On an RTX 5070 Ti (16 GB), Ministral 3 8B in Q4_K_M needs about 6,3 Go of video memory: it should fit. Estimated generation speed: ≈ 141 tokens/s.
Not measured yet: the figures below are estimates, computed from the models already measured on this card. Measure my card
≈ 141 tok/s
Generation (estimated)
≥ 6,3 Go
Recommended VRAM (16 GB avail.)
≈ 154 W
Power (ref. model)
0,08 €/M tok
Electricity / M tokens · 0,25 €/kWh
Calculation: Qwen3 8B measured at 146 tok/s on this card, 4,7 Go read per token → × 4,7 Go ÷ 4,8 Go ≈ 141 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
0,5 Go
5,9 Go
yes
≈ 129 tok/s
8 192
1,1 Go
6,4 Go
yes
≈ 117 tok/s
16 384
2,1 Go
7,5 Go
yes
≈ 99,1 tok/s
32 768
4,3 Go
9,6 Go
yes
≈ 75,7 tok/s
65 536
8,5 Go
13,8 Go
yes
≈ 51,5 tok/s
131 072
17,0 Go
22,3 Go
no
—
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 (Mistral). As soon as a measurement exists, this page shows the real figures.