On an RTX 5070 Ti (16 GB), Nemotron 3 Nano 4B in Q4_K_M needs about 4,2 Go of video memory: it should fit. Estimated generation speed: ≈ 204 tokens/s.
Not measured yet: the figures below are estimates, computed from the models already measured on this card. Measure my card
≈ 204 tok/s
Generation (estimated)
≥ 4,2 Go
Recommended VRAM (16 GB avail.)
≈ 135 W
Power (ref. model)
0,05 €/M tok
Electricity / M tokens · 0,25 €/kWh
Calculation: Qwen3.5 4B measured at 196 tok/s on this card, 2,8 Go read per token → × 2,8 Go ÷ 2,7 Go ≈ 204 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
3,3 Go
yes
≈ 200 tok/s
8 192
0,1 Go
3,3 Go
yes
≈ 195 tok/s
16 384
0,3 Go
3,5 Go
yes
≈ 187 tok/s
32 768
0,5 Go
3,7 Go
yes
≈ 172 tok/s
65 536
1,0 Go
4,2 Go
yes
≈ 149 tok/s
131 072
2,0 Go
5,2 Go
yes
≈ 117 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 (NVIDIA Nemotron). As soon as a measurement exists, this page shows the real figures.