On an RTX 4070 Ti (12 GB), Qwen3.5 9B in Q4_K_M needs about 7,2 Go of video memory: it should fit. Estimated generation speed: ≈ 67,4 tokens/s.
Not measured yet: the figures below are estimates, computed from the models already measured on this card. Measure my card
≈ 67,4 tok/s
Generation (estimated)
≥ 7,2 Go
Recommended VRAM (12 GB avail.)
≈ 153 W
Power (ref. model)
0,16 €/M tok
Electricity / M tokens · 0,25 €/kWh
Calculation: Gemma 2 9B measured at 72,1 tok/s on this card, 5,4 Go read per token → × 5,4 Go ÷ 5,7 Go ≈ 67,4 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
0,1 Go
6,4 Go
yes
≈ 66,1 tok/s
8 192
0,3 Go
6,5 Go
yes
≈ 64,8 tok/s
16 384
0,5 Go
6,7 Go
yes
≈ 62,2 tok/s
32 768
1,0 Go
7,2 Go
yes
≈ 57,5 tok/s
65 536
2,0 Go
8,2 Go
yes
≈ 50,1 tok/s
131 072
4,0 Go
10,2 Go
yes
≈ 39,8 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.