On an RTX 5070 Ti (16 GB), Llama 3.2 1B in Q4_K_M needs about 2,3 Go of video memory: it should fit. Estimated generation speed: ≈ 588 tokens/s.
Not measured yet: the figures below are estimates, computed from the models already measured on this card. Measure my card
≈ 588 tok/s
Generation (estimated)
≥ 2,3 Go
Recommended VRAM (16 GB avail.)
≈ 73 W
Power (ref. model)
0,01 €/M tok
Electricity / M tokens · 0,25 €/kWh
Calculation: Qwen2.5 1.5B measured at 481 tok/s on this card, 0,9 Go read per token → × 0,9 Go ÷ 0,8 Go ≈ 588 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
1,4 Go
yes
≈ 515 tok/s
8 192
0,3 Go
1,5 Go
yes
≈ 450 tok/s
16 384
0,5 Go
1,8 Go
yes
≈ 359 tok/s
32 768
1,0 Go
2,3 Go
yes
≈ 255 tok/s
65 536
2,0 Go
3,3 Go
yes
≈ 162 tok/s
131 072
4,0 Go
5,3 Go
yes
≈ 93,4 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 (Meta Llama). As soon as a measurement exists, this page shows the real figures.