Free Tool · VRAM 計算機

LLM 顯示卡記憶體估算器How much GPU memory does your model need?

揀模型、調量化、較 context 長度，即時計你需要幾多 VRAM。支援 70+ 模型預設、MoE、KV Cache 可視化、GPU 相容性檢查。Select a model, pick quantization, dial in context length get your VRAM estimate instantly. Supports 70+ presets, MoE, KV cache visualization, and GPU compatibility.

STEP 1 揀模型 / 設參數

STEP 2 架構設定

STEP 3 使用場景

1

RESULT 計算結果

--
GB 總 VRAM
Weights--
KV Cache--
Overhead--
Total--

GPU 相容性

計算原理： Weights = 總參數 x bytes/param (Q4=0.56, Q8=1.0, FP16=2.0) KV Cache = 2 x attn_layers x kv_heads x head_dim x context x batch x kv_bytes Overhead = 1.5 GB + 激活記憶體 MoE：用總參數計 weights，活躍參數影響激活記憶體 Hybrid Attention (Qwen3.6)：只用 attention layers 計 KV Cache 數值為估算，實際用量因 runtime 而異。建議保留 10-20% 冗餘。
Formula: Weights = total params x bytes/param KV Cache = 2 x attn_layers x kv_heads x head_dim x ctx x batch x kv_bytes Overhead = 1.5 GB + activations. Leave 10-20% headroom.

📬 每週收本地 AI vs 雲端對比，慳 HK$500+/月

留低 email 即刻收到實戰工具清單，比對本地 vs 雲端真實開支。

立即拎免費清單

Free download · 免費下載

免費拎《2026 Free AI Tools Master List》Get the 2026 Free AI Tools Master List free

每一件工具都寫明用途、限制、同我實際點用。留低 email 即刻寄俾你，之後有新片、新工具第一時間通知。

Every tool listed with what it is for, where it falls short, and how I actually use it. Leave your email and it is sent straight away.

睇埋其他工具包 · See the other kits 拎工具包 · Get the kit

▶ 睇手把手教學片 · Watch the walkthrough