VRAM & Power Bill Calculator

LLM VRAM & Home GPU Electricity vs. Cloud Calculator

Calculate how much GPU memory your LLM requires across FP16, 8-bit, and 4-bit quantization (including KV cache context overhead) and compare your local PC's power bill against on-demand Cloud GPUs.

Interactive LLM VRAM Calculator

FP16 / BF16 Full Precision 154.0 GB VRAM
8-Bit Quantized (Q8_0 / FP8) 80.5 GB VRAM
4-Bit Quantized (Q4_K_M / AWQ) 45.5 GB VRAM
Recommended GPU Tier NVIDIA A100 80GB

Home GPU Electricity Bill Calculator

Monthly Energy Consumption 396.0 kWh/mo
Monthly Home GPU Power Cost $71.28/mo
Deploy on RunPod (From $0.20/hr) Rent GPUs on Vast.ai (Save up to 70%)

🖥️ Check Your Local GPU in Terminal (Zero-Install CLI)

Zero Dependencies · Python 3

Auto-detects your NVIDIA GPU (nvidia-smi) or Apple Silicon Unified Memory and prints which Llama, Qwen, DeepSeek, Gemma, and Mistral models fit—with exact ollama run and vllm serve commands. Supports --vram 24 --ctx 32768 --live.

Linux / macOS (Terminal)
curl -sL https://stigla610.github.io/ai-cost-engine/tokenbench.py | python3 -
Windows (PowerShell)
irm https://stigla610.github.io/ai-cost-engine/tokenbench.py | python -

🛒 Local LLM Hardware Buying Guide (Best GPUs & Unified Memory Rigs)

Compare local GPU & Apple Silicon tiers by usable VRAM, 24/7 electricity draw, and model capacity before buying hardware.

Up to 16 GB VRAM · 165W

NVIDIA GeForce RTX 4060 Ti (16GB VRAM)

7B to 14B models (Llama 3.1 8B, Qwen3 14B, Phi-4, Gemma 3 12B) in 4-bit Q4_K_M

Est. Hardware Price~$449
24/7 Electricity ($0.18/kWh)~$21.38/mo
Check Current Price on Amazon →
Up to 24 GB VRAM · 350W

NVIDIA GeForce RTX 3090 / RTX 4090 (24GB VRAM)

14B to 32B models (Qwen3 32B, Gemma 4 31B, Mistral Small 24B, Codestral 22B)

Est. Hardware Price~$899
24/7 Electricity ($0.18/kWh)~$45.36/mo
Check Current Price on Amazon →
Up to 32 GB VRAM · 575W

NVIDIA GeForce RTX 5090 (32GB GDDR7 VRAM)

32B models at 32K–64K context & 8-bit Q8_0 24B models with 1.79 TB/s bandwidth

Est. Hardware Price~$1,999
24/7 Electricity ($0.18/kWh)~$74.52/mo
Check Current Price on Amazon →
Up to 96 GB VRAM · 90W

Apple Mac Mini M4 Pro (64GB) / Mac Studio (128GB Unified)

70B to 123B models (Llama 3.3 70B, Qwen2.5 72B, Mistral Large, Llama 4 Scout) at ultra-low 90W power

Est. Hardware Price~$2,199
24/7 Electricity ($0.18/kWh)~$11.66/mo
Check Current Price on Amazon →

Open-Weight & Popular Model VRAM Guides

Select any model below to view quantization memory tiers and check if your GPU can run it.

DeepSeek: DeepSeek V3Open-Weight
~671B · 4-bit: 436.2 GB · FP16: 1476.2 GB →
Meta: Llama 4 MaverickOpen-Weight
~400B · 4-bit: 260.0 GB · FP16: 880.0 GB →
Qwen: Qwen3 32BOpen-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
Mistral LargeOpen-Weight
~123B · 4-bit: 80.0 GB · FP16: 270.6 GB →
NVIDIA: Nemotron 3 Nano 30B A3BOpen-Weight
~30B · 4-bit: 19.5 GB · FP16: 66.0 GB →
Cohere: Command AOpen-Weight
~111B · 4-bit: 72.2 GB · FP16: 244.2 GB →
Z.ai: GLM 4.5Open-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
MoonshotAI: Kimi K2 0711Open-Weight
~1000B · 4-bit: 650.0 GB · FP16: 2200.0 GB →
Microsoft: Phi 4Open-Weight
~14B · 4-bit: 9.1 GB · FP16: 30.8 GB →
DeepSeek: R1 0528Open-Weight
~671B · 4-bit: 436.2 GB · FP16: 1476.2 GB →
Meta: Llama 4 ScoutOpen-Weight
~109B · 4-bit: 70.9 GB · FP16: 239.8 GB →
Qwen2.5 72B InstructOpen-Weight
~72B · 4-bit: 46.8 GB · FP16: 158.4 GB →
Mistral: Mistral Small 3.2 24BOpen-Weight
~24B · 4-bit: 15.6 GB · FP16: 52.8 GB →
NVIDIA: Nemotron 3 SuperOpen-Weight
~120B · 4-bit: 78.0 GB · FP16: 264.0 GB →
Cohere: Command A+Open-Weight
~111B · 4-bit: 72.2 GB · FP16: 244.2 GB →
Z.ai: GLM 4.5 AirOpen-Weight
~106B · 4-bit: 68.9 GB · FP16: 233.2 GB →
MoonshotAI: Kimi K2 0905Open-Weight
~1000B · 4-bit: 650.0 GB · FP16: 2200.0 GB →
Meta: Muse Glimmer 30BOpen-Weight
~30B · 4-bit: 19.5 GB · FP16: 66.0 GB →
Google: Gemma 4 31BOpen-Weight
~31B · 4-bit: 20.2 GB · FP16: 68.2 GB →
DeepSeek: DeepSeek V3.2Open-Weight
~671B · 4-bit: 436.2 GB · FP16: 1476.2 GB →
Meta: Llama 3.3 70B InstructOpen-Weight
~70B · 4-bit: 45.5 GB · FP16: 154.0 GB →
Qwen: Qwen3 235B A22B Instruct 2507Open-Weight
~235B · 4-bit: 152.8 GB · FP16: 517.0 GB →
Mistral: Codestral 2508Open-Weight
~22B · 4-bit: 14.3 GB · FP16: 48.4 GB →
NVIDIA: Nemotron 3 UltraOpen-Weight
~550B · 4-bit: 357.5 GB · FP16: 1210.0 GB →
Cohere: Command R (08-2024)Open-Weight
~35B · 4-bit: 22.8 GB · FP16: 77.0 GB →
Z.ai: GLM 4.5VOpen-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
MoonshotAI: Kimi K2 ThinkingOpen-Weight
~1000B · 4-bit: 650.0 GB · FP16: 2200.0 GB →
Google: Gemma 3 27BOpen-Weight
~27B · 4-bit: 17.6 GB · FP16: 59.4 GB →
DeepSeek: DeepSeek V3.1Open-Weight
~671B · 4-bit: 436.2 GB · FP16: 1476.2 GB →
Meta: Llama 3.1 70B InstructOpen-Weight
~70B · 4-bit: 45.5 GB · FP16: 154.0 GB →
Qwen2.5 Coder 32B InstructOpen-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
Mistral: Mixtral 8x22B InstructOpen-Weight
~140.8B · 4-bit: 91.5 GB · FP16: 309.8 GB →
NVIDIA: Nemotron 3.5 Content SafetyOpen-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
Cohere: Command R+ (08-2024)Open-Weight
~104B · 4-bit: 67.6 GB · FP16: 228.8 GB →
Z.ai: GLM 4.6Open-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
MoonshotAI: Kimi K2.5Open-Weight
~1000B · 4-bit: 650.0 GB · FP16: 2200.0 GB →
Google: Gemma 3 12BOpen-Weight
~12B · 4-bit: 7.8 GB · FP16: 26.4 GB →
DeepSeek: DeepSeek V3 0324Open-Weight
~671B · 4-bit: 436.2 GB · FP16: 1476.2 GB →
Meta: Llama 3.1 8B InstructOpen-Weight
~8B · 4-bit: 5.2 GB · FP16: 17.6 GB →
Qwen: Qwen3 14BOpen-Weight
~14B · 4-bit: 9.1 GB · FP16: 30.8 GB →
Mistral Large 2407Open-Weight
~123B · 4-bit: 80.0 GB · FP16: 270.6 GB →
NVIDIA: Nemotron 3.5 LightningOpen-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
Cohere: Command R7B (12-2024)Open-Weight
~35B · 4-bit: 22.8 GB · FP16: 77.0 GB →
Z.ai: GLM 4.6VOpen-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
MoonshotAI: Kimi K2.6Open-Weight
~1000B · 4-bit: 650.0 GB · FP16: 2200.0 GB →
OpenAI: gpt-oss-120bOpen-Weight
~120B · 4-bit: 78.0 GB · FP16: 264.0 GB →
DeepSeek: DeepSeek V3.1 TerminusOpen-Weight
~671B · 4-bit: 436.2 GB · FP16: 1476.2 GB →
Meta: Llama 3.2 1B InstructOpen-Weight
~1B · 4-bit: 0.7 GB · FP16: 2.2 GB →
Mistral: Devstral 2 2512Open-Weight
~24B · 4-bit: 15.6 GB · FP16: 52.8 GB →
NVIDIA: Nemotron 3.5 Super VLOpen-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
Z.ai: GLM 4.7Open-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
MoonshotAI: Kimi K2.7 CodeOpen-Weight
~1000B · 4-bit: 650.0 GB · FP16: 2200.0 GB →
AionLabs: Aion-RP 1.0 (8B)Open-Weight
~8B · 4-bit: 5.2 GB · FP16: 17.6 GB →
DeepSeek: DeepSeek V3.2 ExpOpen-Weight
~671B · 4-bit: 436.2 GB · FP16: 1476.2 GB →
Meta: Llama 3.2 3B InstructOpen-Weight
~3B · 4-bit: 2.0 GB · FP16: 6.6 GB →
Qwen: Qwen2.5 VL 72B InstructOpen-Weight
~72B · 4-bit: 46.8 GB · FP16: 158.4 GB →
Mistral: Ministral 3 14B 2512Open-Weight
~14B · 4-bit: 9.1 GB · FP16: 30.8 GB →
Z.ai: GLM 4.7 FlashOpen-Weight
~8B · 4-bit: 5.2 GB · FP16: 17.6 GB →
MoonshotAI: Kimi K3Open-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
Qwen: Qwen3 235B A22B Thinking 2507Open-Weight
~235B · 4-bit: 152.8 GB · FP16: 517.0 GB →
Mistral: Ministral 3 3B 2512Open-Weight
~3B · 4-bit: 2.0 GB · FP16: 6.6 GB →
Z.ai: GLM 5Open-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
MoonshotAI: Kimi K3.1 ProOpen-Weight
~70B · 4-bit: 45.5 GB · FP16: 154.0 GB →
Meta: Muse Spark OpenOpen-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
Qwen: Qwen3 30B A3BOpen-Weight
~30B · 4-bit: 19.5 GB · FP16: 66.0 GB →
Mistral: Ministral 3 8B 2512Open-Weight
~8B · 4-bit: 5.2 GB · FP16: 17.6 GB →
Z.ai: GLM 5 TurboOpen-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
Qwen: Qwen3 30B A3B Instruct 2507Open-Weight
~30B · 4-bit: 19.5 GB · FP16: 66.0 GB →
Mistral: Mistral Large 3 2512Open-Weight
~123B · 4-bit: 80.0 GB · FP16: 270.6 GB →
Z.ai: GLM 5.1Open-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
Qwen: Qwen3 Coder 30B A3B InstructOpen-Weight
~30B · 4-bit: 19.5 GB · FP16: 66.0 GB →
Mistral: Mistral Large 4Open-Weight
~123B · 4-bit: 80.0 GB · FP16: 270.6 GB →
Z.ai: GLM 5.2Open-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
Qwen: Qwen3 Coder 480B A35BOpen-Weight
~480B · 4-bit: 312.0 GB · FP16: 1056.0 GB →
Z.ai: GLM 5.3Open-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
Z.ai: GLM 5.3 FlashOpen-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
Qwen: Qwen3 Coder NextOpen-Weight
~32B · 4-bit: 20.8 GB · FP16: 70.4 GB →
Z.ai: GLM 5.3 FlashXOpen-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
Qwen: Qwen3 Next 80B A3B InstructOpen-Weight
~80B · 4-bit: 52.0 GB · FP16: 176.0 GB →
Mistral: Mistral NemoOpen-Weight
~12B · 4-bit: 7.8 GB · FP16: 26.4 GB →
Z.ai: GLM 5.3 FlashX LegalOpen-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
Qwen: Qwen3 VL 235B A22B InstructOpen-Weight
~235B · 4-bit: 152.8 GB · FP16: 517.0 GB →
Mistral: Mistral Small 3Open-Weight
~24B · 4-bit: 15.6 GB · FP16: 52.8 GB →
Z.ai: GLM 5.3 PrimeOpen-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
ByteDance: UI-TARS 7BOpen-Weight
~7B · 4-bit: 4.5 GB · FP16: 15.4 GB →
Qwen: Qwen3 VL 30B A3B InstructOpen-Weight
~30B · 4-bit: 19.5 GB · FP16: 66.0 GB →
Mistral: Mistral Small 3.1 24BOpen-Weight
~24B · 4-bit: 15.6 GB · FP16: 52.8 GB →
Z.ai: GLM 5V TurboOpen-Weight
~355B · 4-bit: 230.8 GB · FP16: 781.0 GB →
Qwen: Qwen3 VL 8B InstructOpen-Weight
~8B · 4-bit: 5.2 GB · FP16: 17.6 GB →
Mistral: Mistral Small 4Open-Weight
~24B · 4-bit: 15.6 GB · FP16: 52.8 GB →
Qwen: Qwen3.5 397B A17BOpen-Weight
~397B · 4-bit: 258.1 GB · FP16: 873.4 GB →
Mistral: SabaOpen-Weight
~12B · 4-bit: 7.8 GB · FP16: 26.4 GB →
IBM: Granite 4.2 8BOpen-Weight
~8B · 4-bit: 5.2 GB · FP16: 17.6 GB →
Mistral: Voxtral Small 24B 2507Open-Weight
~24B · 4-bit: 15.6 GB · FP16: 52.8 GB →
Qwen: Qwen3.5-122B-A10BOpen-Weight
~122B · 4-bit: 79.3 GB · FP16: 268.4 GB →
Qwen: Qwen3.5-27BOpen-Weight
~27B · 4-bit: 17.6 GB · FP16: 59.4 GB →