4.0 GB
model weights
~/tools / vram-calculator
GPU VRAM calculator
The question every local-model experiment starts with: will it fit? This calculator does the real arithmetic — weights = parameters × bits, KV cache = 2 × layers × kv-dim × context × 2 bytes — and delivers a straight verdict per GPU, headroom included. No vibes, just bytes.
TOOL 04
Will it fit? — GPU VRAM calculator
the question every local-model run starts with1.07 GB
kv cache @ 8K ctx
5.6 GB
total (+10% overhead)
| card | vram | headroom | verdict |
|---|---|---|---|
| RTX 3060 | 12 GB | +6.4 GB | fits |
| RTX 4070 Ti | 16 GB | +10.4 GB | fits |
| RTX 3090 / 4090 | 24 GB | +18.4 GB | fits |
| A100 40G | 40 GB | +34.4 GB | fits |
| A100 80G / H100 | 80 GB | +74.4 GB | fits |
| MacBook M-series 36G | 36 GB | +30.4 GB | fits |
weights = params × bits/8 · KV cache = 2 × layers × kv_dim × ctx × 2 bytes · +10% runtime overhead. Green rows are cards that fit.
more
tools.siblings
Token & cost metercount tokens · compare API cost across modelsRAG capacity plannercorpus size → vectors, RAM, embed costProvider request translatorOpenAI ⇄ Anthropic/Bedrock request JSONJSONL dataset inspectorfine-tune data sanity check, fully localPrompt diffword-level diff for prompt iterationPrompt linterESLint for system prompts