LLM VRAM Calculator
Pick a model and a GPU to see weight + KV cache memory, estimated token speed, and ready-to-run launch commands.
Concurrency (Batch):
Dense 27B·64L (d=5120)·24Q/4KV·Ctx 256K·Text+Image+Video
Smooth Fit19.33 / 23.5 GB~42.6 tok/s· Q4_K_M · 16K ctx
3. Weight Quantization
4. Per-Request Context16K (16,384 tok)
Verified: 2026-11-20·Roofline bandwidth/compute estimate (Batch=1)
Smooth Fit
19.33/ 23.5 GB usable
Physical 24 GB · Load 82.2%
Theoretical Decode Speed
~42.6tokens/s
Actual speed varies ~±25% by runtime
Max Context at Batch=1
78K/ 256K
Enable Q8/Q4 KV quantization to scale batch concurrency
Weights: 16.37 GBKV Cache (1×16K): 1.34 GBOverhead: 1.63 GB
🏃 Smooth & Snappy~11.7s for a 500-token response
Reference Launch CLI:
OLLAMA_CONTEXT_LENGTH=16384 ollama run qwen3.5:27bRTX 4090 Fits Qwen3.5-27B · Cost & TCO
Approx. $1,899 (~$79/GB VRAM) · 450W TDP
Intelligence vs. Cost / VRAM Pareto Quadrant
Pareto FrontierTop-left corner = maximum benchmark capability for minimum VRAM or API spend
Qwen3.5-35B-A3BAlibaba Qwen · MoE A3BQ4: 23.34 GBIndex 45Pareto Optimal