On an RTX 4070 Ti (12 GB), Qwen3.5 2B in Q4_K_M needs about 2,8 Go of video memory: it should fit. Estimated generation speed: ≈ 215 tokens/s.
Not measured yet: the figures below are estimates, computed from the models already measured on this card. Measure my card
≈ 215 tok/s
Generation (estimated)
≥ 2,8 Go
Recommended VRAM (12 GB avail.)
≈ 96 W
Power (ref. model)
0,03 €/M tok
Electricity / M tokens · 0,25 €/kWh
Calculation: Coder V2 Lite 16B-A2.4B measured at 189 tok/s on this card, 1,5 Go read per token → × 1,5 Go ÷ 1,3 Go ≈ 215 tok/s (generation is memory-bound: speed follows the amount of weights read).
What about a longer context?
Context
KV cache
Estimated VRAM
Fits in 12 GB?
Estimated speed
4 096
0,0 Go
1,8 Go
yes
≈ 208 tok/s
8 192
0,1 Go
1,9 Go
yes
≈ 201 tok/s
16 384
0,2 Go
2,0 Go
yes
≈ 189 tok/s
32 768
0,4 Go
2,2 Go
yes
≈ 167 tok/s
65 536
0,8 Go
2,6 Go
yes
≈ 137 tok/s
131 072
1,5 Go
3,3 Go
yes
≈ 100,0 tok/s
Estimated VRAM = file + KV cache (full-attention layers, f16) + ~0,5 Go of buffers, excluding memory used by the system — hence the slightly higher “recommended VRAM”, which keeps that margin. Speed ≈ estimated speed × weights read ÷ (weights read + added KV cache).
Estimates based on real measurements of other models on this card and on the model's public data (Qwen). As soon as a measurement exists, this page shows the real figures.