On an RTX 5070 Ti (16 GB), Gemma 3 4B in Q4_K_M needs about 3,8 Go of video memory: it should fit. Estimated generation speed: ≈ 225 tokens/s.
Not measured yet: the figures below are estimates, computed from the models already measured on this card. Measure my card
≈ 225 tok/s
Generation (estimated)
≥ 3,8 Go
Recommended VRAM (16 GB avail.)
≈ 112 W
Power (ref. model)
0,03 €/M tok
Electricity / M tokens · 0,25 €/kWh
Calculation: Qwen3 2507 4B measured at 224 tok/s on this card, 2,3 Go read per token → × 2,3 Go ÷ 2,3 Go ≈ 225 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 16 GB?
Estimated speed
4 096
0,1 Go
2,9 Go
yes
≈ 219 tok/s
8 192
0,2 Go
3,0 Go
yes
≈ 212 tok/s
16 384
0,3 Go
3,1 Go
yes
≈ 199 tok/s
32 768
0,6 Go
3,4 Go
yes
≈ 178 tok/s
65 536
1,3 Go
4,1 Go
yes
≈ 147 tok/s
131 072
2,5 Go
5,3 Go
yes
≈ 109 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 (Google Gemma). As soon as a measurement exists, this page shows the real figures.