizationcalculator.com/terms-of-service) - [Cookie & LocalStorage Policy](https://www.quantizationcalculator.com/cookie-policy) - [Hardware & Benchmark Disclaimer](https://www.quantizationcalculator.com/disclaimer) - [XML Sitemap](https://www.quantizationcalculator.com/sitemap.xml)